Automatic classification of white regions in liver biopsies by supervised machine learning

Area of Science:

• Digital pathology
• Computational analysis
• Liver histopathology

Background:

• Non-alcoholic fatty liver disease (NAFLD) assessment relies on human interpretation of histological features, introducing variability.
• Automated methods can offer continuous measurements and reduce inter-observer variability in NAFLD diagnosis.

Purpose of the Study:

• To develop and validate an automated classification algorithm for histological features in liver biopsies.
• Specifically, to accurately identify steatosis and other white regions in H&E-stained slides.
• To assess the algorithm's performance in distinguishing normal histology from NAFLD.

Main Methods:

• Supervised machine learning classifiers were trained on digital images of 47 liver biopsies (20 normal, 27 NAFLD).
• Expert pathologists provided annotations for training the algorithm.
• The algorithm was evaluated on its accuracy in classifying steatosis, central veins, portal veins, portal arteries, sinusoids, and bile ducts.

Main Results:

• The classification algorithm achieved 89% overall accuracy.
• High precision and recall (≥ 82%) were obtained for macrosteatosis, bile ducts, portal veins, and sinusoids.
• The accuracy for macrosteatosis identification is reported as the best to date.

Conclusions:

• Automated histological assessment of NAFLD, particularly macrosteatosis, is feasible and accurate.
• This technology has significant clinical and research applications for NAFLD evaluation.
• Accurate identification of liver anatomical landmarks aids in subsequent lesion localization.