Comparison of segmentation models used to identify mature and immature lymphocytes in blood film -

Abstract

Acute Lymphocytic Leukemia (ALL) affects about 75 thousand people a year, among these 80% are children, being a highly invasive and fatal disease, rapid diagnosis is of great importance, traditional diagnostic methods are expensive and time-consuming, therefore, the use of image segmentation methods using artificial intelligence can help detect the elements of interest in blood slides, the lymphoblasts. This work compared the models: Segnet, Mobilenet Segnet, Vgg Segnet, Resnet50 Segnet, Vgg Unet, Resnet50 Unet, Mobilenet Unet, FCN 8, FCN 32 and FCN 32 Mobilenet, by pixel precision and execution time. The ALL-IDB database was used, containing blood slides from healthy patients and possible ALL. As a result, it was observed that MobileNet networks performed better, among them, Mobilenet Unet stood out, where the result of the mean average precision of the classes was 83.4%.