Deep Learning-Based Diagnosis of Leukemia in Africa: Case of Acute Lymphoblastic and Acute Myeloid Leukemia in Senegal

Early and accurate diagnosis of leukemia in Africa, particularly for children, remains a major public health challenge. In the current paper, we propose an application of deep learning (DL) and computer vision approaches in order to improve the diagnosis of acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML). We rely on two models, namely ViT and YOLOv8, that have been trained, evaluated on public open dataset (C-NMC dataset [1]) and designed for identifying acute lymphoblastic leukemia cell from microscopic images of immature blood cells. The results of this experimental evaluation reveal that YOLOv8 gives better performance in the context of limited data and resources. We then selected YOLOv8 pre-trained models and adopted a transfer learning approach on augmented data we collected from the Senegalese context of the Dalal Jamm Hospital. The results obtained present a precision of 96.6%, a recall of 92.9%, and an mAP50 of 98.6%. These encouraging performances demonstrate the ability of YOLOv8 to accurately identify ALL and AML cells.