Artificial intelligence strategies based on random forests for detecting ischemia-reperfusion injury changes in kidney tissue during intravital imaging.

Abstract

This study presents a supervised machine learning approach using a Random Forest classifier to detect ischemia-reperfusion injury (IRI) in kidney tissue based on intravital two-photon microscopy data. A rodent model of unilateral renal IRI was used, with 30&#xa0;min of pedicle occlusion followed by 15&#xa0;min of reperfusion. Continuous imaging captured nuclear (Hoechst 33342), vascular (FITC-dextran), and mitochondrial (TMRM) changes in real time. From extracted video frames, 2000 manually segmented regions of interest (ROIs), 1000 control and 1000 injured segments, were analyzed using texture analysis. Five textural features were used as input: angular second moment (ASM) and inverse difference moment (IDM) from gray-level co-occurrence matrix (GLCM); short run emphasis (SRE) and long run emphasis (LRE) from run length matrix (RLM); and HH wavelet coefficient energy (EnHH) from discrete wavelet transform (DWT). All showed significant differences (p < 0.001) between injured and control tissue. The Random Forest model achieved 79.8% accuracy, a macro F1-score of 0.79, a Matthews Correlation Coefficient of 0.5959, and an ROC AUC of 0.83. These findings highlight the potential of AI-based texture analysis to detect early nuclear and vascular alterations during IRI. Future work should expand datasets, include 3D analyses, and incorporate multimodal imaging for greater generalizability.