The future direction of imaging in prostate cancer: MRI with or without contrast injection

3.1 European Association of Urology (EAU)

Albeit EAU guidelines⁶ do not recommend mpMRI as first screening step for PCa diagnosis, it recommends mpMRI as first-line study in naïve patients with suspicion of PCa, before considering prostate biopsy, following the PI-RADS recommendations for mpMRI acquisition and interpretation. Naïve patients should undergo targeted and systematic prostate biopsy, when mpMRI is positive, with a PI-RADS score ≥3. When negative (PI-RADS score ≤2) without any clinical suspicion of PCa exist, prostate biopsy is not recommended. Patients who have already undergone prostate biopsies, with negative results, must undergo MRI-targeted biopsy whenever mpMRI is positive.

3.2 National Comprehensive cancer Network (NCCN)

According to NCCN recommendations, mpMRI should be considered before performing transrectal ultrasound-guided biopsies (TRUS-GB), to accurately characterize prostatic tissue and possible neoplastic foci, to guide the procedural planning. However, NCCN panel affirms that the available data are not consistent enough to recommend a generalized use of pre-biopsy mpMRI, mostly due to the lack of an appropriate inter-reader variability. Furthermore, they state that MRI-guided targeted biopsies may be recommended in centers where MRI is available and where there is a high expertise among radiologists in prostate mpMRI reading. NCCN panel highlights the importance of mpMRI to approach high-risk patients with a prior negative prostate biopsy, to efficiently rule-out csPCa. Furthermore, MRI and MRI-targeted biopsies are proving extremely promising performances; however, the NCCN guidelines state that a combination of MRI-targeted biopsy with the systematic technique can better detect csPCa, with higher sensitivity. Nonetheless, the topic is a very controversial point in literature and the use of MRI-targeted techniques, if available, should always be pursued.^{12, 13}

3.3 National Institute for Health and Care Excellence (NICE)

NICE guidelines state that MRI should be routinely proposed to patients when a radical treatment is being considered. On the other hand, MRI must be proposed as first-line examination to all men with clinical suspect of localized PCa. According to NICE, MRI should be scored using the Likert assessment scale, and patients with findings scored as >3 should be counseled for MRI-directed biopsy. MRI-directed biopsies have proved to be more cost-effective than systematic sextant biopsy since csPCa is more likely to be detected. Patients with ciPCa enrolled in active surveillance protocol, should be investigated with mpMRI, especially if they have never undergone it. If MRI reports do not correspond to biopsy results, a new MRI-directed biopsy is recommended.¹⁴

3.4 American Urological Association (AUA)

American Urological guidelines have been also endorsed by the American Society of clinical Oncology (ASCO). No clear recommendations exist for naïve patients. According to AUA, mpMRI should be considered for all patients in active surveillance protocols, in men with localized PCa, and MRI-directed biopsy should be offered to these patients for a correct management.^15-19

5.1 MRI performance predictive factors

The correct protocol and image acquisition are essential for proper interpretation and reporting of prostate MRI. The appropriate protocol acquisition parameters are fully described in the PI-RADS document and more explicitly enlisted in the paper published by Engels et al. in 2019.⁴⁷ The lack of adequate equipment, protocol optimization, and patient preparation often determines suboptimal signal-to-noise ration that does not allow to safely rule in and out all significant lesions. Recently, Giganti et al.⁴⁸ published the first attempt to develop an MRI quality scoring system: The Prostate Imaging-Quality (PI-QUAL) from the PRECISION trial. PI-QUAL might offer clinicians a scoring system for evaluating and reporting the quality of their prostate MRI scans. De Rooij et al.^{34, 35} released a list of consensus statements on MRI images quality. Recommendations included checking and reporting on image quality, to visually assess images adequacy for determining diagnostic acceptability, to control image quality at 6 months intervals or in 5% of studies, and to standardize ADC measurements on phantom.

As poor image quality, also reader expertise is a crucial point of a highly diagnostic prostate MRI report. Beginners can often misinterpret a suspicious lesion on MRI, performing both under- and overdiagnosis and treatment, lowering the MRI potential to detected csPCa. Both reading errors, MRI learning curve, and training needs to be considered.

Pitfalls in prostate MRI exist and paraphysiologic prostate appearances can often mimic cancer.^{49, 50} Most common signs mimicking PCa are the bilateral basal hypointense zones (mustache sign) due to the compression of the central zone by benign prostatic hyperplasia (BPH); the median posterior hypointense area at the middle third of the gland, in which the central zone appears compressed between the transition and peripheral zone (reversed teardrop sign); BPH foci of stromal hyperplasia; ectopic BPH nodules at the periphery of the gland; granulomatous prostatitis after intravesical BCG instillation; hypertrophic anterior fibromuscular stroma; hypertrophic periprostatic venous plexus and neurovascular bundle.

As of young trainees learning curve for prostate cancer diagnosis with MRI and MRI-Directed Biopsy (MRDB), there is little evidence in literature. In 2019, Gatti et al.⁵¹ investigated the detection of csPCa by readers with different experience, comparing performance of bpMRI, with mpMRI as reference. Authors showed that readers with no experience performed significantly worse in bpMRI and indicated 700–800 cases as threshold for reliable interpretation with bpMRI. In 2017, Calio et al.,⁵² showed that after an early learning period, fusion biopsy detected higher rates of csPCa and lower rates of ciPCa compared with systematic biopsy. Later, X. Meng et al.⁵³ showed a significant improvement in csPCa detection with fusion biopsy in 4 years. In 2019, panelists from the European Society of Urogenital Radiology (ESUR) and EAU Section of Urologic Imaging (ESUI) released a consensus statement on prostate MRI quality requirements for image acquisition, interpretation, and radiologists’ training. It represents a starting point for certification of individual radiologists and for accreditation of centers for their prostate MRI diagnostic pathway.³⁴ Twenty-four out of 31 of agreement items and 11/16 of other questions reached consensus using the Delphy method. Considering radiologists’ training, authors concluded that theoretical and hands-on courses, followed by supervised education, regular self-, and external performance assessments are essential. The highest agreement was reached on radiologists’ training, specifically on criteria and minimum requirements for basic and expert prostate MRI knowledge. Among others they set as minimum number of cases per year at 150 and 200, for basic and expert, respectively. Also, suggested agreement in double reads with expert centers at 80% and ≥90%, for basic and expert, respectively.³⁵ Recently, Park et al.⁵⁴ published a systematic review and meta-analysis on PI-RADS v. 2 inter-reader agreement. Authors showed a substantial inter-reader agreement with a PI-RADS cutoff of 4 or higher (k = 0.61) and a moderate agreement with a cutoff of 3 or higher (k = 0.57). They found that the difference in reader experience was the single significant factor affecting study heterogeneity (p = 0.01).

In addition, PCa is well known for its heterogeneity, complexity, and variable outcomes. Such features make its management perfectly suited to a multidisciplinary (MTD) approach. In 2016, Pillay et al.⁵⁵ investigated the impact of multidisciplinary team meetings on patient assessment, management, and outcomes on oncologic patients in a systematic review. Twenty-seven studies were included, and they found that 4%–45% of patients discussed at MTD meetings experienced changes in diagnostic reports. Also, that MTD discussed patients are more likely to receive more accurate and complete pre-operative staging and neoadjuvant/adjuvant treatment planning. However, no improved survival outcomes were demonstrated. Similarly, Heidenreich⁵⁶ stated that the benefit of MTD meetings has been shown to lead to major individual treatment changes in 40–60% of the patients.

6.1 Evidence in literature on non-contrast MRI performance

The diagnostic performance of non-contrast MRI has been investigated by many researchers in the last decade. As mentioned above, eliminating contrast-enhanced sequences offers many advantages in terms of operational logistics, costs, time, capacity, and side effects. When looking at the diagnostic accuracy of non-contrast MRI, one of the largest and earliest cohort was presented in 2017 by Kuhl et al.⁵⁹ in which the bpMRI diagnostic accuracy for detection of csPCa (89.1%, 483 of 542), comparable to that of mpMRI. Obmann et al.⁶⁰ in a prospective single-institution study showed that bpMRI detected csPCa with a sensitivity of 95.1% with an NPV of 95.1% and positive predictive value of 53.2%, taking as standard of reference biopsy results. Also, in 2017 Jambor et al.⁶¹ showed the results from the IMPROD trial where bpMRI and targeted biopsy improved risk stratification of patients with prostate cancer suspicion. Recently, Knaapila et al.⁶² externally validated the results of the IMPROD, showing an NPV for csPCa in bpMRI Likert 3–5 and 4–5 score groups of 93% and 92%, respectively, with PPV of 57% and 72%, respectively. Authors defined as optimal the combination of IMPROD bpMRI Likert score 4–5 or Likert 3 with PSAd ≥0.20 ng/ml/cm³, with 35% of the biopsy avoided and 13 csPCa missed.

In 2019, Boesen et al.⁶³ presented the results from the BIDOC study, in which patients with suspicion of PCa underwent non-contrast MRI. All patients underwent systematic biopsy, patients with positive MRI underwent systematic plus targeted biopsy. Authors showed how performing combined biopsies for men with suspicious bpMRI findings, 30% of men could avoid biopsies. Diagnosis of csPCa improved by 11% (396 vs. 351 men; p < 0.001), and diagnosis of ciPCa were reduced by 40% (173 vs. 288 men; p < 0.001). The NPV of bpMRI in ruling out csPCa was 97% (95% CI, 95%–99%). Van der Leest et al.³⁰ in a prospective, multireader, head-to-head study compared the performance of monoplanar "fast" bpMRI and triplanar bpMRI. Sensitivity for high-grade PCa for all protocols was 95% (180/190; 95% confidence interval [CI]: 91–97%). Specificity was 65% (285/436; 95% CI: 61–70%) for "fast" bpMRI and 69% (299/436; 95% CI: 64–73%) for bpMRI. With fast bpMRI, 0.96% (6/626) more low-grade PCa was detected. In a multicenter study, Choi et al.⁶⁴ on a total of 113 patients with prostate cancer who underwent radical prostatectomy showed that the diagnostic performance of pre-biopsy bpMRI is similar to that of mpMRI, with a moderate inter-reader agreement on PI-RADS scoring for both bpMRI (k = 0.540) and mpMRI (k = 0.478). Results from one of the largest head-to-head study investigating the role of bpMRI, Van der Leest et al. ⁶⁵ prospectively proved that sensitivity for high-grade PCa for all protocols was 95% (180/190; 95% CI: 91–97%), specificity was 65% (285/436; 95% CI: 61–70%) for "fast" bpMRI and 69% (299/436; 95% CI: 64–73%) for bpMRI and mpMRI. However, authors acknowledge the poor generalizability of these results in less experienced centers. Recently, Christophe et al.⁶⁶ investigated on the applicability of non-contrast MRI as staging tool for prostate cancer and did not show any reduction in local staging accuracy. Several systematic reviews and meta-analysis have been published on the topic.^67-74 In 2018, Woo et al.⁷¹ selected 20 studies with 2142 patients. The pooled sensitivity and specificity were 0.74 (95% CI, 0.66–0.81) and 0.90 (95% CI, 0.86–0.93) for bpMRI and 0.76 (95% CI, 0.69–0.82) and 0.89 (95% CI, 0.85–0.93) for mpMRI. The MRI protocol (bpMRI vs mpMRI) was not a significant factor in heterogeneity in any subgroup of patients, stratified according to clinic-pathological data and MRI characteristics (p = 0.25–0.97). In 2020, Bass et al.⁶⁹ included 44 studies and with a pooled sensitivity for clinically significant prostate cancer was 0.87 (95% CI, 0.78–0.93), specificity 0.72 (95% CI, 0.56–0.84) and the AUC value was 0.87. No difference was revealed in the pooled diagnostic estimates between bpMRI and mpMRI. Finally, Knaapila et al.⁶⁷ published a pooled data analysis on the NPV of bpMRI based on clinical data from four prospective studies and showed NPVs of bpMRI Likert scores 3–5 and 4–5 for csPCa were 0.932 and 0.909, respectively, and the corresponding PPV were 0.589 and 0.720.

Several investigations have shown comparable diagnostic accuracy of bpMRI with that of mpMRI in the detection of PCa and the identification of csPCa ^{75, 76} (Table 2). The possibility of applying non-contrast MRI for csPCa diagnosis as default approach is the consequence of many investigations evaluating the diagnostic performance of MRI for prostate cancer detection with and without contrast media. However, the detection of more ciPCa implies the need to define patient subgroups where the benefits and harms of contrast enhancement are aligned to their clinical priorities.⁵⁸

7.1 Prostate Cancer Screening using non-contrast MRI

PCa screening is still a controversial topic in the urology community, this is mostly related to the low specificity of PSA value that has been the cause of overdiagnosis of ciPCa for many years, with lack of survival improvement.^{77, 78} Recently, Van Poppel et al.⁷⁹ proposed a novel early detection algorithm for PCa that sees risk stratification for prostate biopsy with multivariable risk prediction models and/or contrast MRI mandatory for an individualized assessment of the potential risk of having a csPCa detection, defining annual PSA testing as redundant. Non-contrast MRI, on the other hand, has become one of the most promising MRI applications as it is a more sensitive test able to perform PCa early detection. Indeed, the Prostate Cancer Prevention Trial (PCPT) showed that the sensitivity for GGG ≥2 was 58%, at a PSA threshold of 3 ng/ml.⁸⁰ In 2016 by R. Nam et al.⁸¹ performed a pilot study to assess the feasibility of MRI to detect PCa during a screening protocol. In this pilot study, PCa was detected with a significantly higher rate with MRI than with PSA level (2.7, 95% CI 1.4–5.4, p = 0.004 vs. 1.1, 95% CI 0.9–1.4, p = 0.21). One of the most significantly results derived from the ERSPC screening study; in particular, the Gothenburg arm^{82, 83} investigated and reported MRI performance at different PSA thresholds. Authors demonstrated that MRI combined with a PSA threshold >1.8 ng/mL reached a sensitivity of 73% and an NPV of 92% for ruling out and ruling in csPCa. The abovementioned studies were pilot study investigating the feasibility of MRI as screening tool for PCa. Today, the first large ongoing trial has been presented by Eldred-Evans et al.,⁸⁴ the IP1-PROSTAGRAM Study. Authors compared the performance of PSA testing, MRI, and ultrasonography as screening tests for PCa and demonstrated that a positive MRI (PI-RADS 4–5) compared with PSA alone >3 ng/mL was associated with more men diagnosed with csPCa, without an increase in the number of men advised to undergo biopsy or overdiagnosis of ciPCa. Also, Panebianco et al.⁸⁵ proposed an emerging innovative method to meet prostate cancer screening needs, using the Network Medicine (NM) approach. Large, randomized controlled trials are needed to further support the role of non-contrast MRI for the early detection of PCa in the screening setting.

7.2 Predictive factors & nomograms

Currently, the use of nomograms and predictive factors is strongly encouraged by EAU guidelines to further improve the complex risk stratification of patients with suspected PCa. The use of such additional tools has been also investigated in the setting of non-contrast MRI. One of the most accepted and validated clinical parameter for PCa detection in adjunct to MRI is the PSA density (PSAd). In EAU guidelines, the suggested threshold for PCa suspicion is 0.2 ng/ml.² Nonetheless, many research groups are still trying to identify the best risk stratification strategy for csPCa early detection that often include clinical parameters, such as PSAd, and bpMRI. Back in 2015, Rais-Bahrami et al.⁸⁶ were among the first groups to demonstrate that using a specific formula [14 × (PSAD) + (the number of SPL) >4.25], the AUC for PCa diagnosis was 0.87 with an overall accuracy of prostate cancer detection of 79%. Falagario et al.⁸⁷ used the Prostate Magnetic Resonance Imaging Outcome Database (PROMOD) to assess the combined use of PSAd and bpMRI for biopsy decision planning. The authors applied ten strategies based on PSAd values and MRI reporting scores (PI-RADS]/Likert/IMPROD biparametric prostate MRI Likert) to assess the best way of avoiding ciPCa diagnosis. They found that the best strategy to avoid biopsy in about 40% of the cases was by using PI-RADS/Likert 4–5 or PI-RADS/Likert 3 if PSAd >0.2, the missed GGG1 PCa were 44% and the csPCa were 10.9%. Recently, several groups are investigating the role of nomograms and risk calculators.^88-91 Reesink et al.⁸⁹ evaluated men with suspicion of PCa (suspicious PSA and/or DRE) using non-contrast MRI and the European Randomized Study of Screening for Prostate Cancer (ERSPC) risk calculator. Authors interestingly found that using bpMRI as a risk stratification tool outperforms risk-calculator in detecting csPCa. Instead, applying the risk-calculator first to decide on whether to perform MRI, avoids 1 out of 2 MRIs. However, applying this strategy would lead to missing up to 1 out of 5 csPCa. In 2019, Boesen et al.,⁹⁰ achieved impressive results developing a predictive model on a cohort of 876 patients, defined by bpMRI scores, age, tumor stage, and PSAd, that showed a high discriminatory power, a good calibration on internal validation, and resulted in a reduction in ciPCa detection of about 50% with only 7% of significant cancers missed. At the same time, Perez et al.⁹¹ aimed at developing and validating bpMRI-based nomograms using qualitative and quantitative derived variables csPCa prediction. They retrospectively analyzed data from the IMPROD and MULTI-IMPROD trial and showed that on a multivariate analysis the model with the highest performance was the MRI model (PSA, age, use of 5-alpha-reductase inhibitors, and IMPROD bpMRI Likert score). Notably, Li et al.⁹² developed a radiomic-clinicopathologic nomogram (RadClip) for post-surgical biochemical recurrence-free survival (bRFS) and adverse pathology (AP) prediction in men with prostate cancer (PCa), from pre-operative bpMRI. Authors demonstrated that the RadClip performed better than existing nomograms in terms of bRFS and AP.

7.3 Radiomics & Artificial intelligence

The rapid technological innovation in radiology has led to continuous advances causing a strong transformation in the specialty. Soon, the highest impact on medical imaging will likely derive from artificial intelligence (AI) technology developments. Indeed, in the age of the Big Data, AI, and its application, and radiomics have been the object of many investigations, including the diagnosis of PCa with non-contrast MRI (Table 3). In 2018, Niu et al.⁹³ investigated the performance of non-contrast MRI texture analysis (TA) in detecting high-grade prostate cancer in zone-specific region. The study included 184 biopsy-naïve patients who underwent contrast MRI. Two readers scored the images with contrast using the PI-RADS system and biparametric MRI TA. Authors showed that the diagnostic performance of the TA-based model was improved (transition zone AUC, 0.87; peripheral zone AUC, 0.89) in comparison with PI-RADSv2. Similarly, Xu et al.⁹⁴ applied radiomics signature to similar non-contrast MRI findings for discriminating benign from malignant lesions. The MRI signature, based on six selected radiomics features, showed better discrimination (AUC, 0.92) than on each subcategory images (AUC, 0.81 on T2WI; AUC, 0.77 on DWI; AUC, 0.89 on ADC), which improved when applying it to clinical factors (AUC, 0.93). Recently, Baek et al.⁹⁵ evaluated the association of texture features extracted from non-contrast MRI with Gleason score for differentiating ciPCa from csPCa. The gray-level co-occurrence matrix (GLCM) entropy on ADC map allowed to discriminate csPCa with an accuracy of 82%. Also, Algohary et al.⁹⁶ evaluated the role of peri-tumoral radiomic features on non-contrast MRI to distinguish PCa risk categories according to the D'Amico Risk Classification System. Authors suggested that peri-tumoral radiomic features were independent predictive value to intra-tumoral radiomic features for PCa risk assessment, improving it by 3–6%. Finally, Gong et al.⁹⁷ assessed the role of radiomic features derived from non-contrast MRI to retrospectively evaluate the use of different sequences for noninvasively distinguishing high-grade PCa. Authors found that all radiomic models showed significant (p < 0.001) predictive performances and that the DWI model achieved the most encouraging results. Bonekamp et al.⁹⁸ prospectively compared non-contrast MRI free radiomic machine learning (RML), mean apparent diffusion coefficient (ADC), and radiologist assessment on a cohort of 316 patients. The quantitative measurement of the mean ADC values improved detection of malignant lesions, compared with clinical assessment, but no significant differences were detected compared to the radiomic machine learning. Also, Jensen et al.⁹⁹ evaluated the role of a k-nearest neighbor classifier to assess the grade grouping of PCa using features extracted from T2WI and DWI. Authors showed that classifying PCa lesions into grade groups (GG) resulted in AUC of 0.87, 0.88, 0.96, 0.98, and 0.91 for GG1, GG2, GG1 + 2, GG3, and GG4 + 5 for the peripheral zone, respectively. Recently, Vente et al.¹⁰⁰ presented a neural network for the diagnosis and grading of PCa, using ProstateX-2 challenge dataset. The proposed model outperformed standard multiclass classification and multi-label ordinal regression. Winkel et al.¹⁰¹ evaluated the performance of bpMRI on a small cohort of patients (49) using a deep learning method in a screening setting. The DL method yielded a sensitivity of 87% and a specificity of 50%, suggesting that the software solution was able to identify highly suspicious lesions and has the potential to effectively guide biopsy.

The application of radiomics features and/or AI systems would likely aid in PCa diagnostics. Indeed, AI may meet multiple PCa diagnostic open issues. First, the image quality and inter-reader reproducibility issue. Second, MRI reading optimization. AI might work as second reader for less experienced radiologists, evaluate and assess quantitative MRI metrics and reduce variability of predictive values. Third, AI might have a role in determining whether to perform contrast MRI and support the MRI role for opportunistic screening as triage test.⁶³ Fourth, AI might adapt MRI performance to meet patient clinical priorities and to manage multiple imaging time point. Fifth, AI might support radiologists in follow-up MR imaging for disease monitoring.