P39 The role of artificial intelligence in augmenting hepatocellular carcinoma surveillance

<h3>Introduction</h3> Hepatocellular carcinoma (HCC) is one of the leading causes of cancer-related mortality worldwide. Liver cirrhosis is the most significant predisposing factor for HCC. Current guidelines recommend biannual liver ultrasonography with or without alpha-fetoprotein for HCC surveillance in patients with cirrhosis. However, these methods are suboptimal due to limited sensitivity, specificity and operator-dependency. Artificial Intelligence (AI) offers a promising solution to overcome these limitations and improve HCC detection. This narrative literature review explores the effectiveness of AI-based ultrasound techniques for HCC detection and its potential role in current surveillance strategies. <h3>Methods</h3> A literature search was conducted on MEDLINE, Cochrane and EMBASE databases from conception to March 2024 on the OVID platform using the following MESH terms: ((Artificial intelligence) OR (machine learning) OR (deep learning)) AND ((HCC) OR (Hepatocellular carcinoma)) AND ((ultrasound) OR (ultrasonography)). Additional articles were identified by manually searching references in the included articles. Articles were limited to English and human studies and were selected based on relevance to this review’s objective. <h3>Results</h3> AI demonstrates impressive performance in detecting HCC with an accuracy of more than 86% in almost all the reviewed studies (table 1). AI also shows proficiency in distinguishing HCC from cirrhosis and similar focal liver lesions (FLLs) with classification accuracies ranging from 84.0% to 94.5%. In addition, when using ultrasound videos, AI’s ability to detect FLLs (including HCC) was 89.8%, exceeding that of non-radiologists (29.1%) and radiologists (70.9%). Although, the study by Tiyarattanachai et al. (2021) which assessed 2654 ultrasound images with HCC showed a large drop off in HCC detection in cirrhosis (72%) compared to images without cirrhosis (94.7%). <h3>Conclusions</h3> AI shows promise in augmenting HCC surveillance by accurately detecting and classifying HCC from cirrhosis and other FLLs. Integrating AI into current surveillance programmes could provide a valuable second opinion for physicians, improve diagnostic performance, relieving workload from overstretched imaging services and reducing the need for further cross-sectional imaging and invasive tests such as liver biopsy. Despite encouraging findings, there is variability in algorithms and feature extraction methods used across studies and not all HCC cases were explicitly stated to have evolved from cirrhosis in the reviewed studies. Large prospective multicentre controlled trials in the setting of cirrhosis are likely to provide the best evidence in guiding future HCC surveillance guidelines and strategies.