Beyond Radiomics Alone: Enhancing Prostate Cancer Classification with ADC Ratio in a Multicenter Benchmarking Study.

Area of Science:

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

Background:

• Radiomics offers non-invasive quantitative imaging features for prostate cancer (PCa) classification.
• Accurate detection of clinically significant PCa (csPCa) is vital for treatment decisions.
• Existing studies often lack comprehensive benchmarking and external validation of radiomics models.

Purpose of the Study:

• To systematically benchmark radiomics modeling pipelines for csPCa detection.
• To evaluate the added value of combining radiomics with the lesion-to-normal apparent diffusion coefficient (ADC) ratio.
• To assess the generalizability of models across multicenter datasets.

Main Methods:

• Extracted radiomic features from ADC maps using IBSI-compliant pipelines.
• Tested over 100 model configurations combining feature selection, classifiers, and harmonization.
• Evaluated models using cross-validation on a multicenter dataset and external testing on the PROSTATEx dataset.
• Defined ADC ratio as mean lesion ADC divided by contralateral normal tissue ADC for normalization.

Main Results:

• The best model in cross-validation combined radiomics, ADC ratio, LASSO, and Naïve Bayes (AUC-PR = 0.844).
• The top-performing model for external validation used Recursive Feature Elimination (RFE) and Boosted GLM (AUC-PR = 0.722, F1 = 0.741).
• Texture-based features from filtered ADC maps were frequently selected; ComBat harmonization improved calibration but not discrimination.

Conclusions:

• Integrating radiomics with the ADC ratio significantly improves csPCa classification performance.
• This combined approach enhances model generalizability, crucial for multicenter studies.
• The findings support the potential of radiomics and ADC ratio as robust, interpretable imaging biomarkers for PCa.