Leukemia Net: Integrating attention depth wise Separable network-aided stacked feature pooling with weighted recurrent neural network-based leukemia detection model

Leukemia is regarded as one of the fatal disorders that develop in the bone marrow due to uncontrolled changes in the White Blood Cells (WBC). Leukemia usually causes abnormal functioning of WBC and also affects the bone marrow and blood of the human body. However, the classical leukemia identification framework consumes more time, and their detection precision is related based on the working capability. These existing challenges make this work to develop a novel structured method for accurately detecting the leukemia infection in blood cells. In the beginning, the leukemia images are gathered from online sources. The obtained images are segmented using the Multiscale Res-UNet++framework. Then, the segmented images are used for acquiring the features using Attention-based Depth Wise Separable Networks (ADSN), where ResNet and DCNN are integrated. From the extracted features, the target-based stacked features are obtained. Finally, the Weighted Recurrent Neural Network (WRNN)-based model is utilized for detecting leukemia. Here, the hybrid optimization algorithm, called Adaptive Fitness-based Cuckoo Search Artificial Rabbits Optimization (AF-CSARO), is used to select optimal features from the extracted feature and also to optimize the parameters in developed ADSN for enhancing the detection rate. Several analyses are done over the developed leukemia detection model to verify the efficiency of the recommended framework against classical techniques. The empirical outcome of the designed model shows that the developed model has attained 94 % in accuracy analysis and 97 % in precision analysis for detecting leukemia.