Deep Learning for Prediction and Classification of Bone Marrow Blood Cancer

Leukemia is a highly malignant blood cancer manifesting complications in early diagnosis and treatment. Conventional imaging involves interpretation by experts and is time-consuming. It herein proposes an automatic identification of leukemia from peripheral blood smear images using deep-learning AI. A CNN architecture has been built and evaluated using 3 pretrained models, namely MobileNetV2, VGG19, and InceptionV2, for the classification of benign and malignant samples. The raw images were subjected to preprocessing techniques such as resizing, normalization, augmentation, and noise reduction to enhance the model accuracy. The performance metrics computed included accuracy, precision, recall, and F1 score for comparison among models. The model that was proven to be the best one was MobileNetV2 model-for a good trade between high accuracy and computational expense-the proposed system brings into existence a bright framework for real-time screening of leukemia by providing adequate assistance to the health care professionals for rapid diagnosis. The future expansion of the work would involve increasing interpretability in the model and application to larger, more heterogeneous medical datasets.