Computed Tomography-based Radiomics for Risk Stratification in Prostate Cancer

Area of Science:

• Radiology
• Medical Imaging
• Oncology

Background:

• Prostate cancer risk stratification is crucial for treatment decisions.
• Accurate noninvasive methods are needed to assess tumor aggressiveness.
• Computed tomography (CT) is a common imaging modality in radiation oncology.

Purpose of the Study:

• To investigate the utility of CT-based radiomics features for prostate cancer risk stratification.
• To develop and evaluate machine learning models for classifying Gleason score (GS) and risk group (RG).

Main Methods:

• Radiomics features were extracted from planning CT scans of 342 prostate cancer patients.
• Machine learning models were trained for GS and RG classification using repeated cross-validation.
• Model performance was assessed using receiver operating characteristic area under the curve (AUC).

Main Results:

• CT-based radiomics classifiers achieved high AUCs for distinguishing GS ≤ 6 vs. GS ≥ 7 (0.90) and GS 7(3+4) vs. GS 7(4+3) (0.98).
• Excellent performance was observed for low vs. high RG (AUC=0.96) and low vs. intermediate RG (AUC=1.00) on training data.
• Modest performance was noted on holdout data, with the highest AUC of 0.75 for low vs. high RG classification.

Conclusions:

• CT-based radiomics offers a noninvasive approach to assess prostate cancer aggressiveness.
• Models demonstrated high accuracy in differentiating risk groups and Gleason scores on training data.
• External validation and prospective studies are recommended to confirm findings and guide targeted radiation therapy.