Development of random forest machine learning model for the detection of changes in liver tissue after exposure to iron oxide nanoparticles

Abstract

Introduction/Aim: The aim of our study was to create a machine learning model, specifically a random forest model, which uses textural data from liver micrographs to differentiate between normal hepatic tissue and damaged tissue exposed to iron oxide nanoparticles. Material and Methods: Regions of interest in micrographs of hepatic tissue, obtained from mice treated with iron oxide nanoparticles and controls, were analyzed using the gray-level co-occurrence matrix (GLCM) method. The resulting GLCM features were employed as input data for the training and testing of the random forest model using the "Scikit-learn" library in the Python programming language. Additionally, a conventional decision tree model was developed, based on the classification and regression tree (CART) algorithm. Results: The random forest model outperformed the alternative CART decision tree approach in terms of classification accuracy, correctly predicting the class for 73.67% of the instances in the validation ROI dataset. The area under the receiver operating characteristic curve was 0.81, indicating relatively good discriminatory power. The F1 score for the model was 0.74, showcasing fairly good precision and recall, though not perfect. Conclusion: The data obtained from this study may be utilized for further development of artificial intelligence computation systems to identify physiological and pathophysiological changes in hepatic tissue. The results also serve as a starting point for additional research on the automation of histopathological analysis of liver tissue exposed to external toxic agents.