Implementation of a Hybrid Deep Learning Framework for Automated Malaria Classification: Integrating Convolutional Neural Network, ResNet, and Transformer Models

Malaria is one of the most severe infectious diseases in the world, especially in tropical and subtropical areas, where timely diagnosis is essential for effective treatment.While microscopic examination of blood smear images is the gold standard, it is time-consuming, requires skilled personnel, and is prone to human error.Such restrictions have led to the emergence of automated computer-aided diagnostic systems based on deep learning frameworks with better performance.This paper builds upon our earlier published paper on convolutional neural network (CNN)-based and Transformer-based malaria classification and proposes three hybrid deep learning architectures to detect malaria automatically.In addition to the standalone CNN and Transformer (Vision Transformer) models, the proposed model introduces (i) a CNN-Transformer-based hybrid model, (ii) a VGG19-Transformer hybrid model, and (iii) a ResNet-Transformer hybrid model.The goal of these architectures is to integrate the ability of convolutional networks to robustly extract local features with the ability of Transformer-based attention mechanisms to extract global features.Each model is trained and tested on the available NIH Malaria Cell Image Dataset.Experimental results show that the proposed hybrid models outperform the standalone architectures, with the CNN-Transformer architecture achieving a maximum validation accuracy of around 95%.The results indicate that hybrid architectures offer a more comprehensive and efficient system for automated malaria diagnosis and show a promising direction for future medical image analysis research.