Technology of Hierarchical Classification Applied to Ensemble Methods for Liver Fibrosis Staging in Ultrasound

Automated liver fibrosis staging is crucial for timely diagnosis and treatment. This paper presents a novel Technology of Hierarchical Classification (Tech-HC) applied to ensemble methods for accurate staging from ultrasound images. Using a dataset of 806 manually segmented regions of interest from 426 ultrasound images, we trained and evaluated Random Forest, XGBoost, LightGBM, and the “Random Forest of Optimal Complexity Trees” (RFOCT) models. LightGBM achieved 82% classification accuracy for F0-1 vs. F2-4, 86% for F0-2 vs. F3-4, and 96% for F0-3 vs. F4, while RFOCT achieved 77% accuracy for F0 vs. F1-4. Integrating these models into our hierarchical classification framework resulted in 99% accuracy for all patients. This technology has the potential to significantly improve diagnostic accuracy and efficiency in liver fibrosis assessment, particularly in resource-constrained settings.