Optimal radiological gallbladder lesion characterization by combining visual assessment with CT-based radiomics

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Differentiating benign gallbladder diseases from gallbladder cancer (GBC) is challenging due to similar imaging appearances.
• Accurate differentiation is crucial for appropriate patient management and treatment planning.

Purpose of the Study:

• To evaluate the efficacy of CT-based radiomic features analyzed by machine learning (ML) in discriminating benign gallbladder disease from GBC.
• To assess the added value of ML models to conventional radiological visual interpretation of CT scans.

Main Methods:

• Retrospective selection of patients with histopathologically confirmed gallbladder lesions and contrast-enhanced CT scans.
• Extraction of radiomic features from gallbladder lesions.
• Analysis using ML classifiers including Lasso regression, Ridge regression, and XG Boosting.
• Comparison of ML-assisted interpretation with visual CT assessment alone.

Main Results:

• The XG Boosting classifier achieved an AUC of 0.81 for differentiating GBC.
• Combining visual CT interpretation with XG Boosting predictions resulted in a significantly improved AUC of 0.98.
• This combined approach yielded a sensitivity of 91%, specificity of 93%, and accuracy of 92%.

Conclusions:

• CT-based radiomic feature analysis using ML demonstrates potential for distinguishing benign gallbladder disease from GBC.
• The combination of radiomic analysis and visual CT interpretation offers the most effective strategy for accurate differentiation.
• This integrated approach enhances diagnostic performance beyond visual assessment alone.