Use of Radiomics to Improve Diagnostic Performance of PI-RADS v2.1 in Prostate Cancer

Abstract

ObjectiveTo investigate whether a radiomics model can help to improve the performance of PI-RADS v2.1 in prostate cancer (PCa).MethodsThis was a retrospective analysis of 203 patients with pathologically confirmed PCa or non-PCa between March 2015 and December 2016. Patients were divided into a training set (n = 141) and a validation set (n = 62). The radiomics model (Rad-score) was developed based on multi-parametric MRI including T2 weighted imaging (T2WI), diffusion weighted imaging (DWI), apparent diffusion coefficient (ADC) imaging, and dynamic contrast enhanced (DCE) imaging. The combined model involving Rad-score and PI-RADS was compared with PI-RADS for the diagnosis of PCa by using the receiver operating characteristic curve (ROC) analysis.ResultsA total of 112 (55.2%) patients had PCa, and 91 (44.8%) patients had benign lesions. For PCa versus non-PCa, the Rad-score had a significantly higher area under the ROC curve (AUC) [0.979 (95% CI, 0.940–0.996)] than PI-RADS [0.905 (0.844–0.948), P = 0.002] in the training set. However, the AUC between them was insignificant in the validation set [0.861 (0.749–0.936) vs. 0.845 (0.731–0.924), P = 0.825]. When Rad-score was added to PI-RADS, the performance of the PI-RADS was significantly improved for the PCa diagnosis (AUC = 0.989, P < 0.001 for the training set and AUC = 0.931, P = 0.038 for the validation set).ConclusionsThe radiomics based on multi-parametric MRI can help to improve the diagnostic performance of PI-RADS v2.1 in PCa.

Artificial Intelligence Compared to Radiologists for the Initial Diagnosis of Prostate Cancer on Magnetic Resonance Imaging: A Systematic Review and Recommendations for Future Studies

“…The 27 studies, and by extension, the CAD systems presented or evaluated within them, were categorized as either ROI Classification (ROI-C), Lesion Localization and Classification (LL&C), or Patient Classification (PAT-C); the categories are shown diagrammatically in Figure 2. ROI-C refers to (n = 16) studies where CAD systems classified pre-defined regions of interest (ROI), e.g., manually contoured lesions [19][20][21][22][23][24][25][26][27][28][29][30][31][32]44,45], LL&C refers to (n = 10) studies where CAD systems performed simultaneous lesion localization and classification [33][34][35][36][37][38][39][40][41][42], and PAT-C refers to (n = 1) studies where CAD systems classified patients directly [43].…”

“…All 27 included studies used a retrospective study design. The median size of patient cohorts used for evaluation was 98 (range 30 to 417, n = 26) for studies where the size of the evaluation patient cohort was reported [19][20][21][22][23][24][25][26][27][28][29][31][32][33][34][35][36][37][38][39][40][41][42][43][44][45]. Most studies (n = 18) considered clinically suspected patient cohorts [20][21][22][23]27,31,[34][35][36][37][38][39][40][41][42][43][44][45], while fewer studies (n = 9) considered patient cohorts with biopsy-proven prostate cancer [19,[24][25][26]…”

“…Histopathological reference standards used in studies were one or a combination of the following: transperineal template prostate-mapping (TTPM) biopsy, in-bore targeted biopsy, TRUS targeted biopsy, TRUS saturation biopsy, TRUS systematic biopsy, or radical prostatectomy. The majority of studies (n = 22) collected scans using 3T MR scanners [19][20][21][23][24][25][29][30][31][32][33][34][36][37][38][39][40][41][42][43][44][45], while fewer studies (n = 4) used 1.5T MR scanners [22,26,27,35]; one further study used 3T mainly but included one scan acquired at 1.5T [28]. Evaluation using multicenter data was completed in three studies [19,34,38], and only three studies used multivendor MRI data for evaluation [19,23,34].…”

“…In the majority of studies (n = 26), readers scored suspicious lesions using mpMRI , while in one study, biparametric MRI (bpMRI) was used [19]. In 15 studies, patient cases were reported by a single reader [19][20][21][22][23]28,32,33,[37][38][39][40][42][43][44], while in the remaining 12 studies, patient cases were reported by more than one reader [24][25][26][27][29][30][31][34][35][36]41,45]; where multiple readers reported on each patient, the presented reader performance is either an average of reader performance or based on the consensus view of readers. Reader experience varied significantly across studies; in studies where reader performance was stratified by experience level, details of the most experienced reader group were extracted.…”

“…Reader experience varied significantly across studies; in studies where reader performance was stratified by experience level, details of the most experienced reader group were extracted. Considerable heterogeneity was observed in the machine learning algorithms employed by the CAD systems: five studies presented convolutional neural networks (CNN) [32,33,39,40,43], one study evaluated the commercially available Watson Elementary TM system [41], while the remaining 21 studies presented or evaluated CAD systems based on traditional machine learning algorithms [19][20][21][22]24,25,[35][36][37][38]41,45]. Across the studies, a variety of methods were used to construct datasets for training and evaluation.…”