White Blood Cell Segmentation Using DeepLabv3+ for Improved Hematological Disease Detection

The morphological analysis of the White Blood Cell (WBC) plays a crucial role in disease diagnosis using medical image analysis, particularly in treating haematological disorders like leukemia, lymphoma, anemia, and sickle cell disease. Clinical medical image analysis is manual and susceptible to subjectivity and human error. Therefore, deep learning-based methods are used to segment WBCs from microscopic blood smear images. This paper introduces a new DeepLabv3+-based WBC segmentation approach. The performance of the proposed approach is evaluated using the science bowl challenge dataset. The dataset is augmented using traditional techniques such as centre cropping, rotation, grid distortion, horizontal flipping, and vertical flipping. The performance of the proposed approach is measured using the parameters such as dice coefficient, Intersection of Union (IoU), precision, and recall. The results are compared with the results of traditional U-Net-based WBC segmentation. The experimental results show that training loss, training dice coefficient, validation dice coefficient, training IoU, validation IoU, precision and recall of the proposed approach are increased by 23.78%, 2.32%, 0.43%, 4.29%, 0.78%, 0.96%, 1.01%, respectively compared to U-Net-based WBC segmentation. Thus, DeepLabv3+-based WBC segmentation outperforms U-Net-based WBC segmentation.