Answers are grounded in this paper and cite their sources.
Introduction: Transrectal ultrasound (TRUS) radiomics has shown potential for prostate cancer detection, but previous studies have primarily extracted features from lesions or broad anatomical regions. The diagnostic value of ultrasound features surrounding individual biopsy needle trajectories remains unclear. Objective: To determine whether radiomic features extracted from regions surrounding tracked needle trajectories in three-dimensional (3D) TRUS improve core-level discrimination of clinically significant prostate cancer (csPCa) beyond clinical and anatomical location information. Methods: We retrospectively analyzed the public Prostate-MRI-US-Biopsy cohort, including 1,129 patients with 17,559 cores. csPCa was defined as a Gleason score of ≥ 7. Cylindrical regions (5-mm radius, with 3-mm and 7-mm sensitivity analyses) were centered on trajectory paths, and 93 PyRadiomics features were extracted. Three L1-penalized logistic regression models (clinical/location baseline, trajectory-radiomics, and combined) were evaluated using nested patient-grouped cross-validation and patient-cluster bootstrap resampling. Results: After quality control, 16,259 cores (1,561 csPCa) were analyzed. No radiomic feature was selected in any outer fold of the combined model. Areas under the receiver operating characteristic curve (AUCs) for baseline, trajectory-radiomics, and combined models were 0.812, 0.543, and 0.819, respectively. The combined-minus-baseline AUC difference was 0.0065 (p < 0.001), attributable to fold-specific regularization rather than radiomic information, and the combined model had a higher (worse) Brier score than the baseline model (0.196 vs. 0.189). Conclusion: Tracked coordinates enabled region construction, but conventional trajectory radiomics provided no attributable or clinically meaningful incremental value over clinical/location data. Independent validation is required before clinical application.
No takes yet. Share an insight, caveat, or question.
Zheng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: