Quantitative Prostate MRI, From the AJR Special Series on Quantitative Imaging

Abstract

Prostate MRI has traditionally relied on qualitative interpretation. However, quantitative components hold the potential to markedly improve performance. The ADC from DWI is probably the most widely recognized quantitative MRI biomarker and has shown strong discriminatory value for clinically significant prostate cancer (csPCa) as well as for recurrent cancer after treatment. Advanced diffusion techniques, including intravoxel incoherent motion, diffusion kurtosis, diffusion tensor imaging, and specific implementations such as restriction spectrum imaging, purport even better discrimination, but are more technically challenging. The inherent T1 and T2 of tissue also provide diagnostic value, with more advanced techniques deriving luminal water imaging and hybrid-multidimensional MRI. Dynamic contrast-enhanced imaging, primarily using a modified Tofts model, also shows independent discriminatory value. Finally, quantitative size and shape features can be combined with the aforementioned techniques and be further refined using radiomics, texture analysis, and artificial intelligence. Which technique will ultimately find widespread clinical use will depend on validation across a myriad of platforms use-cases.

The use of MRI for the detection, characterization, management, and follow-up of prostate cancer has undergone a sea of change over the past three decades. Initial protocols used conventional T2-weighted imaging (T2WI) for anatomy combined with spectroscopic imaging for functional characterization. The current standard of T2WI combined with DWI and the ADC map, as well as dynamic contrast-enhanced (DCE) perfusion imaging for functional evaluation, has made multiparametric MRI (mpMRI) (Fig. 1) a mainstream for all practitioners managing patients with a suspicion or diagnosis of prostate cancer.

Figure 1:

Example of conventional qualitative and quantitative components from prostate MRI in 67-year-old man with PSA of 4.16 ng/mL. A) Qualitative T2-weighted image shows 0.7-cm circumscribed homogeneously hypointense circumscribed noninvasive mass in right apical peripheral zone (arrows). B) ADC map shows hypointense mass, which is more hypointense compared to other areas (arrows). C) cClculated high b-value DWI shows hyperintensity of mass (arrows) compared to remainder of prostate. D) Restriction spectrum imaging (RSI) overlay corresponds to biomarker value in mass of 102 (arrow), in intermediate-high range for calibration of this statistical metric. E) Early enhancement T1-weighted image shows enhancement in mass that is contemporaneous (but not sooner) than transition zone (arrows). F) Conventional time-intensity curve shows early enhancement and plateau. G) Ktrans color map shows values slightly lower than transition zone (arrows). H) Kep map shows values similar to transition zone (arrows). I) Summary of quantitative lesion values for ADC, Ktrans, Kep, Ve, iAUGC, and ADC including median, mean, SD, minimum, and maximum. Listed kurtosis is statistical kurtosis of histogram map, not diffusion kurtosis. J) Histogram plot of ADC values; such histograms can highlight presence of two populations of values.

Nevertheless, current specifications for the performance, assessment, and reporting of prostate mpMRI include lesion size and lesion contact with the prostatic capsule as the only quantitative components. While ADC is a quantitative metric of diffusion, and quantitation is possible for both T2WI and DCE, variations in scanner performance, imaging protocol parameters, and technical capability, as well as reader familiarity with quantitative assessment, have precluded broad recommendations on their clinical use. However, quantitative assessment has consistently been proposed as a way of standardizing patient management across scanner platforms and site protocols, to minimize variability in the derived assessment metrics.

The prostate is not the only organ where quantitative aspects of MRI have been investigated for disease detection and characterization. The use of Hounsfield units to characterize renal stones and cysts, proton-density fat fraction to stratify fatty liver disease and stiffness, and liver stiffness from MR elastography to measure liver fibrosis, are examples of now established quantitative imaging metrics. Substantial data now support the consideration for incorporation of similar quantitative components in prostate mpMRI, with literature providing the groundwork to this end. This Review describes the current evidence supporting the incorporation of quantitative techniques into cancer detection and tissue characterization by prostate mpMRI and highlights the outstanding challenges that must be addressed before implementation. Table 1 summarizes the basics of each reviewed quantitative mpMRI method.

Table 1:

Quantitative Prostate MRI Components

Component	Conventional Diffusion	Advanced Diffusion	Hybrid Diffusion	Water Relaxation	Dynamic Contrast Enhancement
Examples	ADC	IVIM, DKI, RSI	VERDICT, HM-MRI	Tl, T2, MRF, LWI	Ktrans, washout
Biologic basis	Free water motion restriction	Cellularity, cellular disorder	Tissue composition, diffusion, and relaxation	Water content and tissue composition	Blood flow
Advantages	Simple, robust, on-scanner reconstruction	Better discriminates cancer from benign changes	Reflects multi-compartment tissue properties	Reflects actual tissue relaxation	Post-processing, pick model based on data quality
Limitations	b-range dependence and SNR bias, limited specificity, large overlap between cancer and benignity	Gradient performance demands; unharmonized acquisition and analysis; heuristic model; off- scanner, custom postprocessing	Requires custom pulse sequences, time-consuming; bias and repeatability not reported; off-scanner, custom multi-parameter postprocessing	Poor discrimination of relaxivity; custom pulse sequences required by MRF and LWI; bias of abbreviated acquisitions; off-scanner custom postprocessing	Hemodynamic dependence; many model variables; off-scanner custom postprocessing
Existing evidence for use	Multiple studies suggest ADC < 700-800 μ2/s has both sensitivity and specificity >85% for csPCa	Evidence is single center only, but DKI and RSI improve specificity to >90%, and IVIM shows sensitivity of 92% and specificity of 94%.	Only VERDICT prospectively used; will likely require new pulse sequence, post-processing	Currently in research arena only.	Model dependence on contrast agent, injection rate, temporal resolution, and arterial input function limits generalizability

IVIM = intravoxel incoherent motion, DKI = diffusion kurtosis imaging, RSI = restriction spectrum imaging, VERDICT = Vascular, Extracellular, and Restricted Diffusion for Cytometry in Tumors, HM-MRI = hybrid multidimensional MRI, MRF = MR fingerprinting, LWI = luminal water imaging,

Conventional DWI

The ability of DWI to discriminate normal glands with large cytoplasm from cytopenic cancer cells on the basis of free water motion restriction seems perfectly suited to the prostate. Quantitation of DWI has mostly used ADC as a straightforward, relatively reproducible, easily obtainable metric of diffusion that assumes mono-exponential DWI signal decay with increasing b-value. Starting nearly a quarter century ago, the advantages of adding DWI to prostate MRI were quickly and widely adopted (1).

The ability of ADC to discriminate cancer from non-cancerous prostate tissue was first investigated in 2009, with performance, as measured by the AUC, of 0.69 (2). Subsequent literature has shown an AUC as high as 0.98 (3). Moreover, ADC has shown strong promise for detecting clinically significant prostate cancer (csPCa), with AUC as high as 0.93 (4–11). However, AUC is not a particularly useful metric, as it is unclear how to implement the metric in management. Sensitivity, specificity, PPV, and NPV can be much more useful in terms of understanding a parameter’s likelihood of identifying a cancer or confidence in excluding cancer’s presence. Fewer studies have reported these metrics, which have broad ranges of 61–87% and 67–88% for sensitivity and specificity, respectively, for detection of any cancer (6,7,12–14). Similar sensitivity and specificity performance, ranging from 70–88% and 71–88%, respectively, is seen for csPCa detection (4–7,10,15).

Literature shows an inverse correlation between ADC and the presence of increasingly higher-grade cancers. However, as this relationship is complex and ADC reflects average tissue characteristics, a cutoff value is needed. As ADC values are dependent on DWI protocols, the cutoff values used are highly variable, ranging from 740 to 1332 s/mm², with some authors reporting different cutoffs across or even within publications (4–7,9–12,16). This heterogeneity prompted PI-RADS version 2.1 (v2.1) to suggest a range of 0.75–0.90 μm²/ms as an ADC threshold (68). This range of threshold values reflects inherent variability in the ADC measurement. Due to truly multi-exponential DWI signal decay in prostate, ADC is dependent on the minimum and maximum b-values used for its calculation (with 50 and 1000 s/mm² as the respective recommendations in PI-RADS v2.1) (Fig. 2), the number of excitations or signal averages (as higher b-values generate less signal), the use of intermediate b-values to compensate for signal variation, the diffusion model (linear vs exponential) used, as well as additional factors such as TE and dwell time. Also, sampling error and specific ROI histogram metric used may contribute to ADC variability: mean values within a lesion are influenced by the amount of surrounding higher-value tissue sampled, potentially making the minimum or 10^th percentile or histogram analysis preferred. Additional confounding factors include the prevalence of differing disease grades within a population due to genetics (17–19).

Figure 2:

Example of acquisition parameter dependence for grey-scale T2-weighted images (T2WI) (TE=110 ms, TR=4.7 s) and of b-value dependence for color-coded ADC maps, for 73-year-old patient with Gleason grade group 2 prostate cancer as well as peripheral zone lesion and atrophy on MRI. A) T2WI shows right midgland peripheral zone posterolateral lesion (arrow). Apparent shift of ROI outline based on DWI lesion illustrates susceptibility distortion. B) Qualitative DWI with b=1600 s/mm² (TE=77 ms, TR=7.2 s) shows restricted diffusion in lesion as high signal (arrow). C) Quantitative ADC map for b=1600 s/mm². D) Quantitative ADC map for b=800 s/mm². E) ADC(DK) map. F) K map. Maps in C-F were generated from clinical DWI with abbreviated 4-b-value DWI acquisition. ADC(DK) and K maps were derived from diffusion kurtosis model. Maps illustrate bias introduced by analysis protocol (i.e., increasing ADC for b=800 s/mm² and ADC(DK)). Color-bar indicates common parameter values scale for quantitative maps in C-F.

Given ADC’s modest ability as a discriminator, it is often combined with other quantitative (perfusion) or qualitative (tissue parameters) factors to improve its overall performance. However, the range of performance values from combined quantitative metrics does not significantly differ from those of ADC alone, and some literature has found no additive value of ADC over qualitative assessment (2, 11–13). Given the suggestion of an inverse relationship between ADC and grade, investigators have evaluated the ability of ADC to predict higher-grade disease. One early study found an AUC of 0.69 to discriminate International Society of Urological Pathology (ISUP) grade group 1 from higher-grade disease (18), with subsequent studies showing an AUC as high as 0.82 (18,21,22). Thus, ADC could help determine whether a biopsy is truly warranted (i.e., if low-grade cancer is predicted) or could prompt repeat biopsy after negative biopsy (i.e., if suspicion for csPCa remains high).

Longitudinal use of quantitative ADC metrics for detecting therapy response is an important application (23). One study evaluating anti-androgen therapy found “relative change in tumor ADC correlated significantly with PSA decrease” (24). Other studies have shown a significant change in ADC after radiation therapy and a significant difference between remission and recurrent disease, although the direction of change may indicate the use of hormone therapy (25,26). Moreover, Chatterjee et al. found that pre-treatment ADC, with a cutoff of 0.96 μm²/ms, has a sensitivity of 100% and specificity of 48% for predicting biochemical failure (27). Those authors also found, on multivariable analysis, that ADC was associated with freedom from biochemical failure per standardized follow-up criteria (serum PSA after accounting for NCCN risk category and any androgen deprivation therapy).

ADC has yet to be incorporated into any standardized assessment criteria and currently does not influence clinical prostate cancer management strategies. The reasons are likely multifactorial: variations in acquisition technique, differing performance characteristics, lack of a standard cutoff, and, most importantly, lack of prospective multicenter trials that have integrated its use into management rather than simply correlated it with clinical parameters. Typical ground-truth tissue label assignment for exploratory studies relies on prostate biopsy, which is prone to sampling errors and is not comprehensive. Validation of ADC cutoff requires alignment of imaging with histopathology data, which is sparse (Table 2) and biased toward studies of high-grade cancers. As with any quantitative metrics, establishment of ADC cutoff thresholds further necessitates evaluation of CIs including bias and repeatability (23,28).

Table 2:

Ranges of T₁, T₂, ADC_b<1K, and ADC_b>1.4K, stratified by acquisition protocol, based on mean values from available literature with histopathology confirmation

Prostate Tissue	ADCb<1K (μm2/ms)	ADCb>1.5K (μm2/ms)	T2 (ms)	T1 (s)
GS-6 prostate cancer	0.9–1.4	0.9–1.1	MESE: 80–120; MRF: 45–75	LL: 1.34; MRF: 1.6–1.7
GS-7 or higher prostate cancer	0.9–1.1	0.6 –0.8	MESE: 50–100; MRF: 35–53	LL: 1.25; MRF: 1.45–1.65

Data reported as range. Prostate cancer with GS of 7 or higher shows marked differences between ADC_{b<1K versus} ADC_b>1.4K, between T₂ measurements from MR fingerprinting (MRF) versus multi-echo spin-echo (MESE), and between T1 measurements from MRF versus Look-Locker (LL), indicating the need for bias assessment.

GS = Gleason score

Advanced Diffusion Techniques and Hybrid Diffusion Techniques

The multi-exponential DWI decay at increasing b-values observed for prostate leads to ADC dependence on the b-value range (Table 3). Numerous more complex models of diffusion have been proposed to distinguish these various components, many of which result in distinct new quantitative metrics and require multiple b-value acquisition (17–19). These include intravoxel incoherent motion (IVIM) deconvolution(for b < 100 s/mm²), which purports to separate pure diffusion from the fractional perfusion components; diffusion kurtosis imaging (DKI), defined by the difference between the measured and extrapolated diffusion signal at increasingly higher b-values (>1000 s/mm²) and reflecting the microstructural complexity of the underlying tissue; diffusion tensor imaging (DTI), which investigates fractional anisotropy by measuring more than the six cardinal directions of diffusion signal; and variations on these techniques. DTI can be further enhanced by a multishell acquisition, which acquires multiple b-values of the tensor images, and high angular-resolution diffusion imaging (HARDI), which uses more than 6 diffusion directions. Models that purport to exploit these advanced diffusion techniques include the Vascular, Extracellular, and Restricted Diffusion for Cytometry in Tumors (VERDICT) model, restriction spectrum imaging (RSI), and hybrid multidimensional MRI (HM-MRI), the last two of which combine diffusion with tissue relaxation characteristics (29–31). All of these advanced models rely on modelling DWI signal for multi-b acquisitions (>3 b-values), which prolongs examination times, and are therefore often not practical in current clinical settings.

IVIM

IVIM is one of the best-studied diffusion models across multiple organ systems. While conventional diffusion acquisitions often acquire intermediate b-values to improve ADC calculation, IVIM instead exploits the degree to which the lowest and highest b-values are differentially affected by perfusion and underlying tissue microstructural complexity. Its use in the prostate has undergone validation testing, with AUC for discriminating cancer from benign changes as high as 0.96, as well as sensitivity and specificity of 92% and 94%, respectively; nonetheless, the combination of IVIM with T2WI has even better performance (32,33). IVIM has also been used to discriminate tumor grade, with a similarly impressive performance in some hands (35), but showing no benefit over conventional ADC in other studies (34). In fact, some studies suggest no benefit of IVIM over ADC for cancer discrimination (36,37). Some of the performance variation may be explained by these complex models’ parameter dependence (38).

DKI

DKI is often compared or combined with a conventional monoexponential ADC model, and some investigators have found DKI to have added benefits for cancer detection (36,39,40). DKI has also shown benefit in some investigators’ hands for grade differentiation (18,41). However, this outcome is not universally experienced, with other investigators seeing no benefit of DKI over ADC for either cancer detection or grade discrimination (16,34,42,43). The reported cutoff thresholds for DKI diffusivity value tend to be higher than corresponding ADC thresholds (Fig. 2). A comparative review of studies for ADC versus DKI for prostate cancer observed similar performance between the methods and higher thresholds for DKI versus ADC (39).

DTI

DTI alone has shown benefit as a quantitative metric (44). The method’s added directions allow determination of fractional anisotropy (i.e., the degree of cell shape disorder within tissue). In one study, DTI alone showed a sensitivity and specificity of 81% and 85%, respectively, and an AUC of 0.92, for diagnosing prostate cancer (45); DTI showed slightly increased AUC to 0.96 by its combination with DCE imaging, with the combination of methods optimizing either sensitivity to 100% or specificity to 98% depending on whether requiring positivity of either or both techniques, respectively. However, DTI, too, shows significant parameter dependence, which may influence its performance in other scenarios (46).

Complex Customized Multi-Compartment Diffusion Models

Increasing complexity of models for calculating multi-compartment diffusion raises the possibility for model failure and increases the required DWI acquisitions. However, because these models have often been iteratively crafted and have the potential for commercial distribution outside of a generic pulse sequence, more care has been taken to ensure robustness. Complex models have not seen wide use outside of single centers and remain dependent on the b-values, number of excitations or signal averages, and TR/TE. Nevertheless, standardization of the acquisition parameters along with, in some cases, phantom validation on novel platforms, would facilitate the quantitative results to be generalized to additional institutions.

RSI is an excellent example of a newly validated quantitative metric, the cellularity index. RSI models isolate restricted and isotropic diffusion signal originating from intracellular compartment-cancer or epithelial cell nuclei; the method shows good diagnostic performance (AUC, 0.94), and moderate correlation with Gleason score (ρ = .53) (47). In one study, RSI alone for detecting csPCa, using histopathology as the ground truth, had an AUC of 0.78 compared to 0.77 for PI-RADS and 0.48 for ADC (30). A disadvantage of models such as RSI is that their highly standardized technique and FDA clearance requirements make them largely vendor-dependent, such that the scanner manufacturer must determine the protocol appropriate for a particular scanner.

Relaxivity

T1 and T2 Relaxivity Measurements

T1 and T2 are inherent properties of any tissue, primarily reflecting the interaction of tissue macromolecules and water, respectively. Their measurement methods are long established. One study found a considerable overlap of T1 and T2 values for discriminating cancer from prostatic hyperplasia and unaffected tissue (48). Another early study found significant changes in the T1 and T2 for both cancerous and uninvolved tissue after radiation therapy (49); subsequent validation work supported relaxivity but showed poorer differentiation compared to other parameters (25–27,49). In one study, T2 alone when compared with surgical pathology for prostate cancer detection had an AUC of 0.67 (2); this performance increased to 0.71 when combined with ADC and perfusion features. T1 and T2 measurements have shown good performance for discriminating cancer from benign tissue, with T1 having an AUC as high as 0.92 and quantitative T2 measurements outperforming T2WI (15, 51). Although these metrics lack other parameters’ discriminative power (Table 2), they contribute to an overall AUC boost when combined with diffusion and DCE (15,50–53). Other studies have evaluated quantitative T1 and T2 changes after biopsy or the effect of differing methods of T1 and/or T2 quantitation, but these currently lack clinical application (54–57).

MR Fingerprinting

MR fingerprinting is an increasingly popular way to obtain maps of T1 and T2, whereby a single pulse sequence produces multiple tissue contrasts. The maps all show significant discrimination of cancer from inflammation and uninvolved tissue. Studies have achieved an AUC as high as 0.94 for differentiating cancer versus non-cancerous tissue when combining ADC and T1, and as high as 0.83 for discriminating low- from high-grade cancer in the transition zone (3,6,7,58). The measured values of T1 and T2 for cancer have varied, ranging from 1450–1660 ms for T1 and 36–73 ms for T2. These values are different from those reported from other methods (e.g., 1100–1500 ms for T1 by modified Look-Locker and 50–100ms for T2 by multi-echo spin-echo), likely indicating imaging protocol biases (Table 3) (15,50,52).

VERDICT

The VERDICT process provides the quantitative metric intracellular volume fraction (FIC) by estimating three distinct signal components and associating them to intracellular water, water in the extracellular extravascular space, and water in the microvasculature. This process yields a marked distinction between cancer and uninvolved tissue on voxelwise maps (29,59). A prospective study using this method to avoid biopsy found the highest AUC for FIC of 0.96, compared with 0.85 for ADC and 0.74 for PSA density (9).

Luminal Water Imaging

Luminal water imaging is a multicomponent T2 mapping technique that isolates the luminal water fraction; the method has been applied for detecting tumor in the untreated prostate (60,61). One study suggested that luminal water imaging has superior performance to diffusion-based techniques, with an AUC of 0.98 (62). However, in an earlier study, LWI had AUC, sensitivity, and specificity values of 0.81, 75%, and 87% for differentiating cancer from non-cancerous tissue (14). Additionally, luminal water fraction (LWF) and amplitude of long T2 component showed significantly correlations with grade, achieving AUC of 0.78 for discriminating low- from high-grade disease (63).

HM-MRI

HM-MRI uses variations in both b-value and TE to measure tissue components separately (31). Individual components of HM-MRI have shown high correlation (>0.9) to histopathology fractions and high discriminatory value for cancer, with an AUC of 0.96 for epithelium and 0.94 for lumen (64). HM-MRI parameters also show a strong correlation with grade (65). This technique could improve diagnosis, having shown similar or improved diagnostic performance for csPCa, higher interobserver agreement, and lower interpretation time for less versus highly experienced readers (66).

Diffusion-relaxation correlation spectrum imaging (DR-CSI) uses diffusion-T2 relaxation data similar to HM-MRI; however, in contrast with model-based approaches, DR-CSI finds peaks in the T2-diffusion spectra to identify individual tissue components (67).

Comparison of VERDICT and HM techniques

While each of the described relaxivity techniques has shown value over conventional assessment, with AUC > 0.9 for discrimination of cancer from benign tissue, only VERDICT has been used prospectively to our knowledge. Yet, this technique is not widely available outside of a few research sites. HM-MRI is also proprietary, and all of the techniques are relatively time-consuming. A potential advantage to these techniques is associated standardization of both qualitative and quantitative images; this standardization could prevent algorithm drift if an artificial intelligence (AI) technique were to be applied to image interpretation.

DCE

The neovascularity associated with tumor growth often disturbs blood flow, making DCE a popular quantitative method for prostate cancer detection and characterization. Cancers generally show faster wash-in and wash-out and microvessel density compared with benign tissue. Simple early enhancement detection, whereby multiple T1-weighted scans are acquired rapidly and scrutinized for the time-point when contrast enhancement first appears in the prostate, is often sufficient and commonly performed. Qualitative wash-in and wash-out assessment was a component of PI-RADS version 1 (68,69). However, various systems have been developed to quantify aspects of enhancement. These range from simple measures of the degree of enhancement at an early time point to complex models that distinguish the exchange constants between the intravascular and extravascular-extracellular spaces (EES) (70,71). Given the presence of hemodynamic effects and potential differences in contrast medium, concentration, and injection rate, the number of factors that determine the overall signal is more complex than for other metrics, with some investigations focusing on components such as the arterial input function or pulse sequence, or building phantoms for standardization (72,73). One study found that qualitative DCE assessment may be higher-performing than quantitative metrics for csPCa detection on biopsy (74).

The commonest quantitative DCE parameter investigated is K^trans, the volume transfer constant, which measures the efflux of contrast media into the EES. Other DCE metrics include the flux rate constant (k_ep), the extracellular volume ratio (Ve), and plasma volume fraction (v_p), all based on the Tofts model (70). A simpler quantitative DCE parameter is the integrated area under the gadolinium enhancement curve at 60 seconds (iAUgC₆₀).

For csPCa detection, K^trans has shown similar performance to other parameters, with an AUC as high as 0.89 (5), but other work suggests an AUC as low as 0.59 (2). Other parameters–including simple wash-in metrics–have shown similar discriminatory performance, with an AUC as high as 0.94 when combined with ADC (12).

DCE is also used to discriminate low- and high-grade disease, with modest but variable reported performance and an AUC of 0.71 for discriminating ISUP grade groups 1–2 versus 3–5, but no significant discrimination of all csPCa from ISUP grade group 1 (21,37,75). Therapy response is an additional common scenario to which quantitative DCE has been applied (25,27,76). At least one DCE parameter has shown measurable changes for many therapies, including medications, radiation, and high-intensity focused ultrasound (HIFU) (76). However, the ability to discriminate treatment failure by DCE is variable, with DCE independently predicting treatment failure in only one study to our knowledge (76).

Other Quantitative Components

MRS

MRS was once the premier functional component of prostate MRI. However, the method’s long acquisition time, requirements for technical mastery, and low spatial resolution have resulted in MRS falling out of favor compared with other techniques. When performing MRS, the combined concentration of choline and creatinine (which often overlaps with that of choline alone, a marker of cell surface proliferation) relative to the concentration of citrate (which is enriched in glandular cells but consumed in high metabolic states) may be quantified based on the area-under-the-peak of a spectroscopic waveform, on a standard, or on a simple ratio (77).

Size and Shape

Size and shape are among the simplest features to measure on prostate MRI and are the only quantitative features included in standardized assessment guidelines. Lesion size guides overall suspicion for prostate cancer, and a tumor-capsule interface of greater than 1.0 cm predicts extraprostatic extension (EPE) (68). While shape may be assessed quantitatively, including through circularity and convexity, these metrics have not shown an advantage over qualitative assessments (78). Comparison of MRI examinations with surgical specimens indicates relatively consistent underestimation of tumor volume by all qualitative and quantitative techniques, but good prediction of EPE (79,80). This observation is important because, in a meta-analysis, the length of a positive surgical margin independently predicted biochemical recurrence after prostatectomy (81).

Radiomics

Radiomics is the science of extracting quantitative features from (qualitative) medical images. Often, this is done through the measurement of texture features of the underlying source images, to exploit the difference in signal intensities, heterogeneity, and other imaging features (e.g., energy, entropy, Haralick) of cancer versus benign tissue (82). These techniques, applied to T2WI alone or in combination with DWI, ADC, or DCE, have shown similar performance to other metrics, with AUC for detection of cancer as high as 0.80 and sensitivity of 89% for csPCa (83,84). Quantitative texture features have also shown potential for other applications, from post-treatment assessment to grade discrimination, hypoxia determination, and PTEN expression (85–89). However, quantifying these features requires a number of pre-processing steps, which can influence the measured feature (90). Radiomics applies quantitative analytics methods to qualitative images, rather than being a quantitative acquisition technique in itself.

Chemical Exchange Techniques

Chemical exchange transfer techniques such as amide proton transfer (APT) rely on the exchange between mobile protons in amide (−NH), amine (−NH₂), and hydroxyl (−OH) groups and bulk water. These methods been used for prostate cancer detection, whereby the measured signal-intensity on APT-weighted images is analyzed using maximum, mean, or minimum values (91). Its use has been advocated in the transition zone (TZ), where diagnosis by conventional methods is particularly difficult (92). However, as with relaxivity measurements, its clinical use in the prostate is limited by inherent normal tissue heterogeneity and low CNR.

Clinical Impact and Utility of Quantitative Assessments

The clinical utility of qualitative mpMRI is now accepted for prostate cancer staging, surgical planning, and, in particular, biopsy targeting (93–95). The use of quantitative MRI in these scenarios remains under-researched and deserves more attention. As previously noted, ADC values have shown utility in differentiating low- from high-risk prostate cancer (10,11,17,96). The appeal of the use of advanced quantitative mpMRI methods is to increase specificity for indolent cancer detection, which could reduce biopsies for patients on active surveillance. A truly multiparametric quantitative approach in an era of AI would provide broad objectivity, not only for tumor detection but also for understanding tumor biology, potentially influencing oncologic outcomes. Quantitative parameters, singly or in combination, therefore, require further investigation to determine their utility as biomarkers that influence screening strategies and disease outcomes. Validation studies to ascertain clinical utility of advanced quantitative mpMRI metrics will be aided by accelerated acquisitions using compressed sensing and deep learning, as are being increasingly implemented on clinical scanners.

Unlike a human observer’s adaptability to qualitative variability within images, quantitative assessments require that all scans adhere to a reproducible acquisition and postprocessing protocol, driven by technologists and radiologists, often without a perceived immediate clinical benefit. Scan protocol deviation, scanner platform changes, variable processing and model parameters, lack of normalization to a phantom, or patient factors that distort the magnetic field such as metallic implants, can affect quantitative measurements and must be addressed. Without a prospective clinical trial backing the data from single-center studies to provide evidence of value in disease detection or treatment outcome, the degree to which these biomarkers would necessarily change practice remains subjective. However, the Quantitative Imaging Biomarker Alliance has proposed a standard for ADC and K^trans, which, along with requisite phantom testing, may provide the first step for a universal quantitative standard for conventional DWI and DCE (28). The FDA developed corresponding guidelines for quantitative imaging devices in 2022 [97]. Indeed, a current general limitation of advanced quantitative diffusion, relaxation, and hybrid methods beyond ADC is the lack of standardized multi-platform scan protocols and multi-center, multi-platform studies that establish CIs for derived mpMRI metrics based on bias estimates and test-retest repeatability (Table 1) (28).

General Considerations for Reporting Quantitative Assessment

Currently, none of the reporting systems explicitly require reporting of any quantitative clinical parameter other than size (68,98–99). Regardless, all of these systems at least mention the use of quantitative ADC, and some site-specific reporting systems have used it as a selection criterion (100). Although sensitivity of ADC by itself is low, the addition of quantitative ADC and T2 to qualitative PI-RADS apparently helps boost specificity for csPCa detection(10,11,17,52). ADC validation studies with histopathology confirmation are usually small and biased toward high-grade cancers. Studies evaluating ADC specificity mostly report cutoff values for csPCa, whereas evaluation of false-positive reduction is needed for application to reduce unnecessary biopsies.

As the factors affecting quantitation (scanner platform, pulse sequence, post-processing models) change with time, quality assurance should be periodically repeated. Quality assurance practice is an inherent part of the American College of Radiology Prostate Cancer MRI Center Designation. While this process does not include quantitative assessments, the same process used for quality assurance may also be applied to quantitative metrics.

What is less clear, then, is how to integrate these quantitative parameters into reporting systems that are inherently qualitative. Most use a Likert ranked-category system, which can be considered semi-quantitative but clearly lends itself to different statistical validations. Therefore, integration of quantitative assessments and conventional Likert scales effectively mixes “apples and oranges” and must be applied judiciously. For now, given that the quantitative metrics covered in this article are not incorporated into standardized practice recommendations, it is crucial to communicate such data’s meaning with referrers and other practitioners who may base their management on the interpretation of these data (40). The modulation of the standard qualitative assessment by the interpretation of quantitative components and their resultant implications for the presence of csPCa should be synthesized in the report impression and recommendations.

Ultimately, one hopes for the same kind of level 1 evidence for these quantitative techniques that has emerged for prostate MRI in general. These techniques hold the promise to further elevate the quality of care, but this promise serves as a “double-edged sword,” as poorly acquired or processed data could be given undue weight. Regardless, the widespread desire for increased quantitative reporting will provide the impetus for its validation and incorporation into future recommendations. Perhaps just as importantly, uniform quality and normalization of quantitative data could serve as a powerful substrate for the development of AI algorithms. The inherent calibration required for these quantitative techniques could, in part, offset the potential for data drift that hinders the application of algorithms trained on a set of data that no longer reflects the subsequent inputs. These techniques’ incorporation into current protocols will likely reflect site preference and local demands until they become widely available from vendors and incorporated into societal guidelines.

Highlights.

• Quantitative techniques can be applied to many aspects of prostate MRI, from ADC values to tissue relaxivity, advanced diffusion techniques, and quantitative perfusion.

• Quantitative techniques have consistently shown value over qualitative assessment for detection of clinically significant cancer and recurrent cancer, and discrimination of low- from high-grade disease.

• Integration of quantitative techniques into clinical protocols will require standardization of acquisition parameters and multicenter validation of improved value.