Performance Analysis for Optimized Light Weight CNN Model for Leukemia Detection and Classification using Microscopic Blood Smear Images

Abstract

The objective of this work is to create a diagnostic tool for the early diagnosis of leukaemia which is a serious type of cancer affecting bones and blood. Acute lymphoblastic leukemia (ALL) is the most dangerous form of leukemia. Doctors diagnose it by blood samples under powerful microscopes with enhanced lenses which can be slow and is sometimes affected by disagreements among experts. Therefore, the purpose of this work was to create a profound diagnostic tool for the early diagnosis of leukaemia.We proposes an Optimized Light Weight CNN to detect ALL at the early stage. Fragmentation and classification based on preprocessing are the two main components of the suggested method. Artificial images are created during the segmentation process and then tamed by chromatic modification. The proposed model is used to extract the best deep features from every blood smear image to predict the presence of ALL. The work was tested by two lymphoblastic leukaemia image databases (ALL_IDB1 and ALL_IDB2). Deep-learning (DL) models-based segmentation and classification techniques have recently been introduced for detecting ALL; however they still have certain drawbacks. The proposed approach was assessed with few DL parameters like accuracy, F1 score, precision, recall and area under the curve. In comparison to the most recent research studies already published; the suggested strategy produced exceptional classification accuracy as 99.56%, F1 score as 99.53%.