MobileNet for ALL Detection

Early diagnosis is critical for leukemia prognosis, yet manual microscopic analysis is time-consuming, motivating the use of automated deep learning–based detection systems. This study proposes a lightweight MobileNet-based CNN enhanced with L1 regularization and advanced dataset balancing and augmentation techniques, achieving 95.33% accuracy and an F1 score of 0.95 on a public leukemia dataset. The model demonstrated robustness to added Gaussian noise and outperformed several existing approaches, highlighting its suitability for efficient and reliable clinical decision support, including mobile and embedded applications.