Quantitative magnetic resonance imaging in prostate cancer: A review of current technology

Abstract

Prostate cancer (PCa) imaging forms an important part of PCa clinical management. Magnetic resonance imaging is the modality of choice for prostate imaging. Most of the current imaging assessment is qualitative i.e., based on visual inspection and thus subjected to inter-observer disagreement. Quantitative imaging is better than qualitative assessment as it is more objective, and standardized, thus improving interobserver agreement. Apart from detecting PCa, few quantitative parameters may have potential to predict disease aggressiveness, and thus can be used for prognosis and deciding the course of management. There are various magnetic resonance imaging-based quantitative parameters and few of them are already part of PIRADS v.2.1. However, there are many other parameters that are under study and need further validation by rigorous multicenter studies before recommending them for routine clinical practice. This review intends to discuss the existing quantitative methods, recent developments, and novel techniques in detail.

Key Words: Prostate cancer; Quantitative imaging; Magnetic resonance imaging; Apparent diffusion coefficient

Core Tip: Quantitative imaging has many advantages over conventional qualitative assessment. A few parameters have also been shown to correlate with the Gleason score and can help in deciding disease prognosis and clinical management. A quantitative imaging biomarker could improve prostate cancer (PCa) detection by minimizing inter-observer variability, thereby reducing overdiagnosis of clinically insignificant PCa (Gleason score < 7). This would help avoid unnecessary biopsies and decrease the overtreatment of slow-growing PCa. In addition, with further advancement in the quantitative imaging parameters, they may be used to monitor therapeutic response or to predict response to a particular treatment.

INTRODUCTION

Prostate cancer (PCa) is the second most common malignancy and one of the main causes of cancer related mortalities in males around the globe[1]. Traditional diagnostic methods like digital rectal examination, prostate specific antigen tests, and transrectal ultrasound (US)-guided biopsy have poor sensitivity and specificity and face limitations in staging and grading of PCa. Therefore, for better prognosis effective management, and treatment planning, accurate characterization of PCa is necessary[2].

Current imaging modalities available for imaging of PCa are transrectal US, multiparametric magnetic resonance imaging (MRI), positron emission tomography-computed tomography (PET-CT) scan, and PET-MRI (PET-MRI) scan. US plays a limited role in cancer detection as very small proportion of hypoechoic lesions are revealed to be tumors upon biopsy. Thus, it is used to visualize the prostate during transrectal or trans-perineal US-guided biopsies. CT offers high specificity in detecting very high-grade tumors but is still not ideal for diagnosing PCa as it lacks soft tissue resolution and molecular insights needed for accurate PCa detection[3]. PET scan uses a tracer to target prostate-specific membrane antigen which helps to detect lymph node metastasis in men with high-risk PCa. Prostate-specific membrane antigen PET has proven to be more effective than conventional CT in accurately staging high-risk PCa patients[4]. MRI, is a noninvasive method that integrates functional and morphological information to provide a superior soft-tissue contrast resolution, which permits resolution of intra-prostatic anatomy, discrimination of the prostate capsule, and differentiation of benign from malignant intraprostatic abnormalities which the other mentioned imaging techniques often lack[5].

Imaging modalities may provide structural as well as functional information, depending upon the modality and technique. However, most of the imaging assessment is qualitative i.e., based on visual inspection and thus subjected to inter-observer disagreement. To overcome this and to enhance the objectivity of reporting, quantitative imaging techniques have been recommended which may play a major role. Quantitative imaging aims to assign a numerical value to the observations enabling higher accuracy, reproducibility, standardized reporting, interpretation, and communication. Few such MRI quantitative metrics are already part of PIRADS v.2.1[6]. However, many are still under study and need further validation. This review intends to discuss the existing quantitative methods, recent developments, and novel techniques in detail.

QUANTITATIVE PARAMETERS

Apparent diffusion coefficient

Diffusion-weighted imaging (DWI) is an important part of multiparametric MRI and enables the assessment of changes associate with the tissue organization in progression of PCa. High cellularity tissues by virtue of reduced extracellular space restrict the Brownian movement of water molecules and thus demonstrate diffusion restriction[7]. DWI acquisition sequence can be used to obtain two sets of images: (1) Diffusion weighted imaging acquired at different b-values, where, b-value signifies diffusion sensitizing gradient’s strength with its unit being seconds/mm²[8-10]; and (2) Apparent diffusion coefficient (ADC) map generated from the diffusion weighted images acquired at least two different b-values. When the logarithm of signal intensity decay on the Y-axis is plotted against b-values on the X-axis, the slope of the line which is produced represents the ADC value[7,9,10]. A minimum of two b-values are required for the calculation of ADC. DWI acquired at more than two b-values enhance the precision of ADC estimation at the cost of increased acquisition time[7]. There is a lack of standardization of number of b-values and highest b-value to use, which leads to differences in the calculation of ADC values. Many recent studies have proven the superiority of high b-value DWI (1500-2500 s/mm²) over conventional DWI (b = 1000 s/mm²)[11-18]. However, the disadvantage of a high b value includes a reduced signal-to-noise ratio, more susceptibility artifacts, and image distortion[19]. Specific software is also required for ADC calculation if using such high b-value[20]. Thus, routinely most of the centers prefer conventional DWI. PIRADS v.2.1 guidelines also recommend the use of b values ≤ 1000 s/mm²[21].

Under PIRADS v.2.1, DWI is the primary sequence for assessing peripheral zone (PZ) lesions and a secondary sequence for transitional zone (TZ) lesions[22]. Low ADC values have been reported in TZ in benign stromal hyperplastic nodules, which is the reason for not using DWI as a primary sequence for TZ[22]. Few studies have demonstrated higher ADC values in stromal hyperplastic nodules as compared to PCa, however, there is no clear-cut value to distinguish both and thus the use of quantitative ADC in the diagnosis of PCa in TZ is still not validated. DWI provides both qualitative (by assessing whether the tissue in question is showing restriction or not) and quantitative assessment. Various quantification metrics which are being explored include mean ADC, ADC ratio, ADC minimum, and histogram analysis. Currently, mean ADC is the only quantitative DWI parameter included under PIRADS v.2.1. Available literature has shown that the mean ADC value has an inverse relation with histological Gleason’s score and thus it can be used to predict whether the prostate malignancy is high grade or low grade[23-25]. Lucarelli et al[5] showed that the mean ADC values decrease from International Society of Urological Pathology (ISUP) 2 to ISUP 5 showing significant differences between low, intermediate, and high-grade tumors. ADC values showed a strong positive correlation with ISUP groups in both the peripheral and transition zones; however, the correlation between ADC values and PIRADS groups is less reliable in the transition zone than in the peripheral zone[5]. One major limitation of mean ADC is averaging ADC values and masking of low ADC pixels which contain more aggressive tumor foci[26]. Also, there are no clear cut-off mean ADC values for benign and malignant lesions, as ADC value is highly influenced by many factors including the type of scanner used, b-values, physiological differences in background prostate tissue, etc[27-30]. To compensate for these differences, some authors have advised the use of ADC ratio which is defined as the ADC value of the prostatic area suspicious of malignancy divided by the ADC value in the non-malignant reference area[31-33]. There is no validated reference with respect to which ADC ratio should be calculated, different studies have used different references including normal TZ, and PZ, and some have also used urinary bladder as a reference[33,34]. The advantage of this method is that with no additional scan time, it allows lesion visualization by providing a highly diffusion-weighted image[35]. Also, there are two schools of thought on how to draw region of interest (ROI) for ADC ratio calculation. One group used ROI covering ADC minimum area[23,36] yielding ADC minimum ratio and another group advocated ROI covering whole lesion yielding ADC mean ratio[37,38]. A study compared both and found that ADC mean and the resulting ADC mean ratio was significantly better compared to ADC minimum and resulting ADC minimum ratio in discriminating high-grade from low-grade PCa[33].

ADC minimum and ADC histogram analysis have been proposed as two quantitative DWI methods which are beneficial especially in heterogeneous lesions[26,39]. ADC minimum is more sensitive to detecting aggressive tumor foci in a heterogenous mass with varying ADC values[40]. And, thus can act as a better targeting guide in prostatic biopsies. However, there is mixed data, as to whether ADC mean or ADC minimum correlates better with Gleason score (GS)[26,39,41]. The limitation of minimum ADC is being restrictive and more prone to fallacious due to artifacts[33,40,42]. Also as already discussed, ADC mean values due to averaging can mask the areas with minimum values, thus can underestimate the aggressiveness in a heterogenous lesion. To overcome this, some authors have come up with the concept of histogram analysis. Homogenous composition tumors will have the gaussian type of distribution as opposed to skewed distribution in heterogenous lesions[26]. The low centile values in the histogram have been shown to correlate better with GS[26,39,41]. However, the optimal centile that should be taken for analysis lacks standardization[36,42].

Tumor size and tumor volume by three-dimensional ellipsoid fit

Tumor size is an important part of the decision tree of PIRADS v.2.1 and the 15 mm cut-off differentiates PIRADS 4 from PIRADS 5 lesion[6] (Figure 1). This has been independently validated by many investigators. The measurement of the single longest diameter is easy to perform and has a good positive correlation with tumor volume (TV) at radical prostatectomy[43].

Figure 1 PIRADS 4 vs PIRADS 5 lesions on basis of size: T2 weighted axial image. A: A right peripheral zone ill-defined hypointense lesion measuring 727 mm suggestive of PIRADS 4 lesion; B: T2 weighted axial image showing a left peripheral zone ill-defined hypointense lesion measuring 17 cm suggestive of PIRADS 5 lesion.

TV is associated with the prognosis of PCa[44,45]. TV more than 0.5 cm³ is considered the threshold for significant disease[46]. Index lesion volume helps in deciding management, and guiding focal treatment, and is a good criterion for follow-up during active surveillance[47,48]. Focal therapy options such as high intensity focused US, cryotherapy, laser ablation, brachytherapy, electroporation, and radiofrequency ablation[49] require precise assessment of TV so that maximum energy is deposited in the cancerous tissue with minimal damage to the adjacent normal area[50,51]. Studies have found that pathological TV correlates with pathological stage and GS[43,44,52,53]. So, if TV on imaging can truly represent the pathological volume, it will also show a similar correlation with GS. A study by McNeal et al[54] demonstrated that if TV is less than 4 cm³ the disease is usually confined to the prostate, and when TV is more than 12 cm³ there is a high likelihood of distant metastasis. TV also correlates with the presence of vascular invasion and seminal vesicle involvement[55]. Many studies have shown that TV is better measured on ADC image as compared to other sequences, as ADC image best demarcates the margin of the lesion in contrast with the adjacent normal tissue[25,56]. A study also advocated an ADC value of 0.0016 mm²/second as the cut-off for accurate measurement of TV (49)[57]. PIRADS v.2.1 assessment includes only the two-dimensional method for TV measurement. However, the drawback of using one of the most widely used two-dimensional conventional formulae for TV calculation i.e., the ellipsoid formula is that it is a linear measurement based on the assumption of the lesion being of ellipsoid shape and does not take into account the irregular tumor shape (48)[58]. Three-dimensional measurement techniques of TV have not been thoroughly investigated. TV estimation by the ellipsoid fitting model has the potential to measure the tumor maximum dimension and TV accurately and may upgrade the PIRADS score (Figure 2) from 4 to score 5 (48)[58].

Figure 2 Three-dimensional ellipsoid fit: Three-dimensional fitting of the segmented lesions from. Diffusion-weighted magnetic resonance images (b = 2000 s/mm²) for different PI-RADS scores.

Shape analysis

Shape analysis is one of the potential quantitative parameters, which can have an important role in staging and prognosis[59]. Currently, PIRADS V2.1 includes only qualitative shape criteria, which are separate for TZ and PZ lesions. In TZ, lenticular shape favors malignancy, while round or oval shape favors benign prostatic hyperplasia (BPH)[22,60,61]. While for PZ, wedge, linear, shape, bandlike, diffuse shape have been described to occur more frequently in benign conditions, on the other hand, round or oval shape has been associated with PCa[22,62] (Figure 3). Shape assessment is done on DWI sequence for PZ lesions[6] and on T2 sequence for TZ lesions[6]. Realizing the important role of shape analysis for PZ lesions, PIRADS v.2.1 has incorporated shape analysis of PZ lesions as well into decision making, while it was not included in the PIRADS V 2[6,22,60]. Various studies have been done to validate the importance of shape analysis in TZ and PZ lesions. In a study by Li et al[63], the ability of radiological semantics to discriminate clinically significant PCa (CsPCa) was studied on multiparametric MRI and it was found that kappa score for qualitative shape analysis showed only moderate interobserver agreement. In another shape metrics study for PZ tumors on DWI in 241 patients, the agreement of shape assessment was also moderate. The round or oval shape of the tumor was associated with a higher GS as compared to crescentic tumors and linear or wedge-shaped tumors (P = 0.011). In addition, tumors with round or oval shapes showed a greater degree of extracapsular extension (ECE) and seminal vesicle invasion[64] (extracapsular extension and seminal vesicle invasion: 70.1% and 26.0%) as compared to crescentic tumors (67.3% and 9.1%; P = 0.003) and linear or wedge-shaped tumors (40.6% and 9.4%; P = 0.008)[64]. Quantitative descriptors used for round and oval lesions were roundness and circularity, which were associated with more aggressive PCa[64]. A quantitative shape analysis study by Krishna et al[65] on TZ lesions on T2-weighted imaging compared it with subjective shape analysis. Circularity (correlating with round lesions) and convexity (correlating with lentiform lesions) shape features showed excellent inter reader agreement in differentiating TZ PCa and BPH; however, the quantitative feature representing lesion topology (number of spiculations) was not able to differentiate TZ PCa and BPH and was and thus inferior in accuracy than subjective shape analysis. In a study, it was shown that an advanced radiomic feature derived from an ADC map known as the surface area to volume ratio came out as a highly accurate and promising tool in discriminating CsPCa from non-CsPCa, and even outperforming previously described parameters such as TV and maximum diameter (5,7)[66].

Figure 3 Shape analysis. A-C: T2-weighted axial image shows a wedge-shaped T2 hypointense lesion (arrow in A) in right peripheral zone which shows no diffusion restriction (arrow in B and C) is scored as a category 2 observation according to PIRADS v.2.1.

Length of capsular contact

Length of capsular contact (LCC) is the length for which the prostatic tumor is in contact with the prostatic capsule[67]. It is a simple measurement, relatively independent of the radiologist’s experience, and more objective with good inter-reader reproducibility. This makes it a good quantitative criterion for diagnosing ECE, which is a crucial point in PCa staging and prognosis[68] (Figure 4). ECE in patients with T3bN0 PCa is linked with a greater risk of systemic spread of the disease and mortality. The presence of ECE is also an important treatment deciding point. If present, it upgrades the disease stage from T2 (organ-confined disease) to T3a (locally advanced disease)[68]. The curative treatment is most likely if the stage is ≤ T2c[69]. Currently, the criteria being used for diagnosing ECE in PIRADS v.2.1 include abutment or bulging of the capsule, irregularity of prostatic capsule, adjacent neurovascular bundle thickening, invasion of peri-prostatic fatty tissue and obliteration of the recto-prostatic angle[70]. However, these criteria are subjective and the reader’s experience profoundly affects ECE detection as shown in a study by Wibmer et al[68]. Also, these criteria diagnose visible macroscopic ECE, not microscopic one[70]. Thus, there should be some relatively more objective criterion such as LCC to overcome the shortcomings of above listed subjective criteria.

Figure 4 Length of capsular contact. A: T2-weighted axial image shows T2 hypointense lesion in right peripheral zone having wide contact (9.6 mm) with the capsule i.e., wider length of capsular contact suggesting the possibility of microscopic capsular invasion which was confirmed on histopathology along with its high grade (Gleason score 4 + 5); B: T2-weighted axial image in another case shows T2 hypointense lesion in left transitional and peripheral zone having wide contact with capsule i.e., wider length of capsular contact with adjacent neurovascular invasion (arrow).

LCC is found to be an independent and reproducible predictor of ECE in multiple studies[67,71-73]. In a study by Baco et al[67], the logistic regression analysis revealed that the pathological-LCC (0.821) superiorly correlated with microscopic ECE as compared to the pathological cancer volume (0.685). Spearman correlation between the pathological-LCC and MRI-LCC was significant with r = 0.839 (P < 0.0001). MRI-LCC threshold of 20 mm showed superior accuracy compared to subjective analysis in diagnosing microscopic ECE (82% vs 67%, P = 0.015). This concludes that LCC can outperform subjective criteria in predicting microscopic ECE. A meta-analysis by Kim et al[74] concluded, the more the MRI-LCC value, the higher is the probability of ECE. In a cohort study[75], LCC was found to be an independent predictor of pathological lymph node, and biochemical recurrence along with pathological ECE. A retrospective study also found, LCC and ADC when used in combination, improved the diagnostic performance in demonstrating microscopic ECE[76]. Limitations of LCC are the inter-individual differences in measurement technique (by linear or curvilinear method)[72,73], no defined threshold (varying between 6-20 mm)[72,76,77], no consensus on which sequence to be used for measurement. These points need further validation before its integration into routine clinical practice.

Dynamic contrast-enhanced MRI-based parameters (Ktrans, Kep)

Angiogenesis is considered as a hallmark of any malignancy including PCa and dynamic contrast-enhanced (DCE) imaging enables non-invasive assessment of tissue vascularity at the cellular level. And by assessing that, it has the potential to assess the aggressiveness of the tumor[24,78-81]. In PIRADS v.2.1, DCE imaging has accessory role as compared to the primary sequences such as T2 in TZ and DWI in PZ. Its main role is in PZ PIRADS 3 lesion, positive finding on DCE MRI upgrades score to PIRADS 4[22]. Lack of quantification is one of the major reasons behind its diminished role in PIRADS and quantification can help in utilizing the full potential of this technique. DCE is a functional imaging technique that offers qualitative, semiquantitative as well as quantitative assessment of PCa[82] (Figure 5). Quantitative assessment is preferred, as it provides objective reproducible numerical values and commutation is also easy these days due to easy availability of software packages. Quantitative kinetic parameters encompass Ktrans (volume transfer constant) and Kep (rate constant). Ktrans is equal to the product of permeability and surface area (P × S) per unit volume of the tissue[82]. Kep (Ktrans/Ve), refers to the efflux of contrast from the extracellular compartment back into plasma (Ve refers to extracellular extravascular volume fraction)[82]. A retrospective study of PZ lesions at 3T found a significant correlation between Ktrans, Kep values, and aggressiveness of PCa[83]. The quantitative DCE parameters, Ktrans and Kep are highly correlated with the histopathology of the prostate tumor tissue and act as independent predictors of malignancy. Higher the Ktrans and Kep values, the lesion is more likely to be malignant. The study identified cutoff points of ± 0.2905 for Ktrans and ± 0.3365 for Kep, with sensitivities of 88.2% and 94.1%, specificities of 84.6% for both, positive predictive values of 88.2% and 88.9%, and negative predictive values (NPV) of 84.6% and 91.7%, respectively[84]. Few studies have also assessed the combined diagnostic accuracy of DCE with diffusion tensor imaging (DTI)[85] and also DCE with DWI[86] in PCa diagnosis. The combination helps in assessing different facets of the disease process, DCE assesses the micro-vessel density and permeability, while DTI and diffusion focus on cellular density. One such study by Li et al[85] in PZ lesions on 3T MRI found that DTI + DCE was significantly better than either DTI [area under the receiver operating characteristic curve (AUC) 0.93 vs 0.86, P = 0.0017] or DCE (AUC 0.93 vs 0.84, P = 0.0034) alone. On the contrary, many studies have also found an overlap of Ktrans and Kep values for benign and malignant lesions[87,88] and also there are studies that demonstrated no correlation between any of the DCE parameters and GS[89]. The study performed by Feng et al[90], showed that as a single predictor DCE [the odds ratios (OR) of Kep (0.987, P > 0.05) and Ktrans (0.794, P > 0.05) closer to 1] does not contribute to the CsPCa model. However, the diagnostic and positive predictive value of the multiparametric model were significantly higher than those of the biparametric model (prostate MRI without DCE)[90]. This concludes DCE should be interpreted always in conjunction with other MRI parameters for better results.

Figure 5 Dynamic contrast-enhanced magnetic resonance imaging based parameters. A: Multiphasic magnetic resonance imaging images show PIRAD 5 lesion in right transitional zone, peripheral zone and left peripheral zone (asterisk); B: Dynamic contrast-enhanced magnetic resonance imaging images show various quantitative parameters. Region of interest has been drawn depicting prostatic cancer and benign prostatic hyperplasia (BPH) nodule. The table in B demonstrates higher Ktrans and Ve value in the prostatic cancer region as compared to BPH nodule. While Kep value is lower in the prostatic cancer region as compared to BPH nodule. The qualitative signal intensity curve in B demonstrates rapid wash in (steeper slope) and fast wash out in prostatic cancer region which is the characteristic curve for malignant tumors.

Intravoxel incoherent motion and Kurtosis

Le Bihan et al[91] and Le Bihan[92] first introduced the concept of intravoxel incoherent motion (IVIM), merging the diffusion and perfusion components, for quantitative analysis of the microstructure and microvasculature of the tissue. It is an extended version of DWI and can be used in the detection of PCa in PZ and TZ. DWI is based on the mono-exponential decay theory, assuming the gaussian distribution of Brownian motion of water molecules. However, such an assumption doesn’t always hold true in vivo, where countless numbers of barriers and inter and intracellular compartments restrict diffusion. This leads to non-gaussian distribution pattern[93,94]. IVIM tries to capture this intravoxel diffusion heterogeneity of the tumor. Multiple b-values ranging from 0 to 2000 s/mm² are used. A study by Malagi et al[95] concluded that a combination of 8 b values provides reasonably good accuracy in IVIM parameters. IVIM is based on a bi-exponential model and yields pure diffusion parameters (D) and perfusion-related diffusion parameters (D^* is perfusion-related diffusion coefficient and f is perfusion fraction) (Figure 6). Kurtosis imaging is another upgradation of conventional DWI, which utilizes kurtosis-based diffusion model to characterize the multiexponential behavior of diffusion decay. IVIM-diffusion kurtosis imaging (DKI) model is traditionally a 3 T model. However, a study by Malagi et al[96] demonstrated that using an advanced spatial penalty-based reconstruction method on 1.5 T can be used as an alternative to 3 T with fewer estimation errors in distinguishing PCa and BPH.

Figure 6 Intravoxel incoherent motion: Intravoxel incoherent motion-diffusion kurtosis imaging at b value 2000 second/mm² showing apparent diffusion coefficient map, D map, D* map, f map and k map. The prostatic cancer region is hyperintense in the background of hypointense gland on b 2000 second/mm² image and hypointense on apparent diffusion coefficient image.

Previous studies demonstrated that IVIM parameters were inferior to ADC in the evaluation of TZ lesions[97]; however, could increase the diagnostic yield in PZ lesions[95]. A study by Chang et al[98], found that D mean and D*kurtosis were important predictors of the postoperative ISUP high-risk group (P < 0.05). Liu et al[99] in comparison of the following models i.e., the monoexponential model (conventional DWI), IVIM, IVIM kurtosis, and kurtosis model, found that each model was useful for PCa diagnosis, however, diagnostic efficacy was similar amongst them. A recent study by Das et al[100] evaluated the role of combined IVIM-DKI and their machine-learning-based texture analysis (TA) for the detection and assessment of severity in PCa. The results of the study reported that D, f, and k computed using the IVIM-DKI model with the TV method were able to differentiate PCa from BPH and normal PZ. Texture features of combined IVIM-DKI parameters showed high accuracy and AUC in PCa detection[100]. Few studies have also demonstrated that D performed better than ADC and other parameters in distinguishing high-grade and low-grade PCa[95,101-103]. Thus, there is a controversy as to whether IVIM adds some value over and above ADC or not as there are studies that have found no added role of IVIM over ADC[104].

T1 relaxometry and MR fingerprinting

T1 relaxometry is a potential quantitative technique that enables the assessment of tissue properties by measuring the T1 relaxation time. T1 relaxation is an intrinsic property of any specific tissue and it is defined as the time taken by the longitudinal magnetization to recover 66.6% of its initial value[105]. Conventional sequences can quantify either T1 or T2 relaxation at one time, while the recently introduced MR fingerprinting technique enables quick and concomitant production of quantitative maps of different tissue properties including T1 and T2 relaxation time[106-109]. This technique is based on the pseudorandom variation of MRI properties like repetition time and flip angle so that unique signals are generated for each combination of tissue properties and then a chain of all possible signal evolutions is calculated for that particular sequence[110]. So far, multiple studies have been carried out to verify the role of T2 relaxation time[108,109]. In a recent study including 104 PZ cases, it was found that a combination of T1, T2 relaxometry, and ADC mapping could be useful for quantitative analysis of PCa grades and distinguishing it from benign lesions on T2-weighted images[110].

In comparison to T2 relaxometry, T1 relaxometry is relatively less studied in TZ and PZ lesions. It could be specifically more useful in differentiating prostatitis from PCa in TZ, as these both share a similar appearance on T2 images[111]. In one such study by Panda et al[111] in TZ lesions, the best results were shown by the combination of T1 and ADC in distinguishing benign and malignant lesions (AUC = 0.94) as well as in discriminating CsPCa from non-CsPCa (AUC = 0.81). ADC value lower than 0.70 × 10^-3 mm²/second was highly sensitive and a T1 value lower than 1510 ms was found to be more specific for distinguishing benign and malignant lesions. There was an overlap of T2 values between the groups as expected. A study by Yu et al[112] in 2017 was the first one in the literature to study T1 differences between benign and malignant PZ lesions. In their study, T1 was significantly less in PZ Ca as compared to adjacent normal PZ. They also pointed out that the history of recent biopsy should always be elicited from the patient, as post-biopsy haemorrhage itself shortens the T1 time, resulting in fallacious results. However, the role of T1 relaxometry in PZ lesions still needs validation by larger studies.

TA

Tumour heterogeneity plays an important role in predicting a tumour’s aggressiveness. TA is a statistical tool to measure this tumor heterogeneity through a complex analysis of the spatial distribution of image pixels, grossly not recognizable on conventional sequences[113] (overview of framework for TA shown in Figure 7). Thus, TA could become a non-invasive imaging biomarker for PCa diagnosis, prognosis, and follow-up. The most widely used and verified technique of quantitative TA is the filtration-histogram technique[114]. First-order TA features are based on the extraction of pixel intensity values within a specific area of interest. Second-order statistics measure the association between two pixels while higher-order TA analyses the association among more than two pixels. Some of the texture features which are most widely studied and commonly employed in PCa are entropy (represents histogram’s randomness) and kurtosis (represents histogram’s peakedness/flatness)[115]. A recent study suggested that after applying the filtration technique, a medium texture scale (3-5 mm) is better to assess tumoral heterogeneity compared to a fine texture scale (2 mm)[116]. A study by Wibmer et al[117] showed that ADC-derived higher entropy values (P = 0.0069) and lower energy values (P = 0.0039) showed good correlation with GS, but there was no association of T2-based texture features with GS. Association between any specific texture feature and GS was studied by Baek et al[118] using bi-parametric MRI. They found that ADC-derived grey level co-occurrence matrix entropy was associated with GS in patients with PCa, with moderately good (82%) accuracy for discriminating CsPCa from non-CsPCa. Significant association between ADC entropy and PCa aggressiveness has also been shown by studies, with improved sensitivity for ECE detection compared to subjective analysis[119]. A recent study compared ADC-derived texture features and T2-derived texture features[120]. They found that the combination of three ADC texture features i.e., skewness, kurtosis, and entropy predicted high-grade PCa. Texture features of combined IVIM-DKI parameters showed high accuracy and AUC in PCa detection[100]. All these above-mentioned studies bring us to conclude that texture features based on the ADC map have proven to be better in PCa diagnosis and risk stratification. T2-weighted textural features correlation with GS has varied among various studies, few studies have found a correlation while others have not[117,120]. TA can also better detect intraductal carcinoma which usually remains undetected on conventional MRI sequences[121].

Figure 7 Overview of the framework for texture analysis. MRI: Magnetic resonance imaging; ADC: Apparent diffusion coefficient; DWI: Diffusion-weighted imaging; FOS: First-order statistics; GLCM: Grey level co-occurrence matrix; GLRLM: Grey level run length matrix; KNN: K-nearest neighbor; LTEM: Laws texture energy measures; ROI: Region of interest; SFM: Statistical feature matrix; SVM: Support vector machine.

Magnetic resonance elastography

Magnetic resonance elastography (MRE) is a non-invasive method that uses low-frequency vibrations to quantitatively measure tissue elasticity and stiffness. In comparison to other quantitative MR techniques, it has the potential to overcome the common limitation of all techniques i.e., inter-observer variability. In a prospective study by Li et al[122] involving 73 patients, stiffness and fluidity quantification using tomoelastography improved the diagnostic yield of multiparametric MRI in detecting PCa in both PZ and TZ. Pre-operative MRE-based PCa stiffness measurement can even help in predicting the degree of lymph node metastasis, with sensitivity and specificity as high as 100% and 86.5%, respectively[123].

MRE is a non-invasive method that uses low-frequency vibrations to quantitatively measure tissue elasticity and stiffness. In comparison to other quantitative MR techniques, it has the potential to overcome the common limitation of all techniques i.e., inter-observer variability. In a prospective study by Li et al[122] involving 73 patients, stiffness and fluidity quantification using tomoelastography improved the diagnostic yield of multiparametric MRI in detecting PCa in both PZ and TZ. Pre-operative MRE-based PCa stiffness measurement can even help in predicting the degree of lymph node metastasis, with sensitivity and specificity as high as 100% and 86.5%, respectively[123].

CONCLUSION

Quantitative imaging has many advantages over conventional qualitative assessment. A few parameters have also been shown to correlate with the GS and can help in deciding disease prognosis and clinical management. Many such MRI based quantitative imaging metrics have been discussed thoroughly in this article. Few of these are already part of PIRADS, while many are under study and can be incorporated into clinical practice in the future after validation by rigorous multicentre studies. One major hurdle in achieving the quantification is the fact that numerical values obtained tend to differ from center to center because of many factors including differences in scanners, individual imaging protocols, and techniques, etc. The applications of features like TV and size, LCC, ADC values need a standardized cut off values to achieve their full clinical potential. An ideal metric would be one that is easy to calculate, and less prone to errors because of differences in scanners, individual imaging protocols, and techniques. A quantitative imaging marker could improve PCa detection by minimizing inter-radiologist subjectivity, thereby reducing overdiagnosis. This would help avoid unnecessary biopsies and decrease the overtreatment of slow-growing PCa.