RPS 1307 - Prostate MRI for differential diagnosis

Purpose:

Urinary incontinence is a major post-surgical complication that diminishes prostate cancer patients' quality of life. This study's purpose was to explore the feasibility of diffusion tensor imaging to access age and disease-related differences in the male urethral complex.

Methods and materials:

95 consecutive patients with a clinical suspicion of prostate cancer underwent a multi-parametric MRI examination including a diffusion tensor imaging (DTI) sequence with 16 directions and a b-value=600 s/mm2. The proximal and distal sphincters and the membranous urethra were reconstructed using a deterministic tractography algorithm. DTI metrics, including tract length and density, fractional anisotropy (FA), axial diffusivity (AD), mean diffusivity (MD), radial diffusivity (RD), and 14 additional histogram metrics, were analysed. Pearson or Spearman correlations and linear regressions were performed between age and each DTI metric. Differences in DTI metrics between age and disease groups (healthy controls, benign prostatic hyperplasia, and prostate cancer patients) were accessed by analysis of variance (ANOVA) or the non-parametric Kruskal-Wallis test. Post-hoc t-tests were used to identify statistically significant differences between groups.

Results:

Statistically significant changes with age were found in MD and RD distributions. Particularly, older subjects evidenced lower RD kurtosis in the proximal sphincter, lower MD maximum in the distal sphincter, and higher MD median absolute deviation in the membranous urethra. No statistically significant differences were found in DTI metrics between disease groups.

Conclusion:

Age should be included as a covariate in the model that uses sphincters’ DTI biomarkers to predict the post-surgical urinary continence recovery time.

Limitations:

The minimum patients’ age under study and the patients’ heterogeneity within each disease group.

Ethics committee approval

The Champalimaud Foundation EC approved the DTI acquisitions, additional to the mpMRI.

Funding:

No funding was received for this work.

Purpose:

To investigate the potential role of radiomic features computed on high b-value diffusion-weighted imaging (DWI) to perform risk stratification of patients with a clinical suspicion of prostate cancer (PCa).

Methods and materials:

42 patients of our institution, representing 7 risk levels, were retrospectively enrolled in the study and grouped into 4 classes of risk: (a) clinically significant (CS) PCa split over 4 levels (ISUP=2÷5), (b) non-clinically significant (NCS) PCa (ISUP=1) patients with a negative biopsy, and (c) positive mpMRI (NP) or (d) negative mpMRI (NN). After computing radiomic features on DWI b=2000s/mm2, the correlation between radiomic features and risk level was investigated through two steps, (i) Spearman index (ρ) and (ii) Kruskal-Wallis and Wilcoxon tests (p<0.05), for multi- and pairwise- comparison of the 4 classes, respectively.

Results:

The mean of the local coefficient of variation (CVL-m), a measure of local dispersion of DWI values, resulted in the most discriminant radiomic features among the four classes (p~10-6), able to rank the four increasing risk classes with ρ=0.81, with a high pairwise separability (p≤0.026). ρ=0.81 was also achieved when correlating the CVL-m with all 7 increasing risk level groups.

Conclusion:

This study allows performing an early stratification of all 7 PCa risk levels. Increasing values of CVL-m in DWI images describe a higher degree of local heterogeneity in accordance with tissue over-proliferation and, consequently, an increasing level of tumour aggressiveness.

Limitations:

The number of patients could be low for a proper stratification of the cohort in 7 classes. However, the excellent results achieved when using CVL-m values to correctly rank all risk levels give CVL-m the most promising role in depicting PCa risk progression.

Ethics committee approval

IRB approval and written informed consent was waived.

Funding:

No funding was received for this work.

Purpose:

To compare multiparametric magnetic resonance imaging (mpMRI)-derived abbreviated protocols in detecting prostate cancer (PCa).

Methods and materials:

We retrospectively included 108 patients (mean age 64.8 years) who performed mpMRI to stage biopsy-proven clinically-localised PCa before radical prostatectomy between January 2016-May 2019. Examinations were performed on a 3.0T magnet with a prostate imaging reporting and data system version 2 (PI-RADSv2)-compliant protocol. Two readers (R1 and R2, with an experience of <600 and <250 examinations) blinded to definitive pathology, independently analysed mpMRI examinations, attributing a PI-RADSv2 score to each observation on a per-sequence basis. A study coordinator assessed the final PI-RADSv2 category of each observation by combining readers’ categorisations according to the following combinations: fast-MRI (diffusion-weighted imaging+T2-weighted imaging), fast-MRI+dynamic contrast-enhanced imaging (DCE), biparametric MRI (bpMRI), and mpMRI. Based on pathology results, we calculated the cancer detection rate (CDR) and false-discovery rate (FDR) (false-positives/true-positives + false-positives) of each protocol. The analysis was performed on the index lesion, using category ≥3 as the cut-off.

Results:

108 index lesions predominated in the peripheral zone, showing a median Gleason score 3+4 and pT stage ≤2c in 74% of cases. Fast-MRI, fast-MRI+DCE, and bpMRI equalled mpMRI, showing 78.7% (95%CI 69.6-85.8) and 77.8% (95%CI 68-6-85.0) CDR, and 10.5% (95%CI 5.0-19.3) and 9.6% (95%CI 4.4-18.3) FDR for R1 and R2, respectively. Fast-MRI+DCE upgraded 53.8% and 90% of category 3 assignments achieved with fast-MRI by R1 and R2, respectively. Upgraded cases were true-positives in 95% of cases.

Conclusion:

In a medium-to-low experience setting, fast-MRI+DCE was the abbreviated protocol better balancing CDR and minimising PI-RADS 3 assignments.

Limitations:

A monocentric and retrospective study.

Ethics committee approval

Informed consent acquisition was waived due to the retrospective design.

Funding:

No funding was received for this work.

Purpose:

To investigate the diagnostic performance for index prostate cancer (iPCa) of a double-reading strategy with an abbreviated magnetic resonance imaging (aMRI) protocol in patients candidate to active surveillance.

Methods and materials:

Between February 2015 and November 2018, we retrospectively included 54 men with biopsy-proven low-risk prostate cancer candidate to active surveillance who underwent 3.0T multiparametric magnetic resonance imaging (mpMRI) before a confirmatory biopsy. Three radiologists (R1, R2, and R3) with average-to-low experience in mpMRI, blinded to clinical history and biopsy results, reviewed examinations under the form of aMRI (transverse T2-weighted imaging and diffusion-weighted imaging). Prostate imaging reporting and data system version 2.1 (PI-RADSv2.1) was used for categorising aMRI observations. Using category ≥4 as the cut-off and a confirmatory biopsy as a reference standard, we assessed the sensitivity and specificity for iPCa of single readings (R1, R2, and R3), as well as the readings resulting from the combination of R1+R2, R2+R3, and R1+R3. In double-reading, a case was assessed positive if suspicious for at least one reader.

Results:

Saturation targeted confirmatory biopsy was performed with the transperineal approach. At biopsy, iPCas included 42/54 (78%) low-risk lesions. The remaining 12/54 (22%) patients were reclassified to the intermediate-to-high-risk category. Sensitivity/specificity for R1, R2, and R3 were 67% (95%CI 35-90)/57% (95%CI 41-72), 83% (95%CI 52-98)/60% (95%CI 43-74), and 42% (95%CI 15-72)/79% (95%CI 63-90), respectively. Double-reading provided a sensitivity/specificity of 83% (95%CI 52-98)/50% (95%CI 34-66) for R1+R2, 83% (95%CI 52-98)/57% (95%CI 41-72) for R2+R3, and 75% (95%CI 43-95)/55% (95%CI 39-70) for R1+R3, respectively.

Conclusion:

A double-reading approach using an abbreviated magnetic resonance imaging (aMRI) protocol improved the sensitivity for iPCa in patients candidate to active surveillance.

Limitations:

A retrospective design.

Ethics committee approval

IRB-approved.

Funding:

No funding was received for this work.

Purpose:

Radiomic features extracted from routine MRI sequences are increasingly used for in-vivo tissue characterisation. The aim of this study was to develop a computer-aided diagnosis (CADx) algorithm to predict the Gleason score (GS) of prostate cancers (PCa) in multiparametric MRIs of the prostate (mpMRI).

Methods and materials:

83 PCa reported in mpMRI were included in this retrospective study. Histopathology of TRUS-guided biopsy cores with reported GS served as a standard of reference (SOR). mpMRIs were performed following international guidelines and current practice. Imaging parameters included lesion size, T2w-signal intensity, diffusion restriction, prostate volume, and dynamic parameters (wash-in, wash-out, peak enhancement intensity, the initial area under the curve, and time-to-peak) along with zonal anatomy (peripheral zone vs central gland), patient age, and serum PSA level. Predicting the GS was defined as a regression problem. Using the above-mentioned parameters, an extreme gradient boosting algorithm was trained to predict the lesions’ GS. A leave-one-out cross-validation was applied to ensure generalisability. Performance optimisation was focused on minimising the root-mean-square-error (RMSE). Inter-reader agreement of the imaging parameters was assessed using intraclass correlation coefficients (ICC).

Results:

Histopathological assessment revealed n=17 Gleason-6 PCa, n=45 Gleason-7 PCa, n=8 Gleason-8 PCa, and n=13 Gleason-9 PCa. The CADx reached an RMSE of 0.653. Predicted and observed GS correlated significantly (P<0.0001, r=0.503). All imaging parameters featured excellent inter-reader agreement (all ICC≥0.87).

Conclusion:

This study describes a CADx for GS prediction of PCa in mpMRI with high accuracy and excellent inter-reader agreement.

Limitations:

Limitations include the retrospective and monocentric study design.

Ethics committee approval

The ethics committee approved this study, with a waiver of informed consent.

Funding:

This work was supported by the German Research Foundation (DFG, CRC 1181-Z02).

Purpose:

To investigate whether DWI-based radiomics features could differentiate patients with a clinical suspicion of PCa and negative TRUS-biopsy that have a positive mpMRI from patients where mpMRIs do not show any evidence.

Methods and materials:

The records of 17 patients undergoing 3T-mpMRI for suspected PCa subsequently not confirmed at TRUS-biopsy were extracted from our institutional database. The ground truth was available for only a few. 7 patients did not report evidence at mpMRI, while 10 patients showed suspected PCa lesions, contoured in consensus by two radiologists. 84 image-based radiomic features were computed on high b-value DW-MRI sequences of all patients of the two groups. The ROC curve was computed for each feature and the one yielding the highest AUC was selected. Its discrimination power was also assessed via a Wilcoxon rank-sum test (p<0.001).

Results:

The mean of local skewness (SL-m), related to local inhomogeneities of DWI values, confirms radiologist reports in 94% of cases, with AUC=0.93 (95% CI, 0.56-1.00), specificity=100%, and sensitivity=86% (one false-positive only). Median SL-m values in patients with suspected PCa were greater than 30% (p~10-4) with respect to patients showing no evidence at mpMRI.

Conclusion:

DWI-based radiomic features strongly support mpMRI evidence in cases of suspected, and for some patients clear, PCa although TRUS-biopsy is negative. These outcomes suggest further investigation on the role that these extremely promising features could have in PCa patient’s stratification.

Limitations:

Although it confirmed the mpMRI evidence to be PCa for the few patients where the ground-truth was available, for most of them it was not at our disposal because patients did not belong to a dedicated study.

Ethics committee approval

IRB approval, written informed consent was waived.

Funding:

No funding was received for this work.