Multiparametric MRI‐Based Radiomics for Prostate Cancer Screening With PSA in 4–10 ng/mL to Reduce Unnecessary Biopsies

Abstract

BackgroundWhether men with a prostate‐specific antigen (PSA) level of 4–10 ng/mL should be recommended for a biopsy is clinically challenging.PurposeTo develop and validate a radiomics model based on multiparametric MRI (mp‐MRI) in patients with PSA levels of 4–10 ng/mL to predict prostate cancer (PCa) preoperatively and reduce unnecessary biopsies.Study TypeRetrospective.SubjectsIn all, 199 patients with PSA levels of 4–10 ng/mL.Field Strength/Sequence3T, T2‐weighted, diffusion‐weighted, and dynamic contrast‐enhanced MRI.AssessmentLesion regions of interest (ROIs) from T2‐weighted, diffusion‐weighted, and dynamic contrast‐enhanced MRI were annotated by two radiologists. A total of 2104 radiomic features were extracted from the ROI of each patient. A random forest classifier was used to build the radiomics model for PCa in the primary cohort. A combined model was constructed using multivariate logistic regression by incorporating the radiomics signature and clinical‐radiological risk factors.Statistical TestsFor continuous variables, variance equality was assessed by Levene's test and Student's t‐test, and Welch's t‐test was used to assess between‐group differences. For categorical variables, Pearson's chi‐square test, Fisher's exact test, or the approximate chi‐square test was used to assess between‐group differences. P < 0.05 was considered statistically significant.ResultsThe combined model incorporating the multi‐imaging fusion model, age, PSA density (PSAD), and the PI‐RADS v2 score yielded area under the curve (AUC) values of 0.956 and 0.933 on the primary (n = 133) and validation (n = 66) cohorts, respectively. Compared with the clinical‐radiological model, the combined model performed better on both the primary and validation cohorts (P < 0.05). Furthermore, the use of the combined model to predict PCa could identify more negative PCa patients than the use of the clinical‐radiological model by 18.4%.Data ConclusionThe combined model was developed and validated to provide potential preoperative prediction of PCa in men with PSA levels of 4–10 ng/mL and might aid in treatment decision‐making and reduce unnecessary biopsies.Level of Evidence: 3Technical Efficacy Stage: 3J. Magn. Reson. Imaging 2020;51:1890–1899.

Radiomics Based on Multiparametric Magnetic Resonance Imaging to Predict Extraprostatic Extension of Prostate Cancer

“…Radiomics is defined as high-throughput extraction of mineable, quantitative, and highdimensional medical imaging features using machine learning (13,14). Recently, the emerging radiomics technique has been widely applied in PCa research, which was reported to have added value in PCa detection, aggressiveness assessment, and survival analysis (15)(16)(17)(18)(19)(20). One of the explanations of radiomics' superiority maybe that radiomics could provide more information about the lesion which might be correlated with the intratumor heterogeneity (13).…”

Radiomics in Oncology, Part 2: Thoracic, Genito-Urinary, Breast, Neurological, Hematologic and Musculoskeletal Applications

“…Similarly, Wibmer A. et al [ 36 ] showed promising results of Radiomics in PCa detection and Gleason Score (GS) assessment on mpMRI, in particular Gleason Score (GS) was significantly associated with higher values of Entropy (GS 6 vs. 7: p = 0.0225; 6 vs. >7: p = 0.0069) and lower values of energy (GS 6 vs. 7: p = 0.0116, 6 vs. >7: p = 0.0039) derived from ADC maps. Furthermore, Qi Y. et al [ 37 ] created a radiomic model by using a random forest classifier, based on 2104 features extracted from MRI sequences. The combined model (radiological and clinical data) returned AUC values of 0.956 and 0.933 on the test ( n = 133) and validation ( n = 66) population, respectively, making it an additional potential tool for the clinicians in treatment decision-making.…”

Radiomic and Genomic Machine Learning Method Performance for Prostate Cancer Diagnosis: Systematic Literature Review

“…Those studies [44][45][46][47][48][49][50][51][52][53][54] employed multiple lesions for each patient enrolled in the study. Subgroup 2 gathered those studies [55][56][57][58][59] that enrolled two distinct groups (PCa and controls) and employed a patient-based ML approach. The heterogeneity in the subgroups was greater than 70% (subgroup 1: P<.001, subgroup 2: P=.002).…”

“…High heterogeneity may be due to the different cross-validation techniques used (eg, bootstrapping [16,40,52], Monte Carlo cross-validation [17]) or the choice of number of folders used in cross-validation methods; if an external data set was used [52,60,61,63], differences in the study protocols may have increased the bias among studies. Moreover, few studies in radiomic [50,53,57,59] and genomic [17,67] analysis employed both cross-validation and external testing. Studies employing no validation showed very low heterogeneity (only 2 studies in radiomic analysis), which may be due to the absence of other confounding variables, and high performances may be due to overfitting problems.…”

“…Accuracy measures of radiomic studies for the systematic review. ANN: artificial neural networks; LDA: linear discriminant analysis; RF: random forest; SNCSAE: stacked nonnegativity constraint sparse autoencoders; LR: logistic regression; DCNN: deep convolutional neural networks; LASSO: least absolute shrinkage and selection operator; DNN: deep neural networks; SVM: support vector machine; CART: classification and regression tree.Figure3shows the subgroup analysis among studies that employed internal cross-validation techniques (subgroup 1)[46][47][48][49]51,54,55,59], split validation approaches (subgroup 2)[44,50,52,53,56,57], and no validation (subgroup 3)[45,58]. The heterogeneity for subgroups 1 and 2 was around 80% (P<.001).…”