A Modified AlexNet Approach for Multi-Class Classification of Blood Cancer

Blood cancer affects the production and operation of normal blood cells. To accurately detect and classify this disease, a Computer-Aided Diagnostics system can be a great alternative for expert hematologists. This work aims to detect five types of blood cancer such as four subtypes of the most serious kind of blood cancer, Leukemia and Multiple Myeloma from microscopic blood smear images using modified AlexNet model, a deep convolutional neural network. A dataset generated from three different online sources has been used to train, validate and test the modified model. The validation accuracy, test accuracy, precision, recall, specificity, and F1Score obtained from this model are 94.17%, 98.96%, 98.91%, 98.99%, 99.67%, and 98.95% respectively. By comparing these performance evaluation parameters with other deep learning models' performance, savings in training time and more accurate results are found in the case of the custom AlexNet model.