Enhancing Cervical Vascular Pattern Segmentation Based on Mathematical Morphology

The signs of cervical pathology are often the appearance of atypical epithelium, especially with an abnormal vascular pattern or low or high-grade squamous intraepithelial lesions, depending on the seriousness of the injury. In this paper, we focus on developing an algorithm to enhance the blood vessels and segmentation of blood vessels from the cervical surface, supporting the classification of the degree of pathology. Cervical vascular images were collected under polarized light to help reduce reflections from the surface to support more efficient image processing steps. Then, the pre-processing algorithms based on mathematical morphological and blood-absorption-based color channel combining are applied to eliminate noise and enhance vascular contrast on the cervical surface. Finally, the Sauvola threshold is applied to vascular segmentation. The result of the proposed method shows a high average sensitivity of 96% and a Jouden index of 70%, giving better segmentation results than CLAHE and the method using only morphology algorithms.