AI facilitates early detection of hepatocellular carcinoma

Presented by

Dr Kin Nam Kwok, Hong Kong University, Hong Kong

Conference

ESMO GI 2024

Doi
https://doi.org/10.55788/c6350017

An AI-driven algorithm based on routine blood tests outperformed ultrasonographic screening combined with alpha-fetoprotein (AFP) testing for the early detection of hepatocellular carcinoma (HCC) in high-risk individuals.

Most patients with HCC are diagnosed with advanced-stage disease, when palliative treatment is the only option, because early detection of HCC is challenging. The sensitivity of ultrasonographic screening combined with AFP testing, the most commonly used biomarker for HCC, is limited, especially in the lower stages of HCC. In addition, individuals often avoid screening appointments because of anxiety around an ultrasound test. To improve early detection of HCC, Dr Kin Nam Kwok (Hong Kong University, Hong Kong) and colleagues aimed to develop an AI-driven algorithm based on routine blood tests [1].

The algorithm was trained using data from 3,415 patients with HCC, including complete blood counts, liver and renal function tests, and clotting profiles. This led to a sensitivity of 79.4% in the detection of HCC in blood samples at 1–30 days before clinical diagnosis (compared with a sensitivity of 43.7% for AFP testing). In time, sensitivity decreased to 61.3% for the detection of HCC at 1–3 months before clinical diagnosis, 50.1% for 3–6 months before diagnosis, 44.2% for 6–9 months before diagnosis, and 41.3% for 9–12 months before diagnosis. In comparison, the sensitivity of the AFP test to reach those time goals in the same cohort was 41.9%, 37.9%, 29.8%, and 21.3%, respectively. In addition, the specificity of the algorithm was over 75% in all time intervals.

“This AI algorithm based on routine blood tests might bring forward the diagnosis of HCC in 40% of patients by 1 year, and thus creates a meaningful window for timely intervention. Potentially, this can lead to cancer mortality reduction,” Dr Kwok concluded.

Further prospective studies to validate the potential of the algorithm are needed. Moreover, validation in a non-Asian cohort is essential.

Kwok KN, et al. Early detection of HCC by routine blood based-AI. Abstract 165MO, ESMO Gastrointestinal Cancers Congress 2024, 26–29 June, Munich, Germany.

Related Articles