Abstract: Leukemia is a blood cancer which is the most prevalent childhood cancer type and accounts for approximately 33% of all paediatric cancer. Some cases have occurred previously where zero symptoms are shown by blood until the disease has progresses and reached a dangerous level. This kind of a case mostly causes a misdiagnosis. To address such problem, this paper introduces Convolutional Neural Networks (CNN), which has led to break-through results in computer vision, and is thus a major technique that can be used to solve the Leukemia Classification Challenge. The ability of discovering abstract features with the capability of discrimination of different aspects of interests is possessed by the CNN algorithm. A diagnosis of the disease at early stage leads to an effective treatment. Segmentation from microscopic images has been performed so that the cells represent images in real world. The task is to identify the leukemic blasts at a premature stage. The dataset consists of 15,114 images (Training data = 10,661 images; Validation data = 1,867 images; Testing data = 2,586 images). The proposed method achieves accuracy up to 99% based on the number of epochs and data split.

Keywords: Leukemia, Classification, Convolutional Neural Networks (CNN), Transfer Learning

| DOI: 10.17148/IARJSET.2021.8344