Opportunistic Detection of Hepatocellular Carcinoma Using Noncontrast CT and Deep Learning Artificial Intelligence
Chengzhi Peng1, Philip Leung Ho Yu2, Jianliang Lu1
1Department of Medicine, School of Clinical Medicine, The University of Hong Kong, Hong Kong.
Journal of the American College of Radiology : JACR
|March 5, 2025
Summary
An artificial intelligence model using noncontrast CT scans shows promise for detecting hepatocellular carcinoma (HCC). This AI approach offers comparable accuracy to radiologists, aiding in early HCC screening.
Area of Science:
- Hepatology and Medical Imaging
- Artificial Intelligence in Oncology
- Radiology and Diagnostic Imaging
Background:
- Hepatocellular carcinoma (HCC) presents a significant global health challenge, underscoring the need for effective early detection strategies.
- Opportunistic screening using existing imaging data, particularly noncontrast computed tomography (CT), remains underexplored for HCC diagnosis.
- Developing advanced AI tools is crucial for improving diagnostic efficiency and accuracy in HCC detection.
Purpose of the Study:
- To develop and validate an artificial intelligence (AI) algorithm for the detection of hepatocellular carcinoma (HCC) using only noncontrast CT scans.
- To assess the performance of the AI model in comparison to expert radiological interpretation.
- To investigate the potential of AI-driven analysis of noncontrast CTs for opportunistic HCC screening.
Main Methods:
- A 3-D convolutional block attention module (CABM) model was developed and trained on noncontrast multiphasic CT scans.
- Patient data included 2,223 patients for internal validation and 584 for external testing, with HCC diagnosis confirmed via established guidelines and a 12-month clinical composite reference standard.
- Radiological interpretation and Liver Imaging Reporting and Data System (LI-RADS) annotations were used for comparison and validation.
Main Results:
- The CABM model achieved an area under the receiver operating curve (AUC) of 0.807 in internal validation, comparable to radiologists (AUC 0.851).
- On external testing, the model demonstrated an AUC of 0.789, indicating robust generalizability.
- The AI model showed promising performance across various patient subgroups, including those with definite HCC, indeterminate scans, and small lesions (< 2 cm).
Conclusions:
- The developed CABM AI model demonstrates diagnostic accuracy comparable to that of radiologists in internal validation.
- AI analysis of noncontrast CT scans holds significant potential as a tool for opportunistic screening of hepatocellular carcinoma.
- Further research and validation are warranted to integrate AI into routine clinical practice for HCC detection.
