A convolutional neural network-based learning approach to acute lymphoblastic leukaemia detection with automated feature extraction

Leukaemia is a type of blood cancer which mainly occurs when bone marrow produces excess white blood cells in our body. This disease not only affects adult but also is a common cancer type among children. Treatment of leukaemia depends on its type and how far the disease has spread in the body. Leukaemia is classified into two types depending on how rapidly it grows: acute and chronic leukaemia. The early diagnosis of this disease is vital for effective treatment and recovery. This paper presents an automated diagnostic system to detect acute lymphoblastic leukaemia (ALL) using a convolutional neural network (CNN) model. The model uses labeled microscopic blood smear images to detect the malignant leukaemia cells. The current work uses data obtained from the Acute Lymphoblastic Leukaemia Image DataBase (ALL_IDB) and performs various data augmentation techniques to increase the number of training data which in effect reduces the over-training problem. The model has been trained on 515 images using a fivefold validation technique achieving an accuracy of 95.54% and further tested on the remaining 221 images achieving almost 100% accuracy during most of the trials, maintaining an average of 99.5% accuracy. The method does not need any pre-processing or segmentation technique and works efficiently on raw data. This method can, hence, prove profitable for pathologist in diagnosing ALL efficiently.