Noninvasive Prediction of Perineural Invasion and Lymphovascular Invasion in Prostate Cancer Using bpMRI Radiomic...

Area of Science:

• Radiology and Oncology
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Perineural invasion (PNI) and lymphovascular invasion (LVI) are critical prognostic indicators in prostate cancer (PCa).
• Current diagnosis relies on postoperative histopathology, delaying treatment decisions.
• There is a need for noninvasive preoperative methods to predict PNI and LVI.

Purpose of the Study:

• To develop and validate a noninvasive radiomic model using biparametric MRI (bpMRI) for preoperative prediction of PNI and LVI in PCa.
• To assess the generalizability of the radiomic model across independent cohorts.

Main Methods:

• Retrospective analysis of 256 PCa patients with pathologically confirmed disease.
• Development of radiomic models using features extracted from T2-weighted and diffusion-weighted MRI sequences.
• Feature selection via Spearman's correlation and LASSO algorithm, followed by machine learning classifier construction (MLP, Logistic Regression).
• External validation using an independent cohort from a second center.

Main Results:

• The Multilayer Perceptron (MLP) model achieved an AUC of 0.805 (training) and 0.795 (testing) for PNI prediction.
• Logistic Regression demonstrated superior performance for LVI prediction with AUCs of 0.859 (training) and 0.810 (testing).
• Models showed good calibration and clinical utility, indicating reliable predictive performance.

Conclusions:

• bpMRI-derived radiomic models can accurately and noninvasively predict PNI and LVI in PCa.
• These models show good generalizability, offering a promising tool for preoperative risk stratification.
• The findings suggest a potential shift from reliance on postoperative histopathology for PNI and LVI assessment.