An Optimized Deep Learning Framework for Classification and Detection of Acute Lymphoblastic Leukemia (ALL)

Abstract

Acute Lymphoblastic Leukemia (ALL) a medical threat which principally affects the pediatric population, demands early diagnosis for fruitful treatment results. Deep learning models have set a really high standard in medical imaging by providing extensive applications for the identification and categorization of ALL. Over time, convolutional neural networks (CNN) have made significant stature in medical image analysis, especially when it comes to detection and categorization of ALL. The proposed study presents a pioneering classification model, XIncept-ALL, which benefits from the transfer learning technique. Furthermore, by improving generalization with incorporation of valuable traits extracted a diverse dataset, the projected model mitigates the concerns of overfitting with the help of two pre-trained models, Inception and Xception, both of which are trained on the widespread ImageNet dataset. Additionally, through merger of distinctive features extracted from both models, the proposed system is expected to refine feature representation and overall performance by using feature fusion. Nevertheless, the anticipated model proves itself as a unique and remarkable framework with not only an exceptional accuracy of 99.0% on the ALL dataset, but also the capability to classify various ALL subtypes (Benign, Earl, Pre-Acute and Pro-Acute) with unequalled correctness within a single classifier. Our results show the superior performance of XIncept-ALL for accurate and rapid effective classification of acute lymphocytes images, thereby highlighting its competence as an efficient diagnostic tool. Furthermore, this model has the potential to be employed in diverse medical imaging contexts, beyond leukemia.