Establishing an evidence-based modern breast MRI program

Abstract

Breast MRI has evolved over the past several decades into a cornerstone of breast imaging. Historically, dynamic contrast-enhanced (DCE) MRI has served as the foundation of breast MRI protocols for differentiation of benign and malignant lesions, supplemented by additional sequences to refine diagnostic accuracy. More recently, advanced techniques such as diffusion-weighted MRI, ultrafast DCE-MRI, and deep learning models have further expanded capabilities of breast MRI. These innovations, however, have also contributed to substantial variability in breast MRI protocols across institutions. At the same time, the expanding indications for screening and diagnostic breast MRI are driving higher patient volumes, creating operational challenges for breast imaging centers tasked with balancing efficiency, accuracy, and limited resources. This review outlines the key elements and considerations of modern breast imaging protocols, discusses strategies for protocol optimization, and explores emerging technologies and future trends that are shaping the next generation of breast imaging.

Introduction

Breast MRI has become an essential modality for routine breast imaging after nearly half a century of development. Early breast MRI had limited clinical value in cancer detection due to an inability to visualize breast cancers using intrinsic MR signal features alone^1,2. However, non-contrast MRI techniques did show value for early detection of silicone implant rupture, leading to subsequent FDA recommendation of MRI screening for silent implant rupture 5–6 years after implant placement. In the late 1980s, the introduction of gadolinium-based contrast agents resulted in clear breast lesion visibility and thus renewed clinical interest in breast MRI for cancer screening and detection^3,4. Since then, breast MRI has primarily relied on contrast enhancement to detect malignancy, exploiting increased vessel permeability resulting from tumor angiogenesis.

Over the past 40 years, breast MRI has greatly progressed with improved breast coil technology and strengthened temporal and spatial resolution^5,6. Early on, two approaches to breast MRI emerged: one emphasizing high spatial resolution and morphologic assessment, and another focusing on temporal resolution through dynamic imaging to evaluate enhancement kinetics^3,6,7. Continued advancements have shown that both approaches provide complementary value to modern breast MRI. However, and as described in the BI-RADS manual⁸ morphological assessment (e.g., lesion type, shape, margin, enhancement pattern, and distribution) has become the single most critical feature for accurate breast MRI assessments.

Clinical trials since the 1990s have confirmed the value of breast MRI for both cancer screening and diagnosis^9–16, and modern-day breast MRI programs seek to optimize sensitivity and specificity while maximizing breast MRI access and efficiency. Breast MRI has demonstrated higher sensitivity for breast cancer detection compared to mammography and/or ultrasound⁹, and adjunct breast MRI for screening has been shown to improve early detection and overall survival compared to mammographic screening alone^17,18. This evidence that has led to consensus recommendations for supplemental MRI screening in high-risk women (with greater than 20% lifetime risk of breast cancer)^19,20. Common, evidence-based breast MRI performed for diagnostic purposes includes evaluating extent of disease of known cancer, evaluation of therapeutic response to neoadjuvant treatment, assessment for mammographically and clinically occult malignancy in the setting of axillary nodal presentation of breast cancer, and limited problem-solving applications, such as women presenting with suspicious nipple discharge after a negative mammographic and sonographic evaluation.

As MRI volumes have increased dramatically over the past several decades^21–23 with over 40 million MRI examinations performed annually in the US alone, modern breast MRI protocols must be optimized for efficiency without compromising performance. In this review, we examine the evolving role of breast MRI, highlighting the challenges and limitations of current protocols. We present modern strategies aimed at meeting clinical demands for efficient image acquisition and accurate interpretation. Finally, we explore emerging techniques such as diffusion-weighted imaging, ultrafast imaging, and deep learning that are shaping the future of the field.

Conventional breast MRI protocols

American College of Radiology requirements for breast MRI

An essential aspect of breast MRI protocol optimization is compliance with American College of Radiology (ACR) Breast MRI Program standards. These standards are meant to be minimum requirements and do not preclude performance of additional series that may be valuable for interpretation, such as non-fat suppressed T₁-weighted MRI and diffusion-weighted. The ACR Breast MRI Program standards require a T₂-weighted or bright fluid series (typically fat-suppressed), pre-contrast T₁-weighted series, and multiphase post-contrast T₁-weighted series including early and delayed phases. Furthermore, in-plane resolution should be less than 1 mm so that morphological features can be described accurately in keeping with the ACR Breast Imaging Reporting and Data System (BI-RADS) atlas⁸. Lesion morphology is determined using early-phase post-contrast images, where high spatial resolution is essential to properly demonstrate the margins while avoiding confounding signal washout or progressive enhancement associated with later time phases. Morphological features serve as the primary factors in BI-RADS to help discern level of suspicion, and each lesion is categorized as a non-mass enhancement (NME) or mass with sub-modifiers further describing morphology. Of note, focus is currently defined as a morphological descriptor in the BI-RADS lexicon; however, recent communications indicate it will be removed in the next edition of the BI-RADS manual²⁴. High-spatial resolution acquisitions providing less than 1 mm pixel sizes often result in longer scan times, which can limit the temporal resolution of post-contrast imaging, especially in scanners less equipped with advanced acceleration capabilities (such as partial k-space sampling and parallel imaging).

Current challenges and limitations of conventional protocols

Access to breast MRI is increasingly challenged by growing demand and limited scanner availability. Breast MRI utilization increased nearly 20-fold in the 2000s²⁵, driven by guideline endorsements, expanding clinical indications, technological advances, and broader insurance coverage, with volumes later experiencing a transient decline during the COVID-19 pandemic²⁶. In the post-pandemic medical landscape, there has been a strong recovery of breast MRI^27–29, with demand expected to continue to grow in the face of scanner and workforce shortages. Although there is emerging evidence that supplemental screening can improve outcomes for some patients (e.g., patients with dense breasts and/or patients with a personal history of treated breast cancer) who do not meet current criteria for breast MRI^11,14,20, high cost and relatively long exam times have, in part, slowed wider breast MRI implementation. For example, the cost of a breast MRI ranges from $545–$2,439 per exam³⁰, which can leave patients with substantial out-of-pocket cost depending on insurance coverage. MRI exam times can be lengthy ranging from 30–50 minutes when including setup, IV placement for contrast, and positioning, with only approximately 20 minutes devoted to actual image acquisition³¹. Prolonged prone positioning in a breast coil can add to patient discomfort, anxiety, and claustrophobia. Therefore, efforts to streamline acquisition time must balance goals to improve efficiency, cost, and patient comfort while preserving the image quality and diagnostic performance that make breast MRI clinically valuable, as overly shortened scans can reduce image quality and limit accurate exam evaluation. Additionally, while breast MRI offers extremely high sensitivity for lesion detection and diagnosis, its moderate specificity can lead to additional pressures on access and growth. Positive MRI examinations lead to additional MRI-guided biopsies that compete with diagnostic MRI slots. Furthermore, high false positive rates can erode referring clinician and patient confidence, undermining MRI use. New strategies are necessary to streamline acquisitions and reduce false positives prompted by MRI examinations to make breast MRI a more efficient and practical tool for routine breast cancer screening.

Building a modern breast MRI protocol

Equipment

Foundational to a modern breast MRI protocol is optimal equipment and hardware, including high field-strength magnets and dedicated multi-channel breast coils. Modern breast MRI is performed in a 1.5 T or 3 T magnet with the patient in the prone position. A field strength of 3 T offers increased signal-to-noise (SNR) ratio compared to 1.5 T, allowing higher spatial resolution, as well as improving lesion conspicuity and detection of smaller lesions^32–36. Dedicated breast radiofrequency (RF) coils should contain at least 4 coil elements (or channels)³⁷, with modern day coils containing 16 coil elements, enabling increased SNR and higher parallel imaging factors, reducing scan times³⁸. While increasing beyond 16-channel coils can reduce field homogeneity and introduce aliasing artifacts in image reconstructions, ongoing research aims to advance coil technology to enable faster image acquisition without compromising image quality³⁹.

MRI sequences

Modern breast MRI protocols begin with various calibration scans and a low-resolution scout image to define the field of view, followed by T₂-weighted, T₁-weighted, and dynamic contrast-enhanced (DCE) imaging, with many modern protocols also incorporating diffusion-weighted imaging (Figure 1). T₂-weighted sequences are often fat suppressed to improve visualization of fluid-containing structures and edema, essential for delineating benign and malignant findings. The reference for T₂ hyperintensity in breast lesions remains debated, with some studies comparing signal to adjacent breast tissue and others to lymph nodes; the upcoming BI-RADS 6th edition is anticipated to specify normally appearing lymph nodes as the reference^24,40–43. T₂ hyperintense findings are typically benign, including cysts, lymph nodes, and fibroadenomas (Figure 2). T₂ hyperintense malignancies, including mucinous and papillary carcinomas, necrotic masses, and metastatic lymph nodes, are less common⁴³. T₁-weighted images without fat suppression aid in problem-solving masses, as intrinsic high signal on T₁-weighted images typically indicates benign entities such as breast lipoma, fat necrosis, and hamartoma. Additionally, the non-fat suppressed T₁-weighted series can be used to identify biopsy clips and post-surgical changes through identification of susceptibility artifacts manifested as localized signal voids (Figure 3).

Figure 1.

Conventional vs. Modern Breast MRI Protocols. Conventional breast MRI protocols (panels A and B) have acquisition times ranging from 25–30 minutes. Protocols acquiring a slice thickness close to in-plane spatial resolution (i.e., near isotropic voxels, panel B) allow for multiplanar reformats and obviate the need for additional delayed phase sagittal acquisitions. Modern breast MRI protocols (panels C-E) leverage new compressed sense and AI reconstruction techniques, enabling faster image acquisition and shorter exam times. Abridged breast MRI protocols such as abbreviated MRI⁵⁴ (AB-MRI, panel C) and rapid abridged multiphase (RAMP⁴⁹, panel D), emphasize the post-contrast appearance, with the RAMP protocol also meeting ACR requirements. Finally, panel E highlights an efficient multiparametric MRI protocol that provides additional information (ultrafast DCE-MRI [UF], diffusion-weighted imaging [DWI]) for improved breast MRI specificity.

Figure 2.

Example of scattered cysts in a patient with known cancer presenting for evaluation of extent of disease. Note that the cysts are evident as high signal lesions (arrows) on T2-weighted series with fat suppression (A) but are not well visualized on post-contrast T1-weighted series with fat suppression (B).

Figure 3.

Example of susceptibility artifact from a metallic biopsy marker clip in a 67-year-old woman with biopsy-proven invasive ductal carcinoma in the upper inner quadrant of the right breast (yellow arrow). Note that susceptibility indicating location of the clip (red arrow) at the medial edge of the tumor is much easier to identify on the non-fat suppressed T1-weighted images (A) than on the T1-weighted post-contrast series with fat suppression (B). This is due to both technical factors on fat-suppressed T1-weighted MRI (longer TE) and that susceptibility may be difficult to distinguish from adjacent “dark” fat that is suppressed on the post-contrast images.

Breast DCE-MRI involves the acquisition of typically fat-suppressed T₁-weighted images before and after intravenous administration of a gadolinium-based contrast agent. ACR protocol accreditation requires an early post-contrast sequence obtained within 4 minutes and a delayed post-contrast sequence obtained after 4 minutes. ACR BI-RADS guidelines, however, recommend the initial phase occurs within 2 minutes post contrast injection, as most cancers demonstrate peak enhancement within this window, with k-space ideally centered close to 90 seconds after gadolinium administration. Delayed phase acquisitions enable assessment of contrast washout, a strong independent predictor of malignancy^44–46. Additionally, delayed phase acquisitions provide improved detection of residual in-situ disease post neoadjuvant therapy compared to early-phase post-contrast images^47,48. In addition to adequate temporal resolution, breast DCE-MRI requires high spatial resolution to ensure reliable assessment of morphological features. While ACR criteria allows for acquiring high in-plane resolution (≤1 mm) with thicker slices (≤3mm), which can be achieved in reasonable scan times on most systems, this approach has limitations for evaluating orthogonal views. Some sites may acquire additional delayed series in alternate planes and expanded fields of view to provide more dedicated coverage of the axilla and surrounding regions; however, increased scan time and associated costs must be balanced against the potential clinical benefit. Designing protocols with slice thickness to closely match in-plane resolution (i.e., near isotropic resolution) enables generation of high quality multiplanar reformats, obviating the need for separate sagittal acquisitions and ensuring that morphological assessments across views are performed within the same post-contrast phase (Figure 1). It should be noted that isotropic or near isotropic resolution may be achieved from anisotropic data by various interpolation methods where data are re-sampled to create synthetic slices between the native slice resolution. This approach can allow for thinner slices without a significant time penalty, facilitating high-quality multiplanar reformats.

Traditionally, DCE series are scanned as far out as 9–12 minutes after gadolinium administration to provide the most accurate delayed phase curve assessment. However, it has been shown that delayed phase worst-curve type (i.e., presence of any amount of washout is most concerning; presence of any amount of plateau and no washout is more concerning than presence of entirely persistent delayed phase) is the most critical kinetic feature for problem-solving lesions that are otherwise equivocal based on morphological feature alone⁴⁶. As such, studies have suggested that delayed phase acquisitions at 3–5 minutes post contrast acquisition may be sufficient for discrimination of benign and malignant lesions using only worst-curve features^49,50. This suggests an opportunity to further reduce scan times while still obtaining valuable kinetic features. Additionally, standard DCE-MRI involves administration of a gadolinium-based contrast agent at a dose of 0.1 mmol/kg. However, new high-relaxivity contrast agents such as gadopiclenol achieve comparable lesion visibility at lower doses, mitigating potential risks and/or patient concerns related to gadolinium deposition^51–53. Adopting streamlined protocols with shorter acquisitions and reduced contrast dosing has the potential to expand the accessibility of breast MRI to wider population

Menstrual cycle timing

BPE has been shown to correlate with the amount of circulating hormones^55,56. Thus, BPE can vary based on menstrual cycle timing, menopausal status, and tamoxifen use^56–60. The clinical impact of BPE remains highly debated: early practice recommended MRI to be performed between the 5^th and 12^th day of the menstrual cycle, when hormonally-influenced BPE is thought to be lowest, to avoid obscuration of lesions⁶¹. However, subsequent quantitative studies demonstrated higher BPE levels occurring later in patient’s cycles^60,62,63. Several studies have proposed that higher BPE reduces the diagnostic performance of MRI^64,65. Recently, a large systematic review of 8 studies by Bechyna and Balzter indicated the degree of BPE influences the diagnostic performance of breast MRI, with minimal or mild BPE associated with higher sensitivity and specificity, and moderate or marked BPE being associated with lower diagnostic performance⁶⁶. However, while these studies suggested significant reduction in diagnostic performance with higher BPE, the effect of menstrual cycle timing on BPE could not be determined. Conversely, there have been studies reporting the measurable impact of moderate to marked BPE on diagnostic performance is less than previously described, with no significant impact on final outcome of positive biopsy rate, cancer yield, sensitivity, and specificity^67,68. Recent studies have shown there is no significant benefit to scheduling MRI examinations based on menstrual cycle timing, which can yield challenges to patients and facilities^55,68. Given BPE has not been proven to be consistently lowest within any specific time-frame of the menstrual cycle, and BPE levels have not been proven to have a significant diagnostic impact, MRI can be performed at any point within the menstrual cycle without clear adverse effect on performance, thereby facilitating improved access for women and facilities. Similarly, MRI can be performed in lactating women, in whom BPE is often increased; screening MRI should proceed as indicated for high-risk patients, without deferral on the basis of nursing status. Finally, the use of ultrafast series, discussed in more detail below, likely will provide enhanced sensitivity in patients with higher BPE, mitigating any potential performance challenges due to elevated BPE⁶⁹.

Emerging applications and techniques of breast MRI

Additional clinical indications and patient populations

With more efficient and cost-effective MRI techniques, there is an opportunity to expand breast MRI eligibility beyond diagnostic and high-risk screening indications. Although women with a personal history of breast cancer are at elevated risk of a second future breast cancer, there is currently a lack of consensus recommendation for supplemental MRI surveillance for this population⁷⁰. Evidence for MRI surveillance is mixed, with some studies demonstrating no difference from mammography in cancer detection⁷¹ and others demonstrating increased sensitivity and detection rates for MRI over mammography^72,73. Optimal timing for MRI surveillance continues to be debated and remains an area of active research^20,74,75.

Supplemental MRI screening has been shown to potentially also benefit women with dense breasts, with increased rates of cancer detection and reduced interval cancers compared to those screened with mammography alone¹¹. Approximately 36% of women over 40 have heterogeneously dense breasts, and 7% have extremely dense breasts⁷⁶. Women with dense breasts on mammography have increased false-positive findings, reduced cancer detection rates, and more interval cancers than those with non-dense breasts^77–82. In the MRI sub-study of the ACRIN 6666 trial, investigating supplemental modalities for screening of women with dense breasts, an incremental cancer detection rate (ICDR) of 14.7 cancers per 1000 women was observed, providing increased detection of early breast cancer⁹. The DENSE trial (Dense Tissue and Early Breast Neoplasm Screening) trial also found similar improvement of cancer detection with ICDR of 13.4 cancers per 1000 screening examinations¹¹. Kuhl et al. yielded similar results for average risk women (of varying breast densities) with overall supplemental cancer detection rate of 15.5 per 1000 cases, detecting 60 additional cancers by MRI compared to conventional imaging in near 2,000 women⁸³. Limitations of breast MRI for supplemental screening include its intermediate specificity, leading to false-positives workups and cost. Further protocol optimizations allowing increased specificity with scan time reduction and associated diminished costs could assist in greater clinical adoption.

Abbreviated breast MRI

A full breast MRI exam may require 30–50 minutes in the MRI suite, with approximately 20 minutes dedicated to actual scan time³¹. Reducing scan times would enable shorter exams, which not only would improve patient comfort, but also lower reading times and potentially costs, thereby allowing increased throughput and supporting broader use of MRI as a screening modality. First popularized by Kuhl et al., abbreviated breast MRI (AB-MRI, Figure 1C) uses a shortened MRI protocol, usually acquiring a single post-contrast sequence and generally targeting an overall scan time of less than 10 minutes^54,84. While removing delayed-phase post-contrast sequences limits evaluation of washout kinetics, several studies have demonstrated comparable diagnostic accuracy^85–87, again emphasizing the higher importance of morphology characterization of lesions over kinetic assessment. Compared to digital breast tomosynthesis (DBT), AB-MRI demonstrates higher rates of cancer detection, particularly for invasive cancers, in women with dense breasts^14,88 and those with a personal history of breast cancer⁸⁹. Although concerns about the intermediate specificity of breast MRI persist, incorporating short sequences such as ultrafast DCE-MRI and diffusion-weighted imaging, may reduce false positives and unnecessary biopsies while maintaining exam efficiency⁹⁰. AB-MRI may be more suitable for excluding disease in screening patient populations whereas full protocol exams are often utilized for determining disease extent.

Ultrafast DCE-MRI

Ultrafast DCE-MRI is an emerging MRI technique that prioritizes temporal resolution over spatial resolution, offering unique insights into contrast uptake kinetics⁹¹. Ultrafast MRI involves acquisition of whole volume images every 3–8 seconds for the first 1–2 minutes after contrast delivery^90,92–94, and has been performed using various acceleration techniques, including parallel imaging, view-sharing (e.g., keyhole imaging), and compressed sensing. View-sharing techniques under sample the periphery of k-space to achieve the desired temporal resolution while maintaining contrast resolution and acceptable spatial resolution (e.g., in-plane resolution of 1–1.5 mm, slice thickness of 2.5–3 mm). With increased temporal resolution, kinetic features such as the maximum slope (MS) of the uptake curve and lesion time to enhancement (TTE, measured relative to the aorta TTE) can be derived to assess lesion vascularity and perfusion. These features have demonstrated value in lesion diagnosis without significantly increasing scan time by taking advantage of frequently idle scanner time in the first minute after contrast injection^90,92–96, as malignancies exhibit elevated vascularity and demonstrate earlier contrast uptake (shorter TTE, higher MS) compared to benign lesions^90,92–95. Although ultrafast DCE-MRI remains investigational, studies have demonstrated that ultrafast kinetic features improve specificity, with a range of 78–82% achieved in some studies, versus a full protocol MRI specificity of 52–76%, while maintaining sensitivity^91,95,97. Although no consensus exists on the optimal duration for ultrafast DCE-MRI, Cao et al. reported a scan duration of 67.5 s provides the best balance between imaging efficiency and diagnostic accuracy⁹⁸. Ultrafast DCE-MRI has also shown improved lesion conspicuity due to reduced BPE at earlier post-contrast phases (Figure 4)^69,99, value in evaluation of therapeutic response^100,101, and successful integration into of AB-MRI protocols, improving their diagnostic accuracy^102–105.

Figure 4.

A 50-year-old female presents for diagnostic MRI to evaluate extent of DCIS in the left lower outer breast. There is no identifiable disease on the Maximal Intensity Projection (MIP) due to marked background parenchymal enhancement (A). The mass and nonmass enhancement (circle) is delineated on early post-contrast T₁ -based subtracted series in the left lower outer quadrant (B). Ultrafast series (pre-contrast and 5 post contrast series with 8 second temporal resolution) with subtraction clearly demonstrates the segmental non-mass enhancement representing the malignancy, with enhancement (arrows) beginning as early as 24 seconds after contrast injection (C).

Diffusion-weighted Imaging (DWI)

Diffusion-weighted imaging (DWI) is a relatively rapid technique (< 3 min) that uses diffusion sensitizing gradients to measure the random Brownian motion of water molecules within tissues. Quantitative measures of the apparent diffusion coefficient (ADC) derived from DWI provide insight into tumor microstructural characteristics¹⁰⁶. Numerous studies have demonstrated that incorporating DWI with an ADC cutoff can improve the specificity of breast MRI^107,108. A study by Youn et al.¹⁰⁹ reported a 16% reduction in biopsies could have been achieved using point-of-care ADC measurements collected during patients’ clinical screening MRI and applying a pre-specified ADC threshold (1.53×10⁻³ mm²/s) to downgrade otherwise suspicious lesions. Furthermore, preliminary findings from the multicenter breast DWI screening trial (DWIST, South Korea¹¹⁰) suggest that DWI may serve as a non-contrast screening tool providing higher cancer detection rates than mammography and ultrasound in high risk women¹¹¹. Figure 5 highlights improved lesion conspicuity on DWI compared to DCE maximum intensity project due to high BPE. DWI also demonstrates value in early prediction of neoadjuvant therapy response¹¹². The multicenter ACRIN 6698 trial¹¹² demonstrated that mid-treatment increases in ADC are predictive of pathological response to neoadjuvant therapy, potentially supporting treatment personalization and reducing ineffective therapies. Future work may allow AI-supported integration of DWI along with other non-contrast images to mimic high spatial resolution contrast-based MRIs to facilitate non-contrast screening in some populations¹¹³.

Figure 5.

A 26-year-old woman presents for evaluation of triple-negative invasive ductal cancer in the right breast lower outer quadrant. There is no identifiable disease on MIP due to marked BPE (A). The biopsied malignant mass (circled) in the right lower outer quadrant demonstrates mild enhancement on early post-contrast subtraction images compared to the BPE (B). There is however clear signal from the high-b value DWI series (circled) (C) with corresponding low ADC value of 0.60 mm²/sec (circled) (D).

Despite its advantages, clinical integration of DWI has been limited in-part by variability in technical approach, including region-of-interest selection method for ADC measurement, as well as acquisition protocols resulting in varying image quality^114–116. These inconsistencies lead to reduced reliability of ADC measures, impacting clinical decision making; however, concerted efforts are being made across the field to achieve standardization¹¹⁷. The European Society of Breast Imaging (EUSOBI) recently published a breast DWI consensus, recommending breast DWI should be performed using at least two b-values of 0 and 800 s/mm² (maximum b of 800 s/mm² for ADC calculation), acquired in-plane resolution ≤ 2 mm with ≤ 4 mm slice thickness, and application of fat suppression for reproducible and accurate ADC quantitation¹¹⁷. Also recommended is a TR ≥ 3000 ms to ensure sufficient recovery of longitudinal magnetization and shortest TE, to minimize signal loss and reduce artifacts¹¹⁷. While single-shot echo planar imaging is the most commonly used DWI sequence in clinical practice, it is prone to geometric distortions due to its high sensitive to magnetic field inhomogeneities^118–120. Advanced DWI acquisition techniques, including multi-shot EPI and turbo spin echo, can reduce these distortions, albeit at the cost of increased scan duration. More recent technical advancements, including simultaneous multi-slice acquisition, restriction spectrum imaging, and deep learning image reconstruction, can reduce acquisition time while generally preserving image quality^119–121.

Artificial intelligence in breast imaging

In the past five years there has been substantial progress on the application of artificial intelligence (AI) to breast MRI for a variety of tasks¹²². Most prominently, several studies have demonstrated the utility of AI for cancer detection and diagnosis^123–130, often applying deep learning techniques to generate a final benign or malignant classification for an input image. Such AI tools could optimize radiology workflows, with recent studies exploring the potential of AI to triage negative exams versus those that require further radiologist evaluation^131–133. Additionally, recent work¹³⁰ by Oviedo et al. has further explored explainable AI techniques, to gain insight into the AI decision making process and indicate which image areas were “anomalous” that led to final malignant classification (Figure 6).

Figure 6.

Example output from an explainable AI model using breast maximum intensity projection images, demonstrating identification and visualization of anomalous regions to potentially aid interpretation. Figure adapted from Oviedo et al.¹³⁰

Despite the advancement of AI for breast MRI applications, it is still lagging when compared with other imaging modalities including mammography, digital breast tomosynthesis, and ultrasound¹²². This is due to the complexity of multiple 3D volumes within a breast MR exam and limited data availability for model training. The lack of large and diverse datasets leads to biased and non-generalized AI models, which pose challenges in deploying the models in real clinical settings^134–137. To tackle these challenges, recent studies have explored application of federated and transfer learning processes during AI model development. Federated learning deploys a unified AI model that can be trained on multi-institutional data without the need for that data to leave each institution’s firewall, while transfer learning deploys a pre-trained AI model that is then fine-tuned with a target dataset^138–140. These approaches show promise in overcoming data limitations and may accelerate the development of robust, generalizable AI tools for breast MRI.

Putting it All Together: How to perform a modern breast MRI

At our institution, we have elected to develop a shortened 3T breast MRI protocol that leverages advanced k-space acquisition (compressed sensing) and emphasizes high-value sequences, while removing unnecessary delayed DCE series (Figure 1E). In this protocol, we continue to obtain a traditional DCE series to meet ACR accreditation standards, but only obtain two post contrast examinations in alignment with data that suggest delayed phases beyond 4 minutes after contrast administration do not provide additional accuracy for worst-curve type assessments. We continue to obtain T₂ weighted series, as these series provide improved specificity, particularly for small circumscribed masses⁴³, assessment of lymph nodes, and for characterizing thin rim-enhancing structures likely to be inflamed cysts. We obtain our high-spatial resolution axial-based DCE series with near isotropic voxel sizes (0.5 AP × 0.55 RL × 0.75 SI mm, leveraging interpolation in the slice/SI dimension) so that we can generate high quality multiplanar reformats in sagittal and coronal orientations at the same post-contrast timing as the source images. This allows us to avoid acquiring additional sagittal or coronal acquisitions at later timepoints at a time penalty and when malignant lesions have likely already washed out. We do not adjust breast MRI scheduling based off menstrual cycle timing. We run a shortened ultrafast series during idle scan time (the first 40 sec after contrast injection) to assist with confidence in identifying unique lesions in the setting of higher BPE as well as a problem-solving tool to improve specificity⁹⁰ for lower suspicion lesions. We obtain a limited DWI series to assist with specificity, particularly improving confidence for benign appearing masses by applying a validated ADC threshold derived from multicenter data¹⁰⁹. Finally, we continue to obtain a T₁-weighted non-fat suppressed series to help assess for fat containing lesions and best determine location of biopsy-marker clips. By employing compressed SENSE and other acceleration techniques, and using a low to high cartesian k space sampling strategy for DCE (obtaining center k-space at the beginning of the scans), we are able to obtain a streamlined multiparametric MRI with only 15–17 minutes of scan time, not much longer than that of much simpler AB-MRI exams that often do not include DWI, T₂-weighted series, ultrafast, or conventional DCE series. Other streamlined protocols include the Rapid Abridged Multiphase (RAMP) breast MRI protocol⁴⁹ meeting ACR accreditation requirements while minimizing scan time. Such an approach focuses on even shorter scan times by not including the additional parameters of DWI and ultrafast series. In your practice, it is critical to work with physicists and vendor representatives to optimize breast MRI protocols to emphasize high yield series and decrease unnecessary low value series.

Conclusion

Breast MRI remains the most sensitive modality for breast cancer detection. However, widespread adoption is still limited by factors such as cost, exam duration, and accessibility when compared to mammography and ultrasound. The modern breast MRI protocol must emphasize the most clinically valuable sequences that can help address these challenges and maximize patient benefit. As techniques such as ultrafast DCE-MRI, DWI, and AI continue to demonstrate their advantages, breast imaging centers are poised to expand their use, ultimately improving diagnostic accuracy and expanding access for a broader population of women.

Key Messages:

1. Modern breast MRI protocols should prioritize efficiency without sacrificing diagnostic accuracy, leveraging high-value and omitting low-yield acquisitions to reduce scan time, cost, and patient discomfort while meeting ACR standards.

2. Emerging technologies, such as ultrafast DCE-MRI, diffusion-weighted imaging, and artificial intelligence, are improving MRI specificity, workflow efficiency, and lesion characterization, and have the potential to expand eligibility for screening beyond high-risk patients.

3. A multiparametric, evidence-based approach to breast MRI design, informed by technical advances, standardized protocols, and operational optimization, can increase accessibility and support broader screening applications.