Smartphone-Based Blood Cancer Classification Leveraging MobileNet Architecture

Blood cancer is a serious health concern worldwide, and the better prognosis of patients depends on early detection. Early detection of blood cancer significantly improves patient outcomes, but access to diagnostic infrastructure remains limited, particularly in resource-constrained regions. This paper presents a smartphone-based deep learning system for real-time blood cancer classification using peripheral blood smear (PBS) images. We utilized the publicly available Blood Cell Cancer ALL 4-Class dataset from Kaggle, consisting of 89 patient records across four diagnostic categories. Leveraging transfer learning, we adapted the MobileNetV2 architecture with a custom classification head, including global average pooling, dense, and dropout layers, to improve generalization while maintaining computational efficiency. The model achieved superior performance with 99% accuracy, outperforming VGG-16. The trained model was converted to TensorFlow Lite and deployed into a Flutter-based mobile application for on-device, offline diagnosis. Our solution demonstrates the potential of lightweight mobile AI systems to expand blood cancer screening access and reduce diagnostic disparities in underserved populations.