Classification of Multiple Myeloma Cancer Cells Using Convolutional Neural Networks and Transfer Learning
Sanju, Ashok Kumar · 2021 Asian Conference on Innovation in Technology (ASIANCON) · 2021
An image processing based computer aided diagnostic tool for Multiple Myeloma cancer diagnosis was developed. Multiple Myeloma involves white blood cells in the bone marrow becoming cancerous, which can severely damage the bones and kidneys and cause death. We used 1451 images for our study from a recently publicly released dataset (SegPC-2021 Challenge) of stained microscope images which contained normal and cancerous cells. For the task of classifying images containing cancerous cells or not, we obtained a good classification accuracy of 93.58%, which was further improved using Transfer Learning and Fine Tuning with different pre-trained Convolutional Neural Networks models, namely VGG (Visual Geometry Group), VGG-16, VGG-19, MobileNetV1, and MobileNetV2 that gave an accuracy of 94.00%, 95.50%, 99.90%, and 97% respectively.