Exercise MR of Skeletal Muscles, the Heart, and the Brain

Abstract

Magnetic resonance (MR) imaging (MRI) is routinely used to evaluate organ morphology and pathology in the human body at rest or in combination with pharmacological stress as an exercise surrogate. With MR during actual physical exercise, we can assess functional characteristics of tissues and organs under real‐life stress conditions. This is particularly relevant in patients with limited exercise capacity or exercise intolerance, and where complaints typically present only during physical activity, such as in neuromuscular disorders, inherited metabolic diseases, and heart failure. This review describes practical and physiological aspects of exercise MR of skeletal muscles, the heart, and the brain. The acute effects of physical exercise on these organs are addressed in the light of various dynamic quantitative MR readouts, including phosphorus‐31 MR spectroscopy (³¹P‐MRS) of tissue energy metabolism, phase‐contrast MRI of blood flow and muscle contraction, real‐time cine MRI of cardiac performance, and arterial spin labeling MRI of muscle and brain perfusion. Exercise MR will help advancing our understanding of underlying mechanisms that contribute to exercise intolerance, which often proceed structural and anatomical changes in disease. Its potential to detect disease‐driven alterations in organ function, perfusion, and metabolism under physiological stress renders exercise MR stress testing a powerful noninvasive imaging modality to aid in disease diagnosis and risk stratification. Although not yet integrated in most clinical workflows, and while some applications still require thorough validation, exercise MR has established itself as a comprehensive and versatile modality for characterizing physiology in health and disease in a noninvasive and quantitative way.

Evidence Level

5

Technical Efficacy

Stage 1

Exercise is arguably the most common form of physical stress that humans encounter in daily life. One's ability to perform exercise or even daily‐life tasks is strongly associated with morbidity and all‐cause mortality, and is an important indicator if not predictor of physical health. ¹ Engaging in physical exercise is a voluntary action initiated by the brain, and set in motion through the recruitment of motor units that each consist of a motor neuron and its innervated skeletal muscle fibers. Such myofiber contraction triggers a cascade of biochemical and physiological responses, including the rise of cellular metabolism that leads to recruitment of the local microvasculature, and the cardiac and respiratory reserve, in order to sustain the desired mechanical task while maintaining energy homeostasis and pH balance in active muscle fibers. To do so, the musculoskeletal, cardiovascular, respiratory, and nervous systems must all work together seamlessly (Fig. 1).

FIGURE 1.

Schematic overview highlighting the involvement of skeletal muscles, the heart, and the brain to perform physical exercise as a complex interplay of the musculoskeletal, cardiovascular, respiratory, and nervous systems. Many aspects of human in vivo organ function, perfusion, and metabolism can be evaluated noninvasively with magnetic resonance (MR) techniques during exercise, eg, tissue oxygenation with blood oxygenation level‐dependent (BOLD) MRI of skeletal muscle, myocardial function with cine MRI, and cerebral perfusion with arterial spin labeling (ASL) MRI. Other readouts include proton MR spectroscopy (¹H‐MRS), phosphorus‐31 MRS (³¹P‐MRS), chemical exchange saturation transfer effect of free creatine (CrCEST), dynamic contrast enhanced (DCE) MRI, intravoxel incoherent motion (IVIM) MRI, and phase‐contrast (PC) MRI. To date, essentially no or very limited work on exercise MR of the lungs or any other organs has been reported. Figure created with BioRender.com, and Servier Medical Art (https://smart.servier.com) under a Creative Commons Attribution 3.0 Unported license (https://creativecommons.org/licenses/by/3.0/).

In numerous pathophysiologic conditions, exercise cannot be sustained due to premature mechanical failure of contracting myofibers. Indeed, exercise intolerance is a hallmark of metabolic ² and neuromuscular disorders, ³ as well as complex systemic diseases such as heart failure, ⁴ post‐viral fatigue, ⁵ and diabetes. ⁶ Physical complaints typically present only during exercise, particularly in early stages of disease, and exercise intolerance may worsen as disease progresses. Currently, cardiopulmonary exercise testing (CPET) constitutes a central part of clinical workflows for evaluating the differential contributions to exercise intolerance of the peripheral, cardiovascular, and respiratory systems. ⁷ , ⁸ Exercise testing can also be used to elicit a pathophysiological response to help classify disease severity, eg, in heart failure with preserved ejection fraction (HFpEF) ⁹ and peripheral artery disease. ¹⁰ Such examinations aim to replicate the physical complaints experienced by the patient in daily life, and allow a direct evaluation of the underlying pathophysiology. The merit of exercise testing, beyond organ's anatomy, tissue perfusion and resting metabolic state, has thus been long recognized and accepted in the clinical practice. With an aging population and high prevalence of obesity and diabetes, the number of patients presenting with complaints related to reduced exercise capacity and exercise intolerance will increase. Moreover, there is mounting evidence that regular physical exercise improves musculoskeletal, ¹¹ cardiovascular, ¹² and cognitive health. ¹³ We foresee growing demand for methods that can quantify the response of the human body to exercise stress will grow. Specifically, the combination of physiological stress with simultaneous readouts of in vivo organ function, perfusion, and metabolism that is possible with exercise magnetic resonance (MR) shows great promise to contribute to advancing our understanding of exercise physiology in health and disease.

Medical diagnostic imaging modalities typically require patient immobility during image acquisition, and MR imaging (MRI) is no exception. Given the relatively low sensitivity of MR and the consequent need for repeated acquisitions to collect sufficient signal for image reconstruction, much thought and effort has been put into mitigating motion artifacts that would otherwise reduce image quality. Yet, already in the early days of MR for in vivo human applications, its noninvasive nature and therewith its possibilities to acquire quantitative data in situ without disruptive biopsy procedures were utilized to study muscle energy metabolism during exercise in human extremities with phosphorus‐31 MR spectroscopy (³¹P‐MRS). ¹⁴ , ¹⁵ Driven by developments in hardware, acquisition strategies, and reconstruction algorithms, the use of both MRS and MRI for investigating exercise physiology has since evolved beyond those early ³¹P‐MRS studies of exercising human skeletal muscle in the 1980s.

In this review, we outline and discuss practical and physiological aspects of exercise MR. We present examples of how MR examination with exercise stress can shed light on human physiology, and how it can help to unmask otherwise undetectable pathophysiology. The acute impact of physical exercise on active skeletal muscles, the heart, and the brain is addressed, as well as the physiological and metabolic readouts that can be gained from studying these organs during recovery immediately after exercise. Finally, we reflect on how exercise MR stress testing can be a versatile platform for comprehensive, multi‐level, and multi‐organ investigations of both physiology and pathophysiology.

Modes for Exercise MR Stress Testing

The majority of MR studies on acute effects of exercise have been conducted with physical exercise performed in horizontal body position, with the subject supine or prone on the MR table. Notable exceptions are some studies were subjects exercised in an upright position on a treadmill directly adjacent to the MR system, with measurements being done after quickly transferring the subject onto the table and into the MR scanner after exercise. ¹⁶ , ¹⁷ Other studies have utilized a vertical open‐bore 0.5 Tesla MR system for upright bicycling exercise. ¹⁸ , ¹⁹ With contemporary clinical MR systems typically configured with a horizontal bore, much effort has been put into the design of ergometers that are compatible with the strong magnetic field while still enabling physiological modes of prone or supine exercise within the confined space of the MR bore. ²⁰ , ²¹ , ²² The most simple but somewhat atypical mode of exercise is isometric handgrip exercise. It avoids body movement, but induces only a small increase in heart rate at low‐intensity exercise. Many investigations focus on isometric or concentric muscle contraction at a predefined percentage of maximum voluntary contractile force, either until exhaustion or for a predetermined duration. While such single‐limb or single‐muscle exercise elicits a local physiological response, these modes fall short in realistically capturing any daily‐life physical activity. Moreover, such exercises only involve a small muscle mass that is insufficient to trigger a physiologically realistic systemic cardiovascular response, ²³ which requires recruitment of a much larger muscle mass. ²⁴ In‐magnet leg‐bicycling ²¹ , ²⁵ or arm‐bicycling ²⁶ , ²⁷ paradigms have been developed for use in clinical MR systems. These modes provide more realistic surrogates for whole‐body exercise that can be combined with dynamic MR acquisitions to study systemic aspects of human exercise physiology. Still, the horizontal body posture will impact the hemodynamic response ²⁸ (eg, due to differences in gravity‐induced hydrostatic pressures that affect cardiac preload), skeletal muscle recruitment and perfusion, and exercise capacity. ²⁹ These effects should be considered when comparing MR‐measured parameters against more clinically accepted exercise stress tests such as CPET that are conducted in an upright position.

Physical Exercise and Skeletal Muscles

Exercise MR stress testing has been pivotal for over 50 years in offering quantitative in vivo data for the study of skeletal muscle energy metabolism, peripheral microvascular function, and muscle contraction dynamics. ³⁰ Despite its demonstrated merit in many clinical applications, exercise MR stress testing of skeletal muscles has not (yet) been implemented into clinical routines. For various reasons, many studies have predominantly examined muscles in the lower leg, i.e., the medial and lateral gastrocnemius and tibialis muscles, with single‐leg (rhythmic) dorsi‐/plantarflexion motor tasks. ³¹ Other single‐limb exercise modes include knee flexion or extension exercise that recruits upper leg muscles, or handgrip exercises that involve flexor muscles of the lower arm. ³² Such exercise stress tests are typically performed in a rest‐exercise‐recovery paradigm, comprising a baseline rest period, followed by an exercise period and subsequent recovery. The exercise phase can adopt either an intermittent paradigm, consisting of a single muscle contraction followed by an interval during which MR data are recorded, or a continuous exercise paradigm, where no intervals of exercise cessation are incorporated for data acquisition. The latter paradigm poses a substantial challenge in obtaining meaningful quantitative MR data during exercise, even if acquisitions are triggered to exercise rhythm.

Standardized exercise MR stress testing protocols to investigate specific physiological effects in skeletal muscle are essentially lacking, while literature reports frequently provide limited details regarding type of exercise, intensity, and duration. Furthermore, despite the growing availability of dedicated MR‐compatible ergometers designed specifically for evaluating skeletal muscle, ³³ , ³⁴ many studies still rely on unique, in‐house built MR‐compatible ergometers. ³² Such variations between laboratories and between studies need to be considered when interpreting and comparing reports on exercise MR stress testing of skeletal muscle.

Skeletal Muscle Energy Homeostasis and pH Balance During Exercise

Adenosine 5′‐triphosphate (ATP) metabolism in in vivo skeletal muscle can uniquely be studied noninvasively during exercise with time‐resolved ³¹P‐MRS (Fig. 2). Dynamic ³¹P‐MRS allows the direct detection of inorganic phosphate (P_i) and phosphocreatine (PCr) concentration kinetics, as well as estimation of intracellular tissue pH. ³¹ Muscle contraction causes accumulation of P_i and depletes the cytosolic PCr pool, while mitochondrial oxidative metabolism replenishes this ATP energy buffer by producing ATP and via the creatine kinase shuttle in the presence of vascular oxygen supply. As such, the PCr recovery rate after aerobic exercise serves as a surrogate measure of in vivo muscle oxidative capacity, ³⁵ comprising intrinsic mitochondrial function, number of mitochondria, and oxygen transport. A comprehensive overview of relevant ³¹P‐MRS‐derived outcome parameters and of how to quantify those in skeletal muscle is provided elsewhere. ³¹

FIGURE 2.

Time‐resolved series of ³¹P‐MR spectra of the calf musculature of a healthy subject (female; 67 years) acquired at rest, during plantarflexion exercise, and during subsequent recovery at a temporal resolution of 3 s. Note the exercise‐induced 41.5% depletion of phosphocreatine (PCr) and concomitant rise of inorganic phosphate (P_i) that recovered (PCr recovery time constant, 25.4 s) upon cessation of exercise. α, β, γ‐ATP = α‐, β‐, γ‐phosphate groups in adenosine 5′‐triphosphate; PDE = phosphodiesters. Reproduced from Bakermans et al ¹⁹⁷ with permission from John Wiley & Sons, Inc. under a Creative Commons BY‐NC license (https://creativecommons.org/licenses/by‐nc/4.0/).

Many early studies employed time‐resolved pulse‐acquire surface coil‐localized ³¹P‐MRS to characterize mitochondrial ATP synthesis in health and disease. For instance, early investigations of patients with peripheral artery disease reported greater PCr depletion and intracellular acidosis during exercise, followed by delayed metabolic recovery relative to healthy subjects. Interestingly, limb blood flow measured by venous occlusion plethysmography appeared to be normal in those patients. ³⁶ , ³⁷ A similar approach was combined with incremental concentric plantarflexion exercise that induced a 50% PCr depletion in patients with dermatomyositis (n = 9) and polymyositis (n = 5) to detect substantially longer PCr and adenosine diphosphate (ADP) recovery half‐times compared with healthy volunteers, ³⁸ which can be indicative of mitochondrial dysfunction. Proton efflux, derived from ³¹P‐MRS‐based pH estimates, however, was reduced in those patients, indicating that mitochondrial dysfunction was likely secondary to impaired blood supply to the musculature. Those and other ³⁹ studies illustrate how dynamic ³¹P‐MRS during both exercise and subsequent recovery provides distinct quantitative readouts on in vivo muscle oxidative capacity in disease.

Various studies with time‐resolved ³¹P‐MRS of muscle energy homeostasis and pH balance observed high‐ and low‐pH milieus within skeletal muscles of healthy subjects during high‐intensity exercise, manifesting as an exercise‐induced and pH‐mediated “splitting” of the P_i resonance peak. ⁴⁰ , ⁴¹ It was suggested that such pH heterogeneity within active muscles reflects metabolic differences between oxidative and glycolytic muscle fibers. This particular explanation for pH heterogeneity has often been ambiguous, because signal localization to a single working muscle could often not be ascertained. ³¹ Yet, elegant experiments by Mizuno et al that used curare to selectively block motor units showed that P_i peak splitting may indeed inform on fiber type recruitment, provided that ³¹P‐MR signal can be obtained from a single working muscle with adequate blood flow. ⁴² This phenomenon proved valuable in a study of upper arm muscles of patients (n = 15) with spinal muscular atrophy (SMA). During arm‐bicycling exercise inside a 3 Tesla MR system, a differential intramuscular P_i accumulation in red (i.e., slow‐twitch oxidative) vs. white (i.e., fast‐twitch glycolytic) and intermediate myofibers was observed (Fig. 3), identifying white‐to‐red fiber type remodeling of the residual upper arm musculature in those patients. ²⁷ As such, those data demonstrated how noninvasive ³¹P‐MRS can provide evidence of the selective degeneration of fast and strongly contracting myofibers in patients with SMA, ultimately leading to muscle atrophy and weakness.

FIGURE 3.

Phosphorus‐31 magnetic resonance spectroscopy (³¹P‐MRS) of the proximal arm musculature during supine arm‐bicycling exercise. (a) ³¹P‐MR spectrum of the biceps brachii of a healthy subject acquired 1 minute after onset of supine arm‐bicycling exercise, revealing splitting of the inorganic phosphate (P_i) resonance peak that reflects myofibers operating under different pH values. (b) Phosphocreatine (PCr) and P_i concentration kinetics in red and white myofibers of the biceps brachii of a patient with spinal muscular atrophy (SMA) during arm‐bicycling until exhaustion. (c) Fractional signal amplitudes of P_i in red myofibers, and in intermediate + white myofibers as a percentage of total P_i in the biceps brachii and triceps brachii (d) muscles of the individual SMA patients and healthy subjects at end‐exercise. Note the substantially larger fraction of red myofibers in the patients with SMA. α, β, γ‐ATP = α‐, β‐, γ‐phosphate groups in adenosine 5′‐triphosphate; AU = arbitrary units; HMP = hexose monophosphate. **P < 0.01. Reproduced from Habets et al ²⁷ with permission from Oxford University Press under a Creative Commons BY‐NC license (https://creativecommons.org/licenses/by‐nc/4.0/).

While pulse‐acquire surface coil localization remains a widely favored method owing to its accessibility and effectiveness, it is important to acknowledge that signal obtained with this approach reflects all tissue within the sensitive area of the surface coil. Therefore, precise coil positioning is crucial for obtaining accurate and muscle‐specific data. Given the non‐uniform spatial distribution of both muscle physiology (eg, myofiber types) and pathology, ⁴³ resolving spatial information with ³¹P‐MRS has become increasingly important. A variety of approaches emerged to improve spatial resolution, including two‐dimensional (2D) chemical shift imaging, ⁴⁴ , ⁴⁵ multi‐coil setups to measure along the length of individual muscles, ⁴⁶ multi‐voxel localized MRS acquisitions ⁴⁷ to assess multiple muscles simultaneously, ⁴⁸ and ³¹P‐MRI to map individual metabolites. ⁴⁹ A challenge of such spatially resolved experiments is to achieve sufficient temporal resolution for following metabolite (typically P_i and PCr) concentration kinetics during exercise and recovery. Using multiple frequency‐selective radiofrequency (RF) excitations for ³¹P‐MRI at 7 Tesla, PCr and P_i signals were sampled in an interleaved fashion during isometric plantarflexion exercise at a spatial resolution of 2 mL and a temporal resolution of 6 s. ⁵⁰ Another study used multi‐voxel ³¹P‐MRS to evaluate how muscle oxidative capacity changes with age, revealing substantially slower P_i and PCr recovery kinetics in the calf muscles of elderly subjects (64.6 ± 5.8 years) after isometric plantarflexion exercise. ⁵¹ In patients with heart failure, a combined ³¹P‐MRS and ³¹P‐MRI approach at 3 Tesla was used to reveal reduced PCr recovery rates in soleus and gastrocnemius muscles after dynamic plantarflexion exercise compared with healthy subjects. ⁵² Those studies highlight how localized readouts can assess differences in oxidative capacity and intracellular tissue pH among individual muscles, emphasizing the added value of acquisitions at higher spatial resolution beyond pulse‐acquire surface coil‐localized ³¹P‐MRS. Moreover, it is practically impossible to replicate the performed exercise intensity for obtaining data on multiple muscle separately in repeated sessions.

Alternative MR methods for noninvasively evaluating skeletal muscle energy metabolism are ³¹P‐MRS magnetization transfer (MT) ⁵³ and measuring the chemical exchange saturation transfer effect of free creatine (CrCEST) with ¹H‐MRI. ⁵⁴ MT experiments have been used primarily to estimate the exchange flux between P_i and ATP as a measure of mitochondrial function in resting muscle, which may be difficult to interpret. ⁵⁵ , ⁵⁶ Instead, CrCEST has been used to measure recovery kinetics of free creatine, a byproduct of immediate ATP regeneration through PCr buffering, after exercise. In patients (n = 13) with genetic mitochondrial disorders, CrCEST at 7 Tesla revealed prolonged recovery times of free creatine returning to baseline levels in the medial gastrocnemius muscle after in‐magnet plantarflexion exercise, which is indicative of impaired oxidative capacity in these patients. ⁵⁷ At 3 Tesla, much slower free creatine recovery kinetics were found in the lateral gastrocnemius (exponential decline time constant, 274 s vs. 138 s, P = 0.01) in adults (n = 11) with Friedreich's ataxia compared with healthy subjects (n = 25), ⁵⁸ similarly suggesting reduced oxidative capacity (Fig. 4). The primary advantages of employing CrCEST instead of ³¹P‐MRS are the high signal‐to‐noise ratio that comes with ¹H‐MRI, superior spatial resolution, and no requirement for dedicated ³¹P‐MR hardware. Conversely, CrCEST solely probes free creatine, is susceptible to motion artifacts, and is constrained by a limited temporal resolution of 20–30 s.

FIGURE 4.

Series of maps of the chemical exchange saturation transfer effect of free creatine (CrCEST) acquired from the lower leg of a healthy (a) subject (male; 25 years) and a patient (male; 30 years) with Friedreich's ataxia (b) at rest and during recovery after plantarflexion exercise. Maps are ordered chronologically from left to right, then top to bottom, respectively. The color bars indicate the CrCEST signal intensity proportional to the tissue free creatine concentration. Note the free creatine concentration kinetics (c, d) for several different muscles, revealing prolonged recovery times in Friedreich's ataxia. LG = lateral gastrocnemius; MG = medial gastrocnemius; Sol = soleus. Reproduced with adapted labels from original by Schur et al ⁵⁸ under a Creative Commons BY license (https://creativecommons.org/licenses/by/4.0/).

Skeletal Muscle Perfusion

In order to sustain the mechanical function of skeletal muscle, adequate regulation of muscle perfusion to meet the local demand of oxygen and nutrients is critical. Many diseases are characterized by impairments of local peripheral microvascular function, which is often not detectable in resting skeletal muscle due to low perfusion of inactive muscles. In active muscles, the demand for oxygen and substrates increases, leading to an overall increase in muscle perfusion by 500%–1000%. ⁵⁹ As such, assessments of perfusion dynamics during and after exercise can serve as a surrogate measure for microvascular function, reflecting key aspects of the convective oxygen transport chain.

Various MRI techniques probe different components of this process, such as blood oxygenation level‐dependent (BOLD) MRI, arterial spin labeling (ASL) MRI, and dynamic contrast enhanced (DCE) MRI that requires the infusion of a gadolinium‐based contrast agent. Whereas ASL‐ and DCE‐MRI probe tissue perfusion directly and quantitatively, BOLD‐MRI is a composite readout that is sensitive to a variety of physiological processes including blood flow, blood volume, and level of oxygenation. Notably, in BOLD‐MRI dynamic changes in signal intensity or T₂* relaxation time constants are typically expressed relative to a baseline condition rather than in physiological units. Several studies have explored the origin of BOLD contrast in healthy muscle tissue by comparing BOLD‐MRI with other modalities such as near‐infrared spectroscopy (NIRS), ⁶⁰ transcutaneous oxygen pressure, ⁶¹ and skin laser Doppler flowmetry. ⁶² Those investigations demonstrated good measurement agreement between modalities in terms of hyperemic response magnitude and timing. Notably, a proposed model by Elder et al ⁶³ revealed that BOLD contrast predominantly reflects intravascular rather than extravascular effects. ⁶⁴ Although ASL‐MRI can quantify skeletal muscle perfusion in response to vasoactive stimuli, measuring perfusion in resting skeletal muscle is challenging due to the low signal‐to‐noise ratio resulting from minimal blood flow at rest. Additionally, generating a quantitative perfusion map with ASL‐MRI requires the subtraction of control‐ and label‐images, which is prone to exercise‐induced displacement artifacts. Lastly, the conventional implementation of ASL‐MRI relies on a single post‐labeling delay, thus presuming that arterial blood travels at a uniform speed, which may not accurately reflect actual physiology during exercise.

By deriving quantitative measures from dynamic series of BOLD‐, ASL‐ and DCE‐MRI, valuable insights on microvascular function in healthy and diseased skeletal muscle have been obtained during and after exercise. Using ASL‐MRI at 3 Tesla, a blunted hyperemic response evidenced by a lower % change, a lower gradient and a longer time‐to‐peak after isometric plantarflexion contractions was found in patients with peripheral artery disease, ⁶⁵ indicating impaired microvascular function in the calf musculature. Other work investigated microvascular function with BOLD‐MRI at 3 Tesla during and after submaximal (50% of maximal voluntary contraction) and maximal isometric dorsiflexion exercise in obese subjects (n = 8) and in patients (n = 8) with type 2 diabetes mellitus. Compared with lean age‐matched subjects (n = 8), blood volume was lower after maximal but not submaximal contractions in the extensor digitorum longus muscle of subjects with obesity and diabetes, suggesting exercise intensity‐dependent structural or functional impairment of the microvasculature. ⁶⁶

The arterial transit time (ATT) quantifies the delivery of blood to the muscles by measuring the time delay for arterial blood to travel into the tissue microvasculature. Using DCE‐MRI, progressive shortening of the ATT was found in healthy volunteers during recovery after plantarflexion exercise against increasing workloads (Fig. 5). ⁶⁷ Those findings challenge the conventional implementation of ASL‐MRI ⁶⁸ and emphasize the importance of adequate or multiple post‐labeling delays. In healthy volunteers, ASL‐MRI with multiple post‐labeling delays was used to characterize recovery effects after dynamic dorsiflexion exercise, revealing proximally elevated perfusion and a shorter ATT compared with distal tissue. ⁶⁹ Those data emphasize that multiple post‐labeling delays are needed to quantify skeletal muscle perfusion with ASL‐MRI, and highlight how microvascular function heterogeneity within individual muscles can be studied with exercise MR stress testing of skeletal muscles. Such observations are particularly important for clinical applications, given that pathology can considerably affect ATT. ⁶⁷

FIGURE 5.

Maps of the arterial transit time (ATT) derived from dynamic contrast enhanced (DCE) MRI in calf muscles of a young healthy subject after dynamic plantarflexion exercise against three different loads: 4 lbs (a), 8 lbs (b), and 16 lbs (c). Note how ATT values vary substantially between the anterior and posterior compartments, the individual triceps surae muscles (gastrocnemii and soleus), and with different workloads, showing lower ATT values (i.e., faster perfusion) with more active muscles. AT = tibialis anterior; LG = lateral gastrocnemius; MG = medial gastrocnemius; S = soleus. Reproduced from Conlin et al ⁶⁷ under a Creative Commons BY license (https://creativecommons.org/licenses/by/4.0/).

Alternatively, intravoxel incoherent motion (IVIM) MRI utilizes spin dephasing‐induced signal loss to visualize and quantify various perfusion‐related parameters including perfusion fraction (f _p ), water diffusion (D) and pseudo‐diffusion (D*). ⁷⁰ Compared with BOLD‐, ASL‐ and DCE‐MRI, IVIM‐MRI has a substantially lower temporal resolution of 2–3 minutes. Similar to ASL‐MRI, IVIM‐MRI measurements at rest are challenged by low resting‐state perfusion of skeletal muscle. Only a few studies used in‐magnet exercise, because contrast for IVIM‐MRI is typically generated by comparing baseline and post‐exercise conditions. Using IVIM‐MRI at 3 Tesla following in‐magnet incremental knee extension exercise, a blunted increase (+18 ± 16%) in perfusion was observed (Fig. 6) in upper leg muscles of elderly (60–90 years; n = 4) compared with healthy young subjects (+37 ± 12%; 21–30 years; n = 4). ⁷¹ Those results are in line with other reports that revealed aging‐related reduced muscle perfusion as measured with ASL‐MRI after moderate‐intensity plantarflexion exercise in older subjects, ⁷² and a blunted hyperemic response after vascular occlusion as measured with BOLD‐MRI. ⁷³

FIGURE 6.

Intravoxel incoherent motion (IVIM) MRI‐based perfusion fraction and water diffusion maps from the upper leg of a young (27 years) and an old (90 year) subject at rest and after dynamic knee extension exercise at 3.25‐minute intervals. Regions of interest for the active rectus femoris muscle (red) and for a part of the inactive adductor magnus muscle (white) are indicated. Note the pronounced difference in perfusion fraction between the young and old subjects after exercise. Reproduced from Adelnia et al ⁷¹ with permission from John Wiley & Sons, Inc.

Beyond assessments of microvascular perfusion, (de)oxygenation of myocytes can be probed through the detection of deoxymyoglobin (DMb) with ¹H‐MRS. ⁷⁴ At rest, typically no DMb signal can be observed because myoglobin will be fully oxygenated. Instead, during arterial occlusion ⁷⁴ or after maximal exercise, ⁷⁵ a clear DMb signal arises. Such DMb signal is a composite measure capturing aspects of both convective and diffusive oxygen transport chains, and holds potential as a quantitative parameter that reflects tissue reoxygenation (deficits). ⁷⁶ Indeed, a study in patients with glycogen storage disease used an interleaved ¹H‐MRS, ASL‐ and BOLD‐MRI approach at 4 Tesla to reveal delayed reoxygenation and concomitantly blunted microvascular response after dynamic plantarflexion exercise, suggesting that reduced tissue perfusion may contribute to impaired oxidative capacity and exercise intolerance in those patients. ⁷⁷ Compared with other methods outlined above, the detection of DMb with ¹H‐MRS suffers from low sensitivity and does not provide spatial information. As such, it may not be the readout of choice during an exercise MR stress test. Indeed, tissue oxygenation states may now preferably be assessed clinically and much more reliably using novel optical equipment, i.e., with diffuse correlation spectroscopy ⁷⁸ , ⁷⁹ and photo‐acoustic imaging. ⁸⁰ However, multi‐parametric dynamic MR that includes the acquisition of signal from DMb has recently been implemented on a clinical setup, allowing the evaluation of tissue (re)oxygenation in conjunction with, rather than at the expense of, other readouts such as BOLD‐MRI and ³¹P‐MRS. ⁸¹

As an alternative to physical exercise, cuff occlusion can be used for ischemia‐hyperemia paradigms in order to noninvasively evaluate microvascular reactivity with MRI. Compared to exercise protocols, cuff occlusions are highly reproducible, well‐controlled, and do not rely on active participation from the subject. Although such occlusions do not trigger the hemodynamic and metabolic responses similar to daily‐life physical activities, it may be a valuable alternative for obtaining data on microvascular function in patients who are unable to perform exercise inside an MR scanner. Currently, no standardized cuff occlusion protocols exist, hampering fair comparisons of data from different laboratories and different studies.

Several studies have assessed the precision of these MRI techniques for measuring muscle perfusion, ⁸¹ , ⁸² , ⁸³ yielding variable results for both ischemia‐hyperemia and exercise protocols. To advance the clinical utility of these techniques, a clear perspective of the measurement precision and reproducibility is essential. Precision can be improved by enhancing signal‐to‐noise ratios, and temporal and spatial resolution, potentially through machine learning and accelerated imaging approaches, while reproducibility will improve with protocol standardization. More details about how techniques for functional MRI of skeletal muscles relate to one another and to muscle physiology, and more examples of clinical applications can be found elsewhere. ⁸³ , ⁸⁴

Skeletal Muscle Contractile Function

Conventionally, muscle contractile function has been evaluated by surface or needle electromyography‐based measurements, assessing a muscle's electrical response to incoming stimuli from motor‐neurons resulting in the activation of single or multiple motor‐units. ⁸⁵ With MRI, contractility measures in terms of tissue displacements and derived strains, velocities and strain rates resulting from these electrical activities in active skeletal muscle are visualized and quantified. This is particularly relevant for investigating degenerative neuromuscular disorders, which are characterized by a lower quality of the remaining muscle tissue that is likely caused by altered contractile function.

The spatial resolution of MRI is typically in the order of millimeters, and precluding discerning individual sarcomeres, but it allows the characterization of muscle fascicle displacements and whole‐muscle mechanics. To date, only one study compared surface electromyography with MRI during isometric plantarflexion exercise, reporting good agreement (0.84 < r > 0.88) between contraction dynamics measured by 2D velocity‐encoded phase‐contrast MRI and surface electromyography data. ⁸⁶

By measuring displacements and velocities using MRI in skeletal muscle during exercise, quantitative estimates of muscle contraction dynamics can be obtained. ⁸⁷ From such velocity and displacement data, strain rate quantifies the change in deformation over time, while principal strains are derived by integrating strain rate over time. The latter calculation is valid under the assumption that the voxels of the acquired slice do not move out of the imaging plane. This can often not be ascertained for concentric and eccentric contractions, and therefore most of the displacement‐ and velocity‐encoded MR measurements are typically conducted during isometric contractions that vary from 5% to 60% of maximal voluntary contraction. The sensitivity of such velocity‐ and displacement‐encoded MR techniques in relation to muscle contraction intensity was demonstrated in healthy subjects, detecting contractile heterogeneity between and within individual muscles. ⁸⁸ , ⁸⁹ , ⁹⁰ Those techniques provided valuable insights on muscle contractile function in clinical conditions. Using 2D velocity‐encoded phase‐contrast MRI at 1.5 Tesla, it was shown (Fig. 7) that athletes with a prior hamstring injury exhibited higher principal strains in the whole muscle as well as near the proximal myotendinous junction during active hamstring lengthening and shortening contractions. ⁹¹ Those results suggest that post‐injury remodeling may adversely affect local tissue mechanics and could contribute to the high re‐injury risk in those athletes. With a similar approach at 3 Tesla, lower strain rates were measured during isometric plantarflexion exercise in elderly, and attributed to aging‐related remodeling of the extracellular matrix. ⁹² In disuse atrophy induced by lower limb suspension in healthy volunteers, strain rates during isometric plantarflexion contractions decreased in the myofiber cross‐sectional direction, ⁹³ which was associated with loss of maximal contractile force. Those findings suggest extensive extracellular matrix remodeling in disuse atrophy that affects muscle contractility, illustrating how MRI can noninvasively inform on muscle mechanics.

FIGURE 7.

First principal strain maps from the biceps femoris during peak knee extension under inertial loading conditions overlaid on coronal anatomical MR images obtained in the upper leg of an uninjured athlete and an athlete with prior hamstring injury. Larger strains were observed in the tissue nearest to the myotendinous junction. Note that strains were substantially larger in athletes with prior hamstring injuries compared to uninjured subjects. *P < 0.05. Reproduced from Silder et al ⁹¹ with permission from Elsevier.

Until now, studies predominantly used single or several imaging planes with 2D tensors to mitigate long acquisition times typically required for three‐dimensional (3D) volumetric acquisitions, even though it is known that both architectural and functional characteristics vary between as well as within individual muscles. ⁹⁴ , ⁹⁵ Furthermore, it is generally impossible to align the imaging plane precisely with all muscle fibers. Advancements in acquisition and reconstruction strategies enabled accelerated MRI for facilitating multi‐slice and volumetric whole‐muscle evaluations (Fig. 8) of the 3D strain rate tensor with appropriate temporal resolution (10 msec) and acceptable acquisition times. ⁹⁶ , ⁹⁷ , ⁹⁸ More research in healthy subjects and patients is essential to truly determine the added value of evaluating whole‐muscle volumes and contraction dynamics during exercise in 3D.

FIGURE 8.

Cross‐sectional maps of strain rate along the myofibers during maximal dorsiflexion (top) and maximal plantarflexion (bottom) of a distal (segment 20%–40%), middle (segment 40%–60%), and proximal (segment 60%–80%) slice of the lower leg. Line graphs (right) showing average strain rates along the fiber in the extensor digitorum longus (top) and gastrocnemius lateralis (bottom) muscles throughout the unloaded exercise cycle for the five segments. Note the generally antagonistic behavior of the anterior and posterior compartments of the lower leg during exercise, as well as the difference between the individual segments. See Hooijmans et al ⁹⁸ for details.

As an alternative to probing tissue displacements, the visualization and quantification of contraction‐induced changes in the skeletal muscle microstructure could serve as an outcome measure for contractile function. However, conventional diffusion tensor imaging (DTI) MRI acquisitions for mapping muscle microstructure prove unsuitable for this purpose, given their high sensitivity to bulk motion of both voluntary and involuntary contractions, which leads to substantial signal voids. Recently, Mazzoli et al used an oscillating gradient spin echo (OGSE) diffusion encoding strategy. ⁹⁹ By employing short diffusion weighting times and trapezoid‐cosine OGSE waveforms, this technique allowed the quantification of tissue water diffusion properties in actively contracting skeletal muscle (isometric contractions; acquisition time 90 s) of healthy volunteers. As such, this approach holds potential for noninvasively investigating contraction abnormalities in patients with neuromuscular disorders. Conversely, diffusion‐weighted MRI can also utilize the signal voids that are introduced by muscle contraction ¹⁰⁰ for “motor unit MRI”. This emerging area of research has recently been reviewed elsewhere. ¹⁰¹

Physical Exercise and the Heart

Exercise stress testing of the cardiovascular system, and the heart in particular, constitutes an important part of the clinical evaluation of patients with exercise‐related complaints. While conventional (i.e., with electrocardiography [ECG] only) exercise tests are no longer part of the standard work‐up for ischemic heart diseases, CPETs that measure respiratory oxygen uptake, carbon dioxide production, systemic blood pressures, and ECG are increasingly being used to evaluate symptomatic patients and to quantify cardiorespiratory fitness. ⁸ Transthoracic echocardiography is the imaging workhorse of clinical cardiology to noninvasively evaluate and quantify cardiac performance. Under exercise stress, the image quality of transthoracic echocardiography may deteriorate substantially, particularly in obese subjects, and views of the right ventricle (RV) are difficult to obtain. Such evaluations with MRI under exercise stress are challenging as well. Yet, already in an early stage of the rise of MRI as a clinical imaging modality, devices were designed that would allow for exercise MR stress testing of the heart ²⁰ as a potential alternative to pharmacological stress. The strong magnetic field as well as the narrow bore of the MR system put substantial constraints on the design and construction of such devices, limiting the flexibility and increasing the costs of this approach. Likewise, noninvasive blood pressure measurements cannot be obtained with standard monitors. In addition, safety ¹⁰² and technical ¹⁰³ concerns can limit the feasibility and efficacy of exercise MR stress testing in patients with implantable electronic devices such as pacemakers and defibrillators. Taken together, these constraints make MRI under exercise stress technologically demanding. Yet, four decades of technological advances have now brought the field to an era where MR‐compatible ergometers are commercially available, ¹⁰⁴ , ¹⁰⁵ supine bicycling exercise stress testing is included in guidelines on standardized cardiovascular MRI protocols, ¹⁰⁶ and (patho)physiological cardiovascular responses to exercise have been investigated with a broad range of noninvasive MR readouts. ¹⁰⁷

Blood Flow in the Great Vessels

The first MRI estimates of cardiovascular performance under exercise conditions were obtained from measurements of blood flow in the great vessels. Using a 0.5 Tesla MR system and an apparatus for supine in‐magnet pedaling exercise, velocity maps of the descending thoracic aorta were obtained in healthy volunteers (n = 10) at rest (heart rate, 68 ± 6 beats/min) and immediately after exercise (101 ± 12 beats/min). ¹⁰⁸ At a temporal resolution of 50 msec and an in‐plane spatial resolution of 5 × 5 mm², the phasic nature of systolic aortic blood flow was demonstrated, revealing a >40% elevation of mean and peak flow immediately after exercise (Fig. 9).

FIGURE 9.

Peak systolic velocity maps of the descending thoracic aorta (arrow) and the right (1) and left (2) ventricular outflow tracts acquired at 0.5 Tesla at rest (a) and after exercise (b), showing blood flow velocity in cranial and caudal directions in light and dark shades of gray, respectively. Aliasing at high velocities caused the ventricular outflow tracts to appear black after exercise. Note the phasic nature of systolic blood flow in the aorta, and the elevated peak flow after exercise (c). Reproduced and relabeled from Mohiaddin et al ¹⁰⁸ with permission from John Wiley & Sons, Inc. © SMR, 1995.

At 1.5 Tesla and with a commercially available supine bicycle ergometer, 2D phase‐contrast MRI (1.5 × 1.5 mm² in‐plane spatial resolution, 28 msec temporal resolution) of the abdominal aorta showed an increased flow from 1.4 ± 0.3 L/min at rest (65 ± 7 beats/min) to 7.9 ± 1.1 L/min after exercise (135 ± 22 beats/min) at 131 W. ¹⁰⁹ Retrograde flow present at rest disappeared at higher exercise intensity. Measurements were obtained during breath holds immediately after exercise, in order to minimize motion‐induced distortions of the ECG signal that was used for triggering, while accepting that heart rate drops rapidly ¹¹⁰ by <14% after high‐intensity exercise in those young (27 ± 2 years; n = 9) healthy volunteers. ¹⁰⁹ This approach could potentially be used for assessments of the pulse wave velocity as a surrogate measure for vascular stiffness. Notably, descending aorta pulse wave velocity after exercise correlated positively with age (22–75 years; n = 50, r = 0.63, P < 0.001), ¹¹¹ and main pulmonary arterial wall stiffness increased in response to acute exercise ¹¹² in healthy volunteers. Those data suggest that evaluations of the great vessels under exercise stress can provide information on vascular function, eg, in atherosclerosis and pulmonary hypertension. ¹¹³ Detailed spatial analyses of vascular hemodynamics will be feasible with 4D (i.e., three spatial dimensions and time‐resolved) flow MRI during exercise at 3 Tesla, ¹¹⁴ which was recently used to visualize altered vortices in diastolic RV filling in preterm‐born but otherwise apparently healthy adolescents (n = 16) and adults (n = 10). ¹¹⁵

Flow measurements of the great vessels allow for a quantitative evaluation of flow distribution and any changes therein during exercise. ²⁸ This is particularly relevant in patients with a Fontan circulation, who can suffer from exercise intolerance. Pedersen et al used 2D phase‐contrast MRI in pediatric patients to show that the increased flow from the inferior vena cava remained equally distributed to both lungs immediately after exercise, indicating that vascular resistance rather than geometry is a major determinant of pulmonary flow after total cavopulmonary connection surgery. ¹¹⁶ Later, real‐time (i.e., without ECG triggering) 2D phase‐contrast MRI of different imaging planes enabled the assessment of flow in the aorta, the inferior and the superior vena cava in a free‐breathing regime during steady‐state supine bicycling exercise. ¹¹⁷ Detailed analyses of the complex physiology during inspiration, expiration, and during exercise revealed how inspiration drives blood flow in the inferior vena cava at rest, but not during supine leg exercise when peripheral muscle contraction appears to be an important contributor to driving the circulation in those patients. Using a similar approach by acquiring 2D phase‐contrast MRI data of several abdominal vessels, the complex and differential hemodynamic responses to moderate‐intensity exercise in patients with abdominal aortic aneurysms and with peripheral artery disease were investigated. ¹⁹

By measuring flow in the ascending aorta and pulmonary artery during exercise, ¹⁰⁴ , ¹¹⁸ respective surrogate estimates of left ventricular (LV) and RV stroke volumes can be obtained. ¹¹⁹ , ¹²⁰ This way, cardiac output can quantified without the need for acquiring multiple ventricular cine MRI series. ¹²¹ Phase‐contrast MRI following exercise stress testing revealed a stiffer aortic wall as well as a blunted exercise‐induced augmentation of LV stroke volume in pulmonary hypertension compared with healthy subjects. This lack of stroke volume reserve was compensated for by a higher heart rate in order to maintain cardiac output. ¹²² Real‐time 2D phase‐contrast MRI of blood flow in the ascending aorta has also been used to benchmark stroke volume estimates based on short‐axis cine MR imaging during exercise. ¹²³ Note that stroke volume estimates based on blood flow in these great vessels do not necessarily reflect the volumetric change of the ventricles, particularly in some scenarios where exercise intolerance is an important phenotype, eg, due to valve disease ¹²⁴ or conditions that cause shunting.

Cardiac Function

Cardiac morphology and ventricular function is commonly quantified by multiplying the segmented surface areas of the ventricular blood pools and myocardial wall with the slice thickness and optional slice gap to obtain systolic and diastolic volumes and derived parameters ¹²⁰ from a stack of contiguous LV short‐axis 2D cine MRI series. Such datasets are acquired under breath‐hold conditions at rest over multiple heart beats using (retrospective) ECG gating ¹⁰⁶ in order to achieve the spatial and temporal alignment of the heart in contiguous and consecutive images that is important for accurate volumetry. In‐magnet physical exercise inherently comes with body movement and increased respiration, and the ECG signal becomes unreliable due to distortions by movement of the torso through the static magnetic field, potential heart rate fluctuations, and perspiration. These aspects preclude obtaining cine MRI series with ECG gating ¹⁰⁶ during exercise. Studies have tried to remedy these issues by acquiring data during breath holds in brief intervals of exercise cessation. ²² , ¹²⁵ Such breath holds are difficult to perform, particularly by patients or with high‐intensity exercise, although cine MRI has even been performed with breath holds during exercise in healthy young adults. ¹²⁶

Acquisitions under free breathing conditions during exercise have been performed with “real‐time” MRI ¹²⁷ in order to visualize dynamic processes such as myocardial contraction and relaxation. Using radial k‐space undersampling and sensitivity encoding (SENSE) for parallel imaging in a radial k‐t SENSE sequence, Lurz et al acquired real‐time short‐axis cine MR images at a temporal resolution of 35 msec during various levels of steady‐state pedaling exercise (heart rates up to 154 ± 13 beats/min) without breath holds. ¹²⁸ The biventricular response to supine exercise in healthy volunteers was visualized, showing augmented LV and RV stroke volumes through a reduction in end‐systolic volumes (ESV) rather than an increase in end‐diastolic volumes (EDV); observations that were later confirmed in a meta‐analysis of similar studies. ¹²⁶ Although real‐time MRI mitigates movement disturbances and gating issues for each image, it typically generates a large number of images during minutes of data acquisition that still need to be spatially and temporally aligned according to their respiratory and cardiac phase for volumetric quantifications. These processing steps have been performed visually ¹²⁸ , ¹²⁹ or with the aid of separately recorded respiratory and ECG signals, ¹³⁰ but are labor intensive even if respiratory motion is ignored. ¹³¹ One approach to reduce this processing burden is to lower the number of different imaging planes. ¹²⁰ Rather than acquiring a stack of contiguous short‐axis slices covering the heart, orthogonal long‐axis planes can be used to estimate (left) ventricular volumes during exercise. ¹²⁶ , ¹³² Recent efforts have focused on (semi‐)automating the processing of real‐time MR images, eg, through principal component analysis of image signal intensities to extract the respiratory signal, ¹³³ , ¹³⁴ or based on periodic signal intensity fluctuations across the liver‐lung interface as a retrospectively applied respiratory “navigator”. ¹²³ Nonetheless, to capture end‐systolic and end‐diastolic phases at high heart rates, MRI at high temporal resolution is needed, because low temporal resolution leads to underestimations of cardiac performance. ¹³⁵ Accelerated image acquisition and increased temporal resolution (<30 msec) have been achieved through radial imaging, ¹³⁶ providing retrospective flexibility in temporal resolution during image reconstructions. ¹³⁷

Building on their approach to quantify biventricular volumes during high‐intensity bicycling exercise at 1.5 Tesla, ¹³⁰ La Gerche et al have since conducted a series of studies in various populations. ¹⁰⁷ In endurance athletes, the occurrence of ventricular arrhythmias may be related to the large hemodynamic load put on the RV during high‐intensity exercise, inducing transient RV dysfunction and potentially chronic RV remodeling. By utilizing the good visibility of both ventricles on MRI, it was demonstrated that in athletes with arrhythmias (n = 9), RV contractile dysfunction arose during exercise stress (i.e., attenuated reduction in RV ESV), while cardiac function appeared normal at rest. ¹³⁸ Cardiac dysfunction and reduced exercise capacity are potential adverse consequences of anthracycline‐based chemotherapy in breast cancer. In a randomized clinical trial investigating the impact of a 12‐month exercise training program on cardiorespiratory fitness in women (n = 102) with breast cancer who received such chemotherapy, MRI‐estimated cardiac reserve (i.e., exercise‐induced changes in stroke volumes, cardiac output, ejection fractions) served as a quantitative outcome measure. ¹³⁹ Exercise training improved cardiac reserve, while cardiac reserve declined in the usual‐care arm of the study, providing direct evidence that exercise training during anthracycline‐based chemotherapy can prevent cardiac dysfunction. Moreover, changes in cardiac reserve were mirrored by improvement or decline of VO₂max, respectively. Such studies highlight the quantitative sensitivity of MRI‐based volumetry during exercise, and how its noninvasive nature readily allows for longitudinal measurements in patients as well as healthy volunteers. Many other examples of applications for ventricular volumetry during exercise, such as in patients with a Fontan circulation ¹⁴⁰ or repaired Tetralogy of Fallot, ¹²⁹ have been reviewed elsewhere. ¹⁰⁷

Assessments of myocardial strain can serve as measures of myocardial deformation and (dys)function beyond volumetric analyses, particularly through quantifications based on feature tracking of the endocardial walls in cine MRI. ¹³⁵ , ¹⁴¹ Global longitudinal strain (GLS) can be assessed from long‐axis views and is sensitive to changes in contractility, with more negative values indicating higher contractility (Fig. 10). With exercise, normal LV GLS ¹³² and RV GLS ¹⁴² decrease from approximately −20% at rest to nearly −30%, while this exercise‐induced decrease is attenuated with age ¹³² and in the RV of patients with pulmonary hypertension, ¹⁴² indicating reduced contractile reserve. As an alternative to MRI feature tracking, others have estimated LV long‐axis strain by manually measuring the distance between the mitral valve and the apex in end‐diastolic and end‐systolic images. ¹⁴³ This approach avoids elaborate processing steps to obtain the complete cardiac cycle for feature tracking of the myocardial walls, particularly with real‐time MRI during exercise. Backhaus et al used supine bicycling exercise at 3 Tesla in a clinical trial to establish the accuracy of long‐axis strain measurements for HFpEF diagnosis. ¹⁴³ Left atrial long‐axis strain derived from real‐time MRI correlated strongly with the pulmonary capillary wedge pressures (PCWP) measured invasively with right heart catheterization during exercise (n = 75, r = −0.75, P < 0.001) as an indicator of impaired diastolic emptying in HFpEF during exercise. As such, their study showed that real‐time MRI during physiological bicycling exercise can qualify as a noninvasive alternative to the current reference test of invasive right heart catheterizations for diagnosing HFpEF.

FIGURE 10.

Example of how real‐time cine MRI during supine bicycling exercise stress can unmask myocardial dysfunction. Note that left ventricular (LV) global longitudinal strain (GLS) measured along the endocardial LV wall is augmented (green arrow) upon exercise in the healthy subject (male; 28 years). In the patient (male; 70 years) with heart failure with preserved ejection fraction (HFpEF), LV GLS appears normal at rest (orange), but fails to increase during exercise (red). Data courtesy of the Department of Radiology and Nuclear Medicine, Amsterdam University Medical Centers, Amsterdam, The Netherlands.

While volumetry and global strain analyses provide whole‐chamber functional assessments, more detailed evaluations of myocardial wall deformation can be obtained through MR tagging. ¹⁴⁴ Using a radial tag pattern, real‐time MRI at 3 Tesla was used to reveal substantial diastolic recoil of the LV during early filling in a volunteer (n = 1) at low‐intensity bicycling exercise. ¹⁴⁵ Note that although imaging was done in real‐time, breath holds and triggering to the ECG signal were still required to accurately apply the tag pattern. Recently, real‐time MRI with ECG‐triggered tag line preparation at a temporal resolution of 29 msec was introduced for use immediately after supine bicycling exercise, showing a blunted response of LV deformation to exercise in HFpEF. ¹⁴⁶ Assessments of the regional response of myocardial deformation to exercise may aid in the evaluation of functional consequences of microvascular dysfunction, that may otherwise remain undetected at rest.

Myocardial Perfusion

In patients with suspected coronary artery disease, perfusion imaging is an important approach for diagnosis and risk stratification. ¹⁴⁷ Stress is then used to evoke myocardial ischemia, most commonly through intravenous infusion of a vasodilator (eg, adenosine) or alternatively with an inotropic stress agent (eg, dobutamine). During such pharmacological stress with MRI, the first pass perfusion of a bolus of an gadolinium‐based contrast agent is dynamically imaged, revealing any hypoperfused regions and perfusion defects as an indicator of inducible myocardial ischemia. ¹⁴⁸ While pharmacological stress testing is sensitive, highly reproducible, and allows the subject to lie still during image acquisition, it does not reflect the hemodynamic and neurohormonal response to exercise or the physical activity that may induce exertional symptoms in the patient's daily life. First attempts to address these limitations involved upright treadmill exercise adjacent to the MR system, with the subjects quickly transferred back to their original position inside the MR bore after exercise followed by image acquisition within 1 minute. ¹⁶ In later work, an MR‐compatible ergometer was mounted on the MR table to reduce the transition time between exercise and imaging. ¹⁴⁹ Supine bicycling exercise was performed with the table moved out of the bore, after which the table was repositioned into the magnet for imaging of myocardial perfusion shortly after exercise. Accelerated MR imaging was achieved through a radial steady‐state free precession sequence in combination with respiratory motion tracking ¹⁴⁹ and advanced reconstruction methods for undersampled data, ¹⁵⁰ allowing for an evaluation of myocardial first pass perfusion under free‐breathing conditions. Measurements of first pass perfusion with MRI after treadmill exercise have shown diagnostic merit over treadmill stress testing with single photon emission computed tomography (SPECT) of patients (n = 210) with known or suspected coronary artery disease in a multi‐center clinical trial. ¹⁷ In another clinical trial, the capacity of exercise MR stress testing to detect defects and regional wall motion abnormalities in patients (n = 60) with suspected coronary artery disease was benchmarked against invasive fractional flow reserve (FFR) measurements with angiography, ¹⁵¹ again showing potential to diagnose coronary artery disease noninvasively. Note that although the intravenous administration of a pharmacological stress agent is no longer needed with physical exercise stress, first pass perfusion measurement still requires infusion of a gadolinium‐based contrast agent as a blood flow tracer.

Changes in the myocardial native T₁ relaxation time constant that occur with changes in myocardial blood flow or volume ¹⁵² have been utilized for the detection of ischemia under adenosine stress without the need for a gadolinium‐based contrast agent. ¹⁵³ Following this concept, Nakamori et al then showed that myocardial native T₁ is transiently elevated by 6 ± 3% within 30 s after supine bicycling exercise on an MR‐compatible ergometer mounted on the table of a 1.5 Tesla MR system. ¹⁵⁴ Importantly, the magnitude of T₁ reactivity correlated (n = 28, r = 0.62, P < 0.001) with the maximum heart rate—blood pressure product, an index of myocardial oxygen consumption during exercise. ¹⁵⁵ Those data suggest that native T₁ mapping can provide a surrogate marker of any physiological change in myocardial blood flow during exercise. Indeed, in patients (n = 14) with coronary artery disease, native T₁ reactivity to exercise mirrored the severity of myocardial perfusion abnormalities confirmed with SPECT. ¹⁵⁴ Using a “free‐running” radial spoiled gradient echo sequence with respiratory navigator readouts and retrospective ECG gating, ¹⁵⁶ exercise‐induced T₁ reactivity of +5% to +10% in healthy volunteers was measured at 3 Tesla (Fig. 11). Importantly, all studies with spatial information on myocardial perfusion, either through first pass perfusion imaging or with T₁ parameter mapping, acquired their data quickly after exercise. Such measurements during physical exercise have not yet been demonstrated. Challenges to do so involve the high heart rate on top of fast and irregular or deep respiration, as well as potential upper body movement during leg exercise, all compromising the alignment of consecutive images that is required for quantitative MR of the myocardium, while the exercise protocol needs to be performed inside the MR system for simultaneous measurements. With low‐intensity isometric hand grip exercise that allows in‐magnet exercise without body movement, an increase (+50.7 ± 31.4%) in peak coronary blood flow during was measured with phase‐contrast MRI in the left anterior descending coronary artery of normal volunteers (n = 9) at 1.5 Tesla. ¹⁵⁷ This approach was later used at 3 Tesla to unmask a reduction in coronary flow in patients with coronary artery disease, demonstrating the ability to assess coronary endothelial (dys)function noninvasively. ¹⁵⁸

FIGURE 11.

Maps of apparent myocardial T₁ (T₁*) values in two healthy volunteers at rest and after two bouts of vigorous supine bicycle exercise on an MR‐compatible ergometer mounted on the table of a 3 Tesla MR system. Note the increase and subsequent gradual decrease of myocardial T₁* values post‐exercise, indicating an exercise‐induced elevation of myocardial perfusion. bpm = beats per minute; HR = heart rate. Reproduced from Guo et al ¹⁵⁶ with permission from John Wiley & Sons, Inc. © 2022 International Society for Magnetic Resonance in Medicine.

Myocardial Energy Metabolism

Nearly four decades ago, among the first applications of MR to study the human heart was measuring phosphate metabolite levels in the in vivo myocardium with ³¹P‐MRS. ¹⁵⁹ It was soon realized that such “measurements at rest did not clearly differentiate between normal and diseased myocardium,” ¹⁶⁰ and that studies during exercise would be needed. Today, essentially all ³¹P‐MRS studies follow the pioneering work from the early 1990s, ¹⁶¹ , ¹⁶² , ¹⁶³ and assess myocardial high‐energy phosphate metabolism by quantifying the PCr over ATP signal ratio as a measure of the in vivo myocardial energy status, while the signal from myocardial P_i cannot be resolved. Because ³¹P‐MRS of the heart requires multiple transients to achieve a sufficient signal‐to‐noise ratio as well as spatial localization (eg, with a multi‐shot 3D image‐selected in vivo spectroscopy [ISIS] sequence ¹⁶⁴ ), such acquisitions are lengthy (i.e., multiple minutes) relative to single‐shot dynamic ³¹P‐MRS acquisitions in skeletal muscle (temporal resolution of seconds, see above). This precludes any dynamic detection of how the PCr/ATP ratio changes during transitions between cardiac work rates. Myocardial PCr/ATP ratios therefore only inform on steady‐state conditions of myocardial energy and proton balance, and are measured during submaximal exercise only.

Given these methodological challenges of cardiac ³¹P‐MRS, a change of the myocardial PCr/ATP ratio during exercise or recovery in healthy subjects has not been detected with ³¹P‐MRS. Most studies have been conducted at low‐ or moderate‐intensity exercise, such as isometric hand gripping ¹⁶¹ or prone knee flexion. ¹⁶⁰ Moreover, measurement variability increases substantially during exercise, ¹⁶⁵ further degrading the sensitivity for detecting any exercise‐induced reduction of the myocardial PCr/ATP ratio that computational modeling estimated to be maximally <10% for healthy subjects at maximal exercise intensity. ¹⁶⁶ In contrast, in various patient populations, localized ³¹P‐MRS of the heart has successfully been applied during exercise to expose impaired myocardial energy homeostasis. Using a 1.5 Tesla MR system, Weiss et al showed that the PCr/ATP ratio of the anterior myocardium in patients (n = 16) with coronary artery disease transiently decreased (−37%) from 1.45 ± 0.31 at rest (77 ± 13 beats/min) to 0.9 ± 0.24 during isometric hand grip exercise (89 ± 16 beats/min) ¹⁶¹ ; observations that were later replicated by others. ¹⁶⁷ , ¹⁶⁸ Moreover, revascularization therapy restored the impaired myocardial energy homeostasis that was observed during exercise. ¹⁶¹ In patients (n = 35) with hypertrophic cardiomyopathy (HCM) due to a pathogenic mutation, ³¹P‐MRS revealed a reduced myocardial energy status at rest that exacerbated upon dynamic prone knee flexion (90 ± 12 beats/min). ¹⁶⁹ This observation supports the notion that inefficient mechanical function due to sarcomere mutations increases the myocardial energy demand for maintaining adequate cardiac pumping performance. ¹⁷⁰ In a scenario energy supply is compromised, Levelt et al showed (Fig. 12) that myocardial PCr/ATP was 17% lower in type 2 diabetes mellitus patients (n = 31) at rest, which further decreased by 12% during moderate‐intensity exercise (69 ± 8 beats/min), attributed to hypoperfusion that was demonstrated separately with BOLD‐MRI under adenosine stress. ¹⁷¹ These studies show how noninvasive measurements of the in vivo myocardial energy status during exercise can inform on myocardial pathology.

FIGURE 12.

Phosphorus‐31 MR spectra (³¹P‐MRS) acquired at rest and during steady‐state prone repeated and alternate knee flexion exercise inside a 3 Tesla MR system in a healthy subject and a patient with type 2 diabetes mellitus. Note the low myocardial PCr/ATP ratio in type 2 diabetes mellitus at rest that decreases further upon moderate‐intensity exercise, indicating impaired myocardial energy homeostasis. α, β, γ‐ATP = α‐, β‐, γ‐phosphate groups in adenosine 5′‐triphosphate; PCr = phosphocreatine. Reproduced and rearranged with adapted labels from original by Levelt et al ¹⁷¹ under a Creative Commons BY license (https://creativecommons.org/licenses/by/4.0/).

Not surprisingly, exercise ³¹P‐MRS of the heart has not yet been translated to the clinic. Measurements are lengthy, prone to (motion) artifacts and suffer from poor sensitivity, and require dedicated hardware and specific expertise that is not typically available. There are large differences in reported PCr/ATP ratios between studies and between sites, ¹⁶⁶ and reference values are lacking. More recent efforts have focused on reducing acquisition times for higher temporal resolution and improving measurement sensitivity and precision by using 7 Tesla MR systems. ¹⁷² , ¹⁷³ , ¹⁷⁴ Using single‐shot single‐voxel stimulated echo acquisition mode (STEAM) localization to suppress the 2,3‐diphosphoglycerate (2,3‐DPG) signal in flowing ventricular blood, the detection of myocardial P_i has been reported. ¹⁷⁵ If confirmed, this would allow for noninvasive estimations of myocardial pH changes. ¹⁷⁶ Moreover, a rise in P_i concentration may be a more sensitive marker of impaired myocardial energy homeostasis than the PCr/ATP ratio. ¹⁶⁶ Yet, a first study in healthy subjects (n = 17) under dobutamine stress did not detect any change in myocardial P_i or pH. ¹⁷⁵ Any ³¹P‐MRS measurements of the human heart during exercise in a 7 Tesla MR system have yet to be reported.

Physical Exercise and the Brain

With very limited intracellular stores of energy substrates, the brain is reliant on an adequate and continuous supply of nutrients and oxygen. An adaptive cerebrovascular system that can cope with fluctuations in supply and demand during physiological challenges is therefore crucial to maintain healthy brain function. To assess cerebrovascular health with MR, typically hypercapnic (inhalation of 5% CO₂) ¹⁷⁷ or pharmacological (intravenous infusion of acetazolamide) ¹⁷⁸ stimuli are used to elicit a maximal vasodilatory response such that the cerebrovascular reserve capacity relative to baseline conditions can be estimated. Such laboratory‐controlled stimuli do not reflect daily life conditions. Instead, physical exercise is a physiologically relevant challenge ¹⁷⁹ that has been applied in MR investigations of the human brain, albeit scarcely.

Until recently, imaging studies of the human brain during exercise ¹⁸⁰ were limited to utilizing the accessibility of the middle cerebral artery (MCA) for ultrasound imaging through the temporal acoustic window. ¹⁸¹ By measuring the blood flow velocity with transcranial Doppler ultrasound, the cerebral blood flow (CBF) can be estimated under the assumption of a constant vessel diameter. Using a 7 Tesla MR system to obtain black‐blood images at a high in‐plane spatial resolution of 0.2 × 0.2 mm², Verbree et al showed that the cross‐sectional area of the MCA lumen decreased by 2.1 ± 0.8% during rhythmic hand grip exercise. ¹⁸² Although exercise intensity was very low and induced only a very mild 11.2 ± 1.7% increase in heart rate to 68.4 ± 12.5 beats/min, the high spatial resolution at 7 Tesla provided sufficient sensitivity to detect the minute vasoconstriction of the MCA due to exercise‐induced sympathetic activation. Those data suggest that assuming a constant vessel diameter may lead to an overestimation of the increase in CBF with transcranial Doppler ultrasound during low‐intensity exercise. Using several different imaging planes for 2D phase‐contrast measurements at 3 Tesla, Tarumi et al made a detailed analysis of the interplay between CBF and cerebrospinal fluid (CSF) flow during low‐intensity rhythmic hand grip exercise, showing how both CBF and CSF flow are coupled to maintain intracranial volume‐pressure homeostasis in healthy volunteers. ¹⁸³

Estimating CBF from the blood flow in a single large cerebral vessel is a rather coarse method for assessing whole‐brain behavior, because any differences between brain regions cannot be resolved. ASL‐MRI ¹⁸⁴ provides quantitative maps of tissue perfusion using magnetically labeled inflowing blood as an endogenous “tracer”. ⁶⁸ Based on the subtraction of control‐ and label‐images, ASL‐MRI is deemed to be susceptible to motion‐induced errors. Nonetheless, Mast et al demonstrated that it is feasible to collect quantitative data on cerebral perfusion of the whole brain as well as its different regions using pseudo‐continuous ASL (pCASL) MRI during bicycling exercise in a 3 Tesla MR system. ¹⁸⁵ At a spatial resolution of 2.75 × 2.75 × 5 mm³ and a temporal resolution of 9.1 s per control/label pair, pCASL datasets were acquired at rest, during 5 minutes of steady‐state exercise at moderate intensity with heart rates of ~100 beats/min and at vigorous intensity (~140 beats/min), and during initial recovery. Whole‐brain CBF remained similar during moderate exercise, decreased by >10% during vigorous exercise, and further decreased during initial recovery by 15% compared with resting‐state CBF (Fig. 13). Strikingly different from whole‐brain behavior, the motor cortex CBF increased by >12% during moderate exercise, returned to resting‐state values during vigorous exercise, and decreased by >17% during recovery. This downward parabolic curve of motor cortex CBF relative to exercise intensity agrees with studies of cerebral perfusion during exercise ¹⁸⁰ that used ultrasound measurements of the MCA. Indeed, the motor cortex is fed predominantly by the MCA, and its perfusion may differ from other brain regions during physical exercise as was shown with positron emission tomography (PET) ¹⁸⁶ and now also with MRI. ¹⁸⁵ In the latter study, whole‐brain CBF measurements with pCASL‐MRI were benchmarked against estimates based on 2D phase‐contrast measurements of blood flow in the common carotid arteries. Those large‐vessel surrogate measures overestimated whole‐brain CBF by >30 mL/100 g/min, which proportionally increased for higher CBF values. This became most apparent during vigorous exercise, with a large proportion of the blood flow in the carotid arteries directed through the external carotid artery toward the skin for thermoregulation. Indeed, elevated skin perfusion became apparent in pCASL perfusion maps as enhanced perfusion‐weighted signal in the scalp (Fig. 13). Interestingly, that study reported a lower cerebral metabolic rate of oxygen after vigorous‐intensity exercise, which suggests that brain oxygen demand during recovery was lower than during resting‐state conditions. ¹⁸⁵ It is intriguing to consider that such reduced brain oxygenation may be a mediator for the sensation of fatigue acting as a limiting factor for exercise capacity, but data to support this notion are currently lacking.

FIGURE 13.

Quantitative maps of cerebral blood flow (CBF) obtained with pseudo‐continuous arterial spin labeling (pCASL) MRI at 3 Tesla at rest, during various stages of steady‐state supine bicycling exercise, and during subsequent recovery (a). Regions of interest for hippocampus and motor cortex are outlined in red and blue, respectively, on T₁‐weighted anatomical images with the MNI‐space (Montreal Neurological Institute, Montreal, Quebec, Canada) z‐coordinate indicated. Note the enhanced perfusion of the skin during exercise and subsequent recovery. Time curves (temporal resolution, 9.1 s per control/label pair) of the mean whole‐brain (WB) CBF (b) show a consistent decrease during vigorous exercise that continued during recovery. Reproduced from Mast et al ¹⁸⁵ with permission from Elsevier under a Creative Commons BY‐NC‐ND license (https://creativecommons.org/licenses/by‐nc‐nd/4.0/).

Several studies have used MRI to investigate the brain immediately after a single bout of exercise, i.e., during recovery. A small pioneering study in young healthy adults (24.8 ± 1.5 years; n = 5) showed that whole‐brain CBF measured with pulsed ASL at 3 Tesla was approximately 20% higher 10 minutes into recovery after a 30‐minute bout of moderate‐intensity (125 beats/min) upright bicycling exercise. ¹⁸⁷ Given the notion that habitual physical activity may improve hippocampal structure and function, ¹⁸⁸ regional CBF after exercise has been investigated for the hippocampus in healthy adults, showing either a ~20% ¹⁸⁹ up to 30% ¹⁸⁵ reduction or a >10% ¹⁹⁰ elevation of perfusion. Another study in older subjects (70.7 ± 3.1 years; n = 25) showed that lower hippocampal CBF at 7 minutes after exercise was associated with higher cognitive performance. ¹⁹¹ In stroke survivors, pCASL MRI after 20 minutes of low‐intensity (100 beats/min) or moderate‐intensity (110 beats/min) semi‐recumbent bicycling exercise revealed region‐specific and exercise intensity‐dependent changes in CBF, with low perfusion in ischemic lesions that did not change after exercise. ¹⁹² Others have used BOLD‐MRI to map cardiac‐related brain pulsatility in response to exercise, ¹⁹³ which could serve as a measure of arterial stiffness when evaluating cerebrovascular health. ¹⁹⁴ More research on such acute effects of exercise on the brain is warranted in order to better understand the mechanisms of how physical exercise can be beneficial for cerebrovascular health and cognitive performance, ¹⁹⁵ eg, in aging ¹³ or in stroke rehabilitation. ¹⁹⁶

Outlook

Exercise MR stress testing provides a paradigm for noninvasive assessments of human (patho)physiology. This methodology is noninvasive and nonionizing, and therefore very suitable for (repeated) measurements in healthy volunteers as well as patients. Yet, despite ample demonstration of its feasibility and merits in various scenarios and patient populations, application of exercise MR has predominantly been restricted to studies of relatively small cohorts in a research setting. Broadly accepted protocols for use in a clinical workflow are currently lacking. Practical barriers remain the technological complexities associated with obtaining meaningful quantitative MR readouts during exercise, which require local expertise, dedicated hardware such as suitable RF coils and MR‐compatible ergometers, and typically dedicated software for data processing and analysis. Performing exercise within the confined space of an MR scanner bore while fitted with vital sign sensors, RF coils and an ergometer may be daunting for subjects. On the other hand, the subject remains in control of the self‐applied stress intensity, and may halt exercise instantaneously, with infusion of pharmacological stress agents and catheterizations for pressure measurements no longer required.

With continued technological advances, we foresee improvements of various aspects of exercise MR stress testing. Vendors have put substantial effort into providing authority‐approved equipment and methods for non‐proton (eg, phosphorus‐31) MR evaluations. However, the frequent use of custom‐built ergometer equipment for exercise MR stress testing, particularly for musculoskeletal applications, hamper authority approval for their clinical use. Strong collaboration and coordination between MR ergometer users and vendors is needed to implement authority‐approved devices that are validated for multiple body parts and for broad‐scale clinical applications. With so‐called “interleaved” scanning of several different acquisitions, eg, ³¹P‐MRS and BOLD ¹H‐MRI, ¹⁹⁷ multi‐parametric MR readouts can be obtained essentially simultaneously. This concept is ideal for exercise stress testing, which is typically limited to a single session that is very difficult to replicate precisely, and paves the way for a more integrative approach to study muscle oxygen utilization and ATP turnover. ⁸¹ Such integrative approaches are particularly important for evaluating exercise intolerance and fatigue, where the underlying mechanisms could be misinterpreted when based solely on individual MR readouts. Alternatively, other noninvasive techniques to investigate skeletal muscle physiology such as surface electromyography ¹⁹⁸ and NIRS ¹⁹⁹ may be integrated into exercise MR stress testing examinations, amplifying its clinical utility.

Advancements in machine learning will contribute to pushing the limits of temporal and spatial resolution for MRI during exercise, ²⁰⁰ such that dynamic physiological processes can be captured and analyzed in even more detail. For instance, evaluations of ³¹P‐MRS data in skeletal muscle have predominantly focused on quantifying the PCr recovery rate from series of spectra obtained during recovery after exercise, mostly for practical reasons and because the PCr resonance frequency is robust to any pH changes. ³¹ In contrast, extracting phenotypical information from series of spectra acquired during exercise is much less straightforward, particularly if the exercise mode or intensity recruits white myofibers along with red and intermediate myofibers (Fig. 3). Analyses of such series are labor intensive and sensitive to user input, and therefore represent a bottleneck in the processing workflow, which precludes swift automated evaluation, and thus stalling routine clinical use of skeletal muscle exercise ³¹P‐MRS stress testing. The application of neural networks that capture the known mechanistic biochemical and physiological correlations between state variables in time may render the processing and quantitative analysis of such datasets amenable for automation. Moreover, machine learning can help in overcoming the labor‐intensive burden of image processing and segmentation ²⁰¹ for quantification of cardiac function during exercise MR stress testing, improving its applicability in the clinical workflow. Indeed, commercial software packages for cardiac function analysis already include machine learning algorithms to generate ventricular contours automatically. While trained to segment conventional cine MRI series of the heart, neural networks provided automated segmentations the LV and RV volumes in real‐time MRI series at rest with an accuracy that is similar to the inter‐observer variability for manual segmentation. ²⁰² Although promising, the performance of those deep learning methods is not yet sufficient for a fully automated analysis of real‐time MRI series acquired during exercise. ²⁰² Using data from a free‐breathing multi‐parametric mapping sequence, a neural network can be used to substantially reduce the time‐consuming image reconstruction process, by reconstructing myocardial T₁ and T₂ maps nearly instantaneously. ²⁰³ Its applicability to exercise MR stress testing, eg, for probing myocardial perfusion with T₁ mapping ¹⁵⁶ or blood oxygenation with T₂ mapping, ²⁰⁴ still needs to be demonstrated. Advanced machine learning approaches increasingly find their way in many aspects of MR examination ²⁰¹ such as data acquisition, image reconstruction, and quantitative analysis. We foresee that exercise MR, which typically requires fast image acquisitions, generates large datasets, and potentially suffers from low signal‐to‐noise ratios and motion artifacts, will particularly benefit from the maturation of machine learning.

Driven by advancements in data acquisition and reconstruction techniques, applications of low magnetic field (i.e., ~0.5 Tesla) MR systems have regained interest. ²⁰⁵ For instance, MRI at low magnetic field allows imaging of the lungs due to its superior magnetic field homogeneity that reduces susceptibility artifacts near air‐tissue interfaces. Indeed, Seeman et al used supine pedaling exercise in clinical MR system ramped down to 0.55 Tesla to demonstrate dynamic accumulation of lung water after vigorous‐intensity exercise and its subsequent clearance in healthy volunteers (n = 15) and patients with heart failure (n = 2). ²⁰⁶ Such measurements of lung water could aid in examinations of (exercise‐induced) transient pulmonary congestion in HFpEF. ²⁰⁷ Additionally, the low magnetic field reduces motion‐induced distortions of the ECG signal, may alleviate some of the constraints on the design and construction of ergometers and auxiliary equipment such as blood pressure monitors, and more readily affords the design of novel MR systems with a vertical open‐bore configuration. ²⁰⁵ Open‐bore MR systems offer more space to perform physical exercise, and allow for evaluations of the musculoskeletal system under weight‐bearing conditions. Moreover, such setups may open up opportunities to revive past approaches for upright exercise MR stress testing ¹⁸ for a more integrated, systemic investigation of exercise physiology. ¹⁹ , ²⁰⁸

The soft tissue contrast of MRI provides noninvasive access to essentially all organs of the human body. As such, it allows for an integrated evaluation of peripheral and central function beyond an individual organ. Indeed, a recent study used ³¹P‐MRS of PCr recovery kinetics to link better mitochondrial function in skeletal muscle with a reduced risk of mild cognitive decline and dementia, suggesting that mitochondrial dysfunction may play an important role in Alzheimer's disease. ²⁰⁹ Whereas early in vivo ³¹P‐MRS measurements during mechanical tasks that involved single limb muscles revealed profound intramuscular acidification, ⁷⁵ studies of skeletal muscle energy homeostasis and pH balance during exhaustive two‐legged bicycling exercise showed a dampened intramuscular pH drop. ²⁵ , ²¹⁰ This phenomenon was attributed to the superior cardiovascular and respiratory support during exercise that recruits a large muscle mass relative to single‐limb exercise. ²¹⁰ Such measurements could be extended to investigate the systemic effects of whole‐body exercise on other organs, such as the brain. ²¹¹ Moreover, the impact of any disease that affects cardiovascular or pulmonary function on peripheral organs that rely heavily on adequate substrate supply and removal of metabolic waste products can be quantitatively assessed with exercise MR stress testing, ²⁵ potentially serving to inform therapeutic strategies.

Despite many advancements in the past decades, exercise MR arguably still represents a niche area in the tremendously versatile field of MR. It is up to our community of MR researchers, clinicians and physiologists to keep moving forward by providing innovative solutions and meaningful examples. At the same time, we advocate for continued support from vendors of MR systems, processing and analysis software packages, and ergometers by ensuring compatibility of their tools and equipment with exercise MR stress testing. Such joint efforts are warranted to consolidate exercise MR stress testing as a valuable contributor to our understanding of human exercise physiology and exercise intolerance.