Acute Lymphoblastic Leukaemia Diagnosis Using EfficientNet B3 Transfer Learning Model
Rudresh Pillai, Neha Vaishnavi Sharma, Rupesh Gupta · 2023
Acute Lymphoblastic Leukaemia (ALL) is a kind of cancer that affects the leukocytes in the bone marrow and blood. Improved treatment results and patient survival depend on the early identification of ALL. A computer-aided diagnosis method for ALL detection was created in this work using a dataset of 15,135 pictures from 118 individuals with two labelled classifications (normal cell and leukaemia blast). For classification, the EfficientNetB3 model of transfer learning was used. The training (70%), validation (12%), and testing (18%) sets were created by randomly dividing the dataset. To enhance the dataset size and avoid overfitting, data augmentation techniques were used on the training set. With a sensitivity of 97.58% and a specificity of 92%, the model's accuracy on the testing set was 95.75%. These findings show the potential of deep learning systems for ALL early diagnosis. Overall, this work shows that leukaemia blast cells in peripheral blood smears can be effectively classified using the EfficientNetB3 transfer learning model. The creation of automated image-based diagnostic tools can assist in shortening the manual inspection process and save time and money while also producing more accurate and repeatable findings. This would help in reducing preventable deaths and neonatal mortality while improving the healthcare sector.