Bladder Urothelial Carcinoma: Machine Learning-based Computed Tomography Radiomics for Prediction of Histological...

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Histological variant (HV) of bladder urothelial carcinoma (UC) significantly impacts therapy management.
• Accurate identification of HV is crucial for optimal patient treatment strategies.

Purpose of the Study:

• To assess the predictive performance of machine learning (ML)-based Computed Tomography (CT) radiomics for identifying HV in UC.
• To develop and evaluate ML models for predicting HV, a factor critical for treatment planning.

Main Methods:

• Manual segmentation of 37 bladder UC tumors (21 pure, 16 HV) and extraction of 117 radiomic features.
• Application of 15 ML algorithms using Python and PyCaret, with data augmentation and outlier removal.
• Ensemble modeling using Voting Classifier on the best performing Gradient Boosting and CatBoost models.

Main Results:

• ML models achieved high performance on the training set, with AUC ranging from 0.79-0.97 and accuracy from 50%-90%.
• The best individual models, Gradient Boosting and CatBoost Classifiers, showed AUCs of 0.95 and 0.97, respectively.
• The ensemble Voting Classifier achieved an AUC of 0.93 and accuracy of 79% on the test set.

Conclusions:

• ML-based CT radiomics effectively predicts histological variants (HV) in urothelial carcinoma (UC).
• This non-invasive radiomic approach offers a valuable tool for identifying prognostic factors that may be missed by conventional radiological evaluation or preoperative biopsies.