Automated Detection of Canine Babesia Parasite in Blood Smear Images Using Deep Learning and Contrastive Learning Techniques

This research introduces a novel method that integrates both unsupervised and supervised learning, leveraging SimCLR (Simple Framework for Contrastive Learning of Visual Representations) for self-supervised learning along with different pre-trained models to improve microscopic image classification of Babesia parasite in canines. We focused on three popular CNN architectures, namely ResNet, EfficientNet, and DenseNet, and evaluated the impact of SimCLR pre-training on their performance. A detailed comparison of the different variants of ResNet, EfficientNet, and Densenet in terms of classification accuracy and training efficiency is presented. Base models such as different variants of the ResNet, EfficientNet, and DenseNet models were utilized within the SimCLR framework. Firstly, the models were pre-trained on unlabeled images, followed by training classifiers on labeled datasets. This approach significantly improved the robustness and accuracy, demonstrating the potential benefits of combining contrastive learning with conventional supervised techniques. The highest accuracy of 97.07% was achieved by Efficientnet_b2. Thus, detection of Babesia or other hemoparasites in microscopic blood smear images could be automated with high accuracy without using a labelled dataset.