Digital medical images are used in healthcare for image processing and machine learning, allowing computers to analyze various phenomena. Traditional microscopic image segmentation by hematologists is labor-intensive, repetitive, and costly. A crucial stage in hematology imaging is the detection and segmentation of white blood cell nuclei, which serves as a foundation for deep learning to assist in diagnosing many blood-related diseases. This study proposes a novel, supervised segmentation method by hybridizing threshold with Mahalanobis distance, offering an accurate and lightweight solution validated across three diverse WBC image datasets. Thresholding is applied to a training image to distinguish between white blood cell nuclei and the surrounding background, thereby creating supervised learning datasets. Subsequently, the Mahalanobis distance technique is employed to automatically and efficiently segment the nucleus from the background in other test images using the established supervised data. This novel method is compared against traditional thresholding technique as well as other widely used clustering methods, including hierarchical clustering and k-means clustering for performance evaluation. The segmentation processes were applied to five distinct types of white blood cells: neutrophils, eosinophils, basophils, monocytes and lymphocytes, under three varying image conditions sourced from different databases. The performance evaluation results show that the proposed method outperforms the other three alternative techniques in two of the three databases. In contrast, the thresholding technique exhibited the shortest execution time among all the methods evaluated. Nevertheless, when assessing the visual segmentation results, it is evident that the proposed method improved the accuracy of the image region of interest.
Ismail et al. (Wed,) studied this question.