An Explainable AI and Optimized Multi-Branch Convolutional Neural Network Model for Eye Anemia Diagnosis

This project proposes a deep learning-powered system for the non-invasive diagnosis of anemia by analyzing eye images. The dataset used, "Eyes Defy Anemia" from Kaggle, consists of a limited number of samples, making data augmentation a crucial step. Techniques such as rotation, flipping, brightness enhancement, and zooming were applied to artificially expand the dataset and improve model generalization. Multiple convolutional neural network (CNN) architectures were evaluated, including VGG19, MobileNet, and InceptionNet, for effective feature extraction and classification. Additionally, a hybrid approach combining MobileNet with Support Vector Machine (SVM) was implemented to enhance classification accuracy by leveraging deep features with traditional machine learning. The best-performing models were integrated into a web application using Flask for the backend and HTML, CSS, and JavaScript for the frontend. Users can upload an eye image and receive an instant diagnosis indicating whether the subject is anemic or not. The system is lightweight, responsive, and designed for accessibility in real-time environments. This work demonstrates a practical, efficient, and low-cost solution for preliminary anemia screening, especially beneficial in remote healthcare scenarios where traditional diagnostic tools are not easily accessible.

 

Keywords:
Eye Anemia Detection, Data Augmentation, VGG19, MobileNet, InceptionNet,