Feasibility of large language models for CEUS LI-RADS categorization of small liver nodules in patients at risk for hepatocellular carcinoma
2024

Using AI to Diagnose Small Liver Tumors

Sample size: 403 publication 10 minutes Evidence: high

Author Information

Author(s): Huang Jiayan, Yang Rui, Huang Xiaotong, Zeng Keyu, Liu Yan, Luo Jun, Lyshchik Andrej, Lu Qiang

Primary Institution: West China Hospital of Sichuan University

Hypothesis

Can large language models improve the diagnosis of small hepatocellular carcinoma in high-risk patients using CEUS LI-RADS?

Conclusion

Large language models integrated with CEUS LI-RADS can effectively diagnose small hepatocellular carcinoma in high-risk patients, with ChatGPT-4.0 showing superior sensitivity compared to human readers.

Supporting Evidence

  • ChatGPT-4.0 showed higher sensitivity in detecting small hepatocellular carcinoma than ChatGPT-4o.
  • ChatGPT-4.0 demonstrated superior sensitivity compared to human readers in diagnosing small hepatocellular carcinoma.
  • LLMs achieved substantial to almost perfect intra-agreement for CEUS LI-RADS categorization.

Takeaway

This study shows that AI can help doctors find small liver tumors better, especially in patients who are at high risk for liver cancer.

Methodology

The study evaluated the performance of four large language models in diagnosing small hepatocellular carcinoma using structured CEUS LI-RADS reports from high-risk patients.

Potential Biases

Potential biases in LLM outputs due to reliance on structured reporting and the subjective nature of radiological assessments.

Limitations

The study relied on structured reports, limiting access to full imaging data, and had a small sample size for certain CEUS LI-RADS categories.

Participant Demographics

Mean age of participants was 52.3 years, with 80.1% being men.

Statistical Information

P-Value

p<0.004

Confidence Interval

95% CI: 73%, 90%

Statistical Significance

p<0.004

Digital Object Identifier (DOI)

10.3389/fonc.2024.1513608

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication