Abstract
Background: Liver fibrosis staging is critical for patient selection and management prior to transplantation, but biopsy is invasive and serum biomarkers lack accuracy. Near-infrared spectroscopy (NIRS) is an emerging non-invasive technology that can detect liver fibrosis via changes in tissue composition. Machine learning (ML) enables analysis of NIRS data for diagnostic modeling.
Purpose: To review the potential of NIRS-ML approaches for rapid, point-of-care liver fibrosis detection, including technological principles, promising applications, current limitations, and future directions.
Main body of the abstract: NIRS leverages unique near-infrared absorbance patterns reflecting collagen accumulation, lipid reduction, and other chemical alterations in fibrotic liver. Handheld/hyperspectral systems acquire tissue spectroscopic data in minutes. Multiple human studies correlate NIRS with histological fibrosis scores. ML techniques like partial least squares regression, neural networks, support vector machines, and random forests analyze spectra to develop optimized diagnostic algorithms. Initial models differentiate mild versus advanced fibrosis and stage cirrhosis with high accuracy, outperforming traditional biomarkers. Recent advances include smartphone-based scanning, cloud computing, and integrated user-friendly platforms. However, further large validation trials, standardization, assessment of confounding factors, improved ML methodology, and cost-effectiveness data are required before widespread clinical implementation.
Conclusion: With ongoing research to address remaining barriers, NIRS-ML approaches hold great disruptive potential for rapid, non-invasive point-of-care quantification of liver fibrosis, including optimizing transplant surgery planning and management.
Keywords
Liver fibrosis, Liver Transplantation, Near-infrared spectroscopy, Machine learning, Point-of-care, Biopsy, Non-invasive