CapsENet: Deep Learning based Acute Lymphoblastic Leukemia Detection Approach
K Lalithkumar, Naveen Raj B, M. Priyanga, S Sandhya, M. Karthiga · 2024
The abnormal formation of lymphocytes in the bone marrow causes one of the dangerous forms of blood cancer, which is called Acute LymphoblasticLeukaemia (ALL). While it is most commonly diagnosed in children, when it is often curable, it can also occur in adults; and poor outcomes are typically due to a late stage at diagnosis. This work proposes a high-level screening method that incorporates an intelligent deep learning (DL) system to enhance the first-stage detection of this possibly fatal disease. The described method is based on the integration of CapsuleNetworks (CapsNet) with EfficientNet (ENet) for the classification of leukemic and healthy blood cells by analyzing the images of small blood smears. Based on the microscopic image databases obtained from the public domain, the designed CapsENet model was thoroughly tested and trained. Training was conducted on the GoogleColaboratory platform at optimal performance with the Nvidia Tesla P-100 GPU. The outcomes achieved by the model are as follows; 96. 95% accuracy, 96. Specificity is 88%, sensitivity – 96%, Cohen's kappa coefficient – 96. 97% F1-score, and 97. 96% precision. These results proof the high efficiency of the classifier in detecting leukemic cells and its applicability for the peripherical blood test and CBC preliminary screening. This could lead on to earlier interventions in patient management and improved patient outcomes.