Multiple Myeloma. Detection from Histological Images Using Deep Learning

Abstract

Multiple myeloma (MM)is a blood cancer that develops in the bone marrow. which is the soft spongy tissue found in the center of bones. It is the second-most common form of blood cancer in the KSA, comprising about 1% of all cancers. It necessitates an early diagnosis to lower the mortality risk. The conventional diagnostic tools are resource intensive. As a result, these solutions are difficult to scale in order to reach a wider audience. There are many ways to diagnose multiple myeloma (MM) in bone marrow such as complete blood count test (CBC) or counting myeloma plasma cells in aspirate slide images. However, using these manual detection techniques are very time-consuming and the result will depend upon the experience of the pathologist. So, computer-aided techniques are used for fast processing and accuracy. That’s beneficial to consultant and especially to pathologist who plays the role of an assistant in the diagnosis of multiple myeloma by a state of art method developed with deep learning. The Deep learning model uses the Convolutional Neural Network (CNN) that automatically extracts the features and classifies the image using the fully connected network. In this study, we proposed an ensemble deep learning-based approach for automatic binary classification of myeloma histology images, which can improve the early diagnosis of multiple myeloma patients. Also, we explored the capability of pre-trained VGG-16, VGG19, and Resnet-50 models with transfer learning for Multiple Myeloma image categorization and obtained a prediction accuracy of 99.77%, 97.92%, and 97.40%, respectively, and explained reasons that helped the accuracy increase.