Advanced CNN Architectures for Enhanced Detection of Leukemia Cells in Bone Marrow Images

The rapid advancement of deep learning techniques has significantly improved the accuracy of medical image analysis, particularly in the detection and classification of leukemia. In this study, an advanced Convolutional Neural Network (CNN) is proposed to automate the process of cell detection in bone marrow images to improve leukemia cell detection. The accuracy of the classification by using the weighted ensemble learning approach incorporating the multiple models CNN ResNet-152, AlexNet, VGGNet, Inception, ResNet-50, ResNet-18, and DenseNet-121 is superior. Finally, the system is trained using images of Peripheral Blood Smear (PBS) labels, ensuring robustly extracted features and the differentiating benign and malignant cells. There are two data preprocessing techniques, such as normalization, resizing, and augmentation, that help to generalize the model. The methodology is supposed to use stratified k-fold cross-validation for steady training and testing and validate its effectiveness by performance measures such as accuracy, precision, recall, and F1 score. Towards a web-based platform, using Flask/Django, the system is deployed and offers real-time diagnostic support to clinicians. It is found that the ensemble model outperforms individual CNN architectures when comparing the accuracy and reliability attained in the system. Located in the resource-constraining poor settings, this AI-palled diagnostic tool provides an early leukemia detection by a non-invasive, efficient, accessible, timely, and accurate clinical decisions.