Improving the Potential for Predicting Prostate Cancer Progression in Patients on Active Surveillance Using...

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Half of prostate cancer (PCa) patients have low- or intermediate-risk disease suitable for active surveillance (AS).
• A significant number of patients on AS experience pathological progression.
• Predicting this progression is crucial for effective PCa management.

Purpose of the Study:

• Develop and validate predictive models for histopathological PCa progression in patients undergoing AS.
• Integrate radiomic features from MRI and clinical data for enhanced prediction.
• Improve clinical decision-making for AS patients.

Main Methods:

• Utilized a dataset of biopsy-confirmed PCa patients with at least two years of follow-up and multiple MRI scans.
• Defined histopathological progression as a grade group upgrade on repeat biopsy.
• Employed machine learning (ML) models integrating radiomic and PSA density (PSAd) variables.
• Applied SHapley Additive exPlanations (SHAP) for feature interpretability.

Main Results:

• A baseline model (radiomic + PSAd) achieved an AUC of 0.793.
• A delta model (feature changes + PSAd) improved prediction with an AUC of 0.913.
• A time series model (all scans + PSAd) showed the highest performance with an AUC of 0.917.
• All models demonstrated strong ability to differentiate progressors from non-progressors.

Conclusions:

• Radiomic analysis combined with ML offers a powerful tool for predicting PCa progression in AS patients.
• These models can significantly enhance PCa management and personalized treatment strategies.
• The findings support the integration of advanced imaging and ML in routine clinical practice for AS.