Optimal Attention based Deep Learning with Segmentation Approach for Automated Leukemia Detection and Classification

Abstract

Leukemia is a deadly disease that compromises the exists of numerous people worldwide. Leukemia does not form strong cancers, yet constitutes a considerable amount of anomalous white platelets that cluster out the typical platelets. Hematologists analyze blood smears from people to detect this anomaly appropriately. The method utilized for diagnoses is impacted by various factors such as the level of weariness and the hematologist's experience, which leads to inaccuracies and nonstandard results. Automatic analysis of acute lymphoblastic leukemia (ALL) is a vital and difficult task. Presently, machine learning (ML) and deep learning (DL)-based identification have become a commendable means in medical image analysis. DL algorithm is mainly utilized in leukemia treatment, for identifying if leukemia exists in a patient. In this work, we propose an Optimal Attention-based DL with Segmentation for Automated Leukemia Detection and Classification (OADLS-ALDC) technique. The OADLS-ALDC technique primarily undergoes image pre-processing in two phases: Wiener Filter (WF) and Dynamic Histogram Equalization (DHE). At the same time, the Watershed Segmentation approach is applied to segment the pre-processed images. Next, the OADLS-ALDC technique uses EfficientNetV2M for the identification of useful feature vectors. Moreover, the recognition and identification of leukemia takes place by the use of the Attention Convolutional Recurrent Neural Network (ACRNN) model. Furthermore, the hyperparameter selection of the ACRNN model is performed using a root mean square propagation (RMSProp) Optimizer. To highlight the improved detection outcomes of the OADLS-ALDC system, a sequence of experiments was involved. The experimental values identified that the OADLS-ALDC approach gets better performance than other techniques.