Development and Validation of a Deep Learning and Radiomics Combined Model for Differentiating Complicated From...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Acute appendicitis (AA) diagnosis can be challenging, with complicated cases requiring prompt intervention.
• Distinguishing between complicated and uncomplicated AA is crucial for appropriate patient management and surgical planning.

Purpose of the Study:

• To develop and validate a combined deep learning and radiomics model for differentiating complicated from uncomplicated acute appendicitis (AA).
• To assess the diagnostic performance of the novel model against conventional methods and radiologist interpretation.

Main Methods:

• A retrospective multicenter study involving 1165 adult AA patients with abdominal pelvic CT scans.
• Development of a CatBoost-based model integrating clinical, CT visual, deep learning, and radiomics features.
• External validation and comparison with conventional combined model, DLR model, and radiologist visual diagnosis using ROC analysis.

Main Results:

• The combined model achieved an AUC of 0.816 in the training cohort and demonstrated robust performance in the validation cohort (AUC=0.799).
• The model outperformed the conventional combined model (AUC=0.723), DLR model (AUC=0.755), and radiologist diagnosis (AUC=0.679) (P < 0.05).
• Decision curve analysis indicated superior net benefit for the combined model in predicting complicated AA.

Conclusions:

• The developed combined deep learning and radiomics model enables accurate differentiation of complicated and uncomplicated AA.
• This AI-powered tool shows significant potential for improving diagnostic accuracy in acute appendicitis cases.