Identification of Parameters Associated with Hepatic Hemangioma Growth Patterns Using a Machine Learning Model: A Single-Centre Analysis

Abstract

Hemangioma, the most common benign tumour of the liver, generally has a clinical course that causes no physical discomfort. However, some patients may experience rapid growth. Despite its limited role in guiding management, there is currently no established criterion or trigger factor to predict lesion growth in daily practice. This retrospective, single-centre study included patients diagnosed with hepatic hemangioma who were followed up and treated between January 2002 and March 2024. Logistic regression analysis was performed to assess the relationship between changes in lesion size and clinical parameters. Machine learning models were then applied to improve predictive accuracy, and feature importance analysis was performed to identify key variables influencing tumour progression. A total of 38 patients diagnosed with hemangioma were included in this study. Female patients (n = 29 (76.3%)) predominated in the present analysis. Mean follow-up was 3.5 years (median, 1.5 years). The mean lesion size was 41.9 mm (median, 29.5 mm). Some patients had solitary lesions, while others presented with multiple hemangiomas (18.4% had more than one lesion). Correlation analysis between follow-up time and change in tumour size showed a very weak negative association (r = -0.069, p = 0.695), indicating that longer follow-up did not significantly predict progression of the hemangioma. Follow-up time plays an important role in detecting tumour progression, but may not be a direct driver of tumour growth. Instead, longer follow-up is likely to increase the detection of incidental changes in size rather than indicating true tumour acceleration.