Machine learning-based multi-class classification of bladder pathologies using fused 3D CT radiomic and 3D...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Accurate multi-class classification of bladder pathologies from non-contrast CT is crucial for timely diagnosis and treatment.
• Current methods may lack the precision and automation required for efficient clinical integration.

Purpose of the Study:

• To develop an automated analytical framework integrating hybrid radiomics and deep learning features from non-contrast CT images.
• To achieve multi-class classification of bladder pathologies, including normal, calculi, cancer, and cystitis.

Main Methods:

• A retrospective analysis of 902 CT scans was performed.
• An integrated pipeline involved automatic bladder segmentation (3D-UNet), hybrid feature extraction (radiomics + deep learning), feature selection (LASSO), and classification (XGBoost).
• Performance was evaluated using AUROC with a one-vs-rest strategy and cross-validation.

Main Results:

• The framework achieved high one-vs-rest AUROCs: 0.94 for calculi, 0.92 for cancer, 0.90 for normal bladder, and 0.83 for cystitis.
• The micro-average AUROC for four-class discrimination was 0.94.
• Radiomic features were key for calculi/normal differentiation, while deep features were critical for cancer/cystitis classification.

Conclusions:

• The hybrid CT analysis framework demonstrates clinically relevant performance in automated multi-class bladder pathology classification.
• The complementary roles of radiomic and deep features offer an interpretable diagnostic aid.
• This framework shows potential for integration into clinical workflows to support differential diagnosis.