Quantifying the Heterogeneity of TRUS-Visible Lesions for Predicting Prostate Cancer: A Multicenter Preliminary Study

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Artificial Intelligence in Medicine

Background:

• Transrectal ultrasound (TRUS)-visible lesions in the prostate present diagnostic challenges.
• Quantifying intratumoral heterogeneity (ITH) is crucial for accurate cancer diagnosis.

Purpose of the Study:

• To develop and validate an ITH model for differentiating prostate cancer (PCa) from benign prostate lesions.
• To quantitatively characterize lesion heterogeneity using subregional imaging analysis.

Main Methods:

• Retrospective study involving TRUS-guided biopsy data from three medical centers.
• K-means clustering for lesion subregion partitioning and radiomic feature extraction.
• Development of whole-lesion and subregional predictive models, integrating subregional models into an ITH model.

Main Results:

• The developed ITH model achieved an AUC of 0.879 (internal) and 0.818 (external validation).
• The ITH model outperformed whole-lesion (Intratumor) and clinical models in diagnostic performance.
• The ITH model demonstrated superior performance compared to junior radiologists and enhanced their diagnostic accuracy.

Conclusions:

• An ITH model was successfully developed to quantify heterogeneity in TRUS-visible lesions.
• The ITH model shows significant potential for improving the diagnostic accuracy of prostate cancer detection.