Detection of Leukemia Cancer in Microscopic Blood Samples Using Hybrid-Based Classification Paradigm

Leukemia cancer detection in microscopic blood samples (MBS) involves the method of identifying the presence of leukemia cancer cells in a blood sample using a microscope. This method is done by identifying the abnormal characteristics of the cancer cells in the blood sample, such as size, shape, and structural changes. This creates one of the problems in leukemia cancer detection using MBS, whereby it is usually very subjective and dependent upon the expertise of the person examining the blood sample. This can result in misdiagnosis and mistreatment. In this regard, this research work will focus on developing a new method to detect leukemia cancer. The proposed model shall have five main stages: preprocessing, segmentation, feature extraction, feature selection, and early detection and diagnosis. First, the raw blood samples collected will be preprocessed using a Harmonic Trimmed Weighted Mean Filter with Decision-based Unsymmetrical Mask and Dualistic Sub-Image Histogram Equalization. The preprocessed samples will then be segmented with a weight-optimized U-Net. Segmentation-based feature extraction includes texture features from the Gray-Level Co-occurrence Matrix (GLCM) and Local Phase Quantization (LPQ) methods; color features from Color Moment; and shape features from Chain Codes and Curvature Scale Space. A novel Hybrid Remora Backtracking Search Optimization (HRBSO) model has been proposed for the selection of the best optimal features under the approach. The recommended method is predicated on modeling the early identification and treatment stage with a Hybrid Deep Learning (HDL) classifier comprising a Recurrent Neural Network (RNN), a Convolutional Neural Network (CNN), and an Attention Layer. The optimization of CNN is done for the weight and bias hyperparameters using a newly designed hybrid optimization approach called HRBSO. Results show the feasibility and excellence of the developed model, which has been implemented in MATLAB.