Integration of Deep Learning Architectures With GRU for Automated Leukemia Detection in Peripheral Blood Smear Images

This project introduces an advanced web-based system for automated leukemia detection and classification from peripheral blood smear images using deep learning techniques. The system focuses on identifying four distinct classes—Benign, Early, Pre-leukemic, and Pro-leukemic—using high-performance neural network architectures. A curated dataset from Kaggle serves as the foundation for training and validating the models. The implemented models include Xception combined with Bi-GRU, EfficientNetB3 fused with Bi-GRU, EfficientNetB3 with Vision Transformer (ViT), and MobileNetV3 paired with LSTM. These architectures enable robust spatial and sequential feature extraction, enhancing diagnostic accuracy. The backend is developed using Python and Flask, responsible for training, prediction, and secure user management. The frontend, designed with HTML, offers a seamless user experience, allowing individuals to register, log in, and upload blood smear images. Once an image is submitted, the system processes it through the trained models to deliver an accurate leukemia stage classification. This project bridges deep learning with healthcare, providing an accessible, fast, and intelligent tool for preliminary leukemia screening. It aims to assist medical practitioners in early diagnosis and reduce reliance on manual microscopic analysis, which is often subjective and time-consuming.

Keywords:
Leukemia detection, Blood smear classification, Deep learning, Xception, EfficientNetB3, Bi-GRU, ViT, MobileNetV3, LSTM, Flask, Medical imaging, Web application, Early diagnosis, AI in healthcare.