Subtyping of hepatocellular adenoma: a machine learning-based approach

Area of Science:

• Hepatobiliary pathology
• Machine learning in medicine
• Oncologic diagnostics

Background:

• Accurate subtyping of hepatocellular adenoma (HCA) is crucial for patient management, but definitive diagnosis relies on complex immunohistochemical and molecular testing.
• Previous attempts to correlate morphologic/clinical features with HCA subtypes have yielded controversial predictive performance.
• There is a need for reliable, accessible methods to aid in HCA subtyping, particularly in resource-limited settings.

Purpose of the Study:

• To utilize machine learning to identify a concise set of morphologic/clinical features for differentiating hepatocellular adenoma (HCA) subtypes.
• To assess the predictive performance of these selected features in classifying HCA subtypes.
• To provide a tool for initial HCA diagnosis and subtyping, complementing existing diagnostic modalities.

Main Methods:

• Analysis of 50 liver HCA resection specimens with assessment of 26 morphologic/clinical features.
• Application of LASSO (least absolute shrinkage and selection operator) for feature selection to identify key differentiating characteristics.
• Validation of selected features using SVM (support vector machine) analysis on an independent cohort of 20 liver resection samples.

Main Results:

• Distinct combinations of morphologic/clinical features were identified as predictive for different HCA subtypes.
• Support Vector Machine (SVM) analysis demonstrated that the selected features achieved an overall accuracy of at least 80% in classifying HCA subtypes.
• The study successfully identified key features for differentiating HNF1α-mutated HCA, inflammatory HCA, and beta-catenin activated HCA.

Conclusions:

• Morphologic and clinical features, when analyzed with machine learning, can effectively aid in the subtyping of hepatocellular adenoma (HCA).
• The identified feature sets offer a promising, less invasive approach for initial HCA classification.
• These findings support the use of selected features for HCA subtyping in clinical settings lacking advanced diagnostic assays.