Executing Spark BigDL for Leukemia Detection from Microscopic Images using Transfer Learning

Abstract

Acute Leukemia disease is the bone marrow problem common both in children and adults. Medical image analytics is applied in the field of Digital Image Processing (DIP) and Deep Learning (DL). The role of deep learning in medical research with big data has been a huge benefit, opening new doors and possibilities for disease diagnostics procedures. Now the medical specialists like pathologists, hematologists, mammalogists and researchers are working in deep learning area. The proposed methodology is Leukemia detection by implementing apache spark BigDL library from the microscopic images of human blood cells using Convolutional Neural Network (CNN) architecture GoogleNet deep transfer learning. The proposed system is an efficient enough to detect 4 types of leukemia Acute Myeloid Leukemia (AML), Actuate Lymphocytic Leukemia (ALL), Chronic Myeloid Leukemia (CML) and Chronic Lymphocytic Leukemia (CLL) and normal from the microscopic images of human blood sample. The proposed methodology after using Spark BigDL framework with Google Net architecture, we achieved 97.33% accuracy in case of training and 94.78% of validation respectively. Moreover we are also compared our model without BigDL GoogleNet. The accuracy of training and validation accuracy are 96.42% and 92.69% respectively. The BigDL model outperformed the Keras model with more efficient and accurate results.