Detection of acute lymphoblastic leukaemia using extreme learning machine based on deep features from microscopic blood cell images

Leukaemia is the medical term for blood cancer. This paper proposes an automatic disease diagnosis model to detect leukaemia from microscopic blood cell images by classifying these images into malignant and benign cells. It uses extreme learning machine (ELM) as the classifier and uses the transfer learning on AlexNet to obtain the 4,096 features required to train the classifier. The training of AlexNet is performed on 864 and 2,080 images, obtained after augmentation. The experiments are repeated five times each for nine different values of 'number of hidden neurons' in the hidden layer of the classifier, to obtain nine average accuracies. The best average accuracy obtained for IDB1 is 99.4% at 3,000 and 4,500 hidden neurons, while for IDB2, it is 99.8% at 3,500 hidden neurons. The grand average is calculated over these nine averages and is found to be 98.6% and 99.2% for IDB1 and IDB2 respectively, while obtaining best accuracy as 100% for both the datasets.