White Blood Cells Classification Using MBOA-Based MobileNet and Coupling Pre-trained Models with IFPOA

The use of AI has greatly improved medical diagnostics, especially in the field of cancer. Diagnosing acute myeloid leukemia (AML) is a time-consuming process that can be mistaken by both humans and machines. Even after a thorough evaluation by a seasoned pathologist, it might be challenging to reach a definitive conclusion in some cases. While diagnosing AML may be a time-consuming and error-prone process, computer-aided diagnostics (CAD) can assist. One of the most important parts of diagnosing AML is finding white blood cells (WBCs), and most cutting-edge methods for doing so is deep learning. The quality of the features that are extracted and used to train the pixel-wise classification models has a strong correlation with the accuracy of WBC detection. Investigating the various patterns of change linked to WBC numbers and characteristics is crucial for CAD. To segment the nuclei for the second level, this research study employs two simultaneous convolutional neural networks using the MobileNet structure. Monarchy Butterfly Optimization Algorithm (MBOA) is employed for the purpose of fine-tuning the segmentation classifier. Next, a hybrid model utilizing ResNet and DenseNet is used for classification. The hyper-parameters are ideally determined using the Improved Flower Pollination Optimization Algorithm (IFPOA). Based on the cross-validation findings, the suggested model achieves an Area Under Curve (AUC) of 99% and a performance measure of around 94.17% across recall, precision, accuracy, besides F1-score. As an alternative to CAD tools, the suggested model can help pathologists evaluate white blood cells in blood smear pictures in the clinical laboratory.