The Value of Deep Learning in Gallbladder Lesion Characterization.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Hepatobiliary Surgery

Background:

• Differentiating gallbladder cancer (GBC) from benign gallbladder lesions is diagnostically challenging due to their similar appearances.
• Accurate diagnosis is crucial for timely and appropriate patient management.

Purpose of the Study:

• To evaluate the efficacy of a convolutional neural network (CNN) in differentiating GBC from benign gallbladder diseases.
• To determine if incorporating adjacent liver parenchyma information enhances CNN performance for GBC detection.

Main Methods:

• Retrospective analysis of contrast-enhanced CT scans from 127 patients (44 GBC, 83 benign lesions).
• A CT-based CNN was trained on gallbladder images alone and on images including a 2 cm adjacent liver parenchyma.
• Performance was assessed using Area Under the Curve (AUC) and compared between models.

Main Results:

• The CNN trained with adjacent liver parenchyma achieved a higher AUC (0.81) compared to the CNN trained on the gallbladder alone.
• The inclusion of liver parenchyma demonstrated a >10% improvement in performance.
• Combining CNN results with radiological visual interpretation did not further enhance diagnostic accuracy.

Conclusions:

• CT-based CNN models show potential for differentiating GBC from benign gallbladder lesions.
• Adjacent liver parenchyma provides valuable information that improves CNN performance in characterizing gallbladder lesions.
• Further validation in larger, multicenter studies is recommended to confirm these findings.