Automatic segmentation and classification of blood components in microscopic images using a fuzzy approach

ABSTRACT

INTRODUCTION: Automatic detection of blood components is an important topic in the field of hematology. Segmentation is an important step because it allows components to be grouped into common areas and processed separately. This paper proposes a method for the automatic segmentation and classification of blood components in microscopic images using a general and automatic fuzzy approach. METHODS: During pre-processing, the supports of the fuzzy sets are automatically calculated based on the histogram peaks in the green channel of the RGB image and the Euclidean distance between the leukocyte nuclei centroids and the remaining pixels. During processing, fuzzification associates the degree of pertinence of the gray level of each pixel in the regions defined in the histogram with the proximity of the leukocyte nucleus centroid closest to the pixel. The fuzzy rules are then applied, and the image is defuzzified, resulting in the classification of four regions leukocyte nuclei, leukocyte cytoplasm, erythrocytes and blood plasma. In post-processing, false positives are reduced and the leukocytes (including the nucleus and cytoplasm), erythrocytes and blood plasma are segmented. RESULTS: A total of 530 microscopic images of blood smears were processed, and the results were compared with the results of manual segmentation by experts and the accuracy rates of other approaches. CONCLUSION: The method demonstrated average accuracy rates of 97.31% for leukocytes, 95.39% for erythrocytes and 95.06% for blood plasma, avoiding the limitations found in the literature and contributing to the practice of the segmentation of blood components.