Custom Residual SE-CNN for Acute Lymphoblastic Leukemia Detection from Microscopic Images

Acute Lymphoblastic Leukemia (ALL) is a rapidly progressing blood cancer that primarily affects children and demands timely diagnosis for effective treatment. In this work, we propose a custom deep learning model built from scratch, in- tegrating Residual Convolutional Neural Networks with Squeeze- and-Excitation (SE) blocks, to classify blood smear images into leukemic (blast) and healthy (normal) categories. Unlike conventional approaches that rely on transfer learning with pre-trained models, our architecture is specifically tailored to the characteristics of microscopic medical images. The model was trained on the ALL-IDB1 dataset using a robust pipeline that includes data augmentation, class balancing, and mixed- precision training to enhance generalization and reduce over- fitting. Experimental results demonstrate that our model not only achieves superior accuracy but also offers computational efficiency, making it suitable for deployment in real-world clinical environments, especially in resource-constrained settings.