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

Area of Science:

• Computational Biology
• Histopathology
• Zebrafish Models

Background:

• Convolutional Neural Networks (CNNs) and image pre-processing can improve antioxidant effect classification in zebrafish intestines.
• Existing methods require enhancement for accurate histological image analysis.

Purpose of the Study:

• To propose a hybrid deep learning (DL) technique combining Xception CNN, autoencoder, custom CNN, and Vision Transformer (ViT) for zebrafish intestine classification.
• To evaluate the efficacy of Support Vector Machine (SVM), Random Forest (RF), and k-Nearest Neighbors (kNN) for feature classification.
• To assess the impact of Contrast-Limited Adaptive Histogram Equalization (CLAHE) on classification accuracy.

Main Methods:

• A hybrid DL model integrating Xception, autoencoder, custom CNN, and ViT was developed.
• CLAHE and AI algorithms were applied for pre-processing histological images of zebrafish intestine.
• SVM, RF, and kNN were used for classification of DL-generated features.
• Binary classification tasks included control vs. oxidative stress (OS), OS vs. OS + theobromine (TB), and OS vs. OS + caffeine (CAF).

Main Results:

• Hybrid models utilizing pre-processed images showed improved classification accuracy compared to raw images.
• The best performance was achieved with a hybrid Xception and SVM model for the OS vs. OS + TB classification.
• This combination yielded an accuracy of 84.6% for pre-processed images, compared to 78.4% for raw images.

Conclusions:

• The proposed hybrid DL approach, combined with CLAHE pre-processing, significantly enhances the accuracy of classifying zebrafish intestinal morphology.
• The Xception-SVM hybrid model demonstrates superior performance, particularly for distinguishing oxidative stress and theobromine effects.
• This technique offers a promising tool for analyzing chemical treatment effects on zebrafish intestines.