Hybrid Deep Learning Models for Analyzing Histological Images of the Zebrafish Intestine Under Oxidative Stress

Background/Objectives: Convolutional Neural Networks (CNNs) and advanced image pre-processing can enhance the classification of antioxidant effects applied on zebrafish intestine. This study proposes a hybrid technique that combines four deep learning (DL) models: the pre-trained Xception CNN, a custom-built autoencoder, a custom-built CNN, and a Vision Transformer (ViT). Methods: For classification features generated by DL models, the Support Vector Machine (SVM), Random Forest (RF), and k-Nearest Neighbors (kNN) algorithms were proposed. Contrast-Limited Adaptive Histogram Equalization (CLAHE) and the mentioned DL artificial intelligence (AI) algorithms were applied to improve the accuracy of the classification of histological images of zebrafish intestinal morphology. Results: In a binary classification, the following classes were studied on zebrafish intestine: (i) control and experimental-induced oxidative stress (OS); (ii) OS versus OS and theobromine (TB); and (iii) OS versus OS and caffeine (CAF). The novelty of the research lies in applying CLAHE to enhance image quality and utilizing four hybrid models to improve classification accuracy compared to raw images, when a private dataset of zebrafish intestine histology under certain chemical treatments (OS, TB, CAF) was employed. Conclusions: The best results are obtained in a binary classification with a hybrid combination of Xception and SVM for OS versus OS and TB classes, with an accuracy of 84.6% for pre-processed images, better than raw images, when the accuracy was 78.4%.