Improving Detection of Intrahepatic Cholangiocarcinoma with a Contrast-enhanced US-based Deep Learning Model.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Hepatobiliary Oncology

Background:

• Intrahepatic cholangiocarcinoma (iCCA) diagnosis relies heavily on imaging.
• Contrast-enhanced ultrasound (CEUS) offers valuable insights but requires expert interpretation.
• Deep learning (DL) has the potential to enhance diagnostic accuracy in medical imaging.

Purpose of the Study:

• To develop and evaluate a DL model utilizing CEUS data for improved iCCA diagnosis.
• To compare the diagnostic performance of the DL model against human radiologists.
• To assess the impact of DL-assisted interpretation on radiologist performance.

Main Methods:

• Retrospective analysis of 1148 CEUS examinations from multiple centers (July 2017-December 2023).
• Training and validation of four DL algorithms (BNInception, MobileNet-v2, ResNet-50, VGG-19).
• External validation using two test sets, including comparison with CEUS and MRI radiologists.

Main Results:

• The ResNet-50 DL model achieved the highest performance (AUC, 0.92) in initial testing.
• The DL model demonstrated diagnostic performance similar to senior CEUS (AUC, 0.91) and MRI radiologists (AUC, 0.91).
• DL assistance significantly improved junior (AUC 0.72 to 0.89) and midlevel (AUC 0.78 to 0.90) CEUS radiologists' performance.

Conclusions:

• A CEUS-based DL model shows expert-level diagnostic performance for iCCA.
• The DL model serves as a valuable tool to augment radiologist capabilities, particularly for less experienced practitioners.
• This AI approach holds promise for enhancing the diagnostic workflow in hepatobiliary oncology.