Leukemia Classification through Deep Learning Techniques and Generative AI

Leukemia is a cancer originating in the bone marrow and leads to rapid proliferation of abnormal blood cells. The main objective of this study is to implement a Convolutional Neural Network (CNN) to detect and classify leukemia from microscopic cell images. The proposed framework combines a Generative Adversarial Network (GAN) that generates synthetic images of healthy cells to address class imbalance and training on a balanced leukemia dataset, with four different CNN architectures (InceptionV3, ResNet50, EfficientNetB3 and InceptionV4) - the effectiveness of this approach is validated on a Breast Cancer tumor dataset consisting of ultrasound images. Unlike prior studies that rely on standard augmentation, our approach incorporates synthetic image quality metrics (FID, IS, SSIM) to validate realism and structural fidelity.The results reveal GAN architecture achieving 16% higher performance on cell images compared to tumor images. Additionally, results obtained for each model were 76%, 80%, 75%, and 75% respectively, with RestNet50 attaining the best result. Obtained results underline potential contribution of deep learning in cancer detection and improving clinical outcomes through GAN-augmentation, addressing class imbalance effectively.