Automated Lesion Grading and Analysis of Uterine Cervix Images For Cervical Cancer Diagnosis

Abstract

In order to accurately assess the quality of reconstructed images, it is necessary to employ consistent methods for the extraction of information and the application of metrics. The quantitative analysis of regions in medical images that include a certain tissue can be made possible by partitioning the images into the many types of tissues. The evaluation makes it much simpler to evaluate the capability of an imaging algorithm to reconstruct the properties of a variety of tissue types and find anomalies. To aid in the diagnosis of cervical neoplasia, a magnified visual examination of the uterine cervix using a colposcope, which is a low-power, stereoscopic, binocular field microscope with a bright light source, is performed using a colposcope. Positive screening tests, such as positive cytology and positive visual inspection with acetic acid, are the most typical indicator that a patient should be referred for a colposcopy. This publication outlines a reliable and adaptable method of medical image segmentation that can be used to divide uterine cervix image images into different types of tissue in order to make the process of evaluating uterine cervix images more straightforward. The strategy utilises a combination of statistical methods as well as an adaptive k-means algorithm. When compared to other methods, such as a threshold-based segmentation method, the most significant benefit of utilising the algorithm is that it enables this quantitative analysis to be performed without any prior preparation. In addition to this, it is applicable to circumstances in which there is a lack of data that may be used for supervised learning. In order for the segmentation technique to be successful, it is necessary to be able to differentiate between different types of tissue across a wide range of image qualities. This ability is demonstrated in the second half of the study. Through the process of segmenting pictures into regions of interest according to the different types of tissue, the surface of the cervix is broken up into a number of distinct tissue areas. A colour score-based metric is utilised in order to differentiate between regions that have been retrieved from reconstructed images and ground truth models. The quantitative data provide evidence of the accuracy with which aberrant and normal tissue qualities can be distinguished. 200 standardized colposcopic images are examined as part of our research. The findings of the segmentation are compared and contrasted with the assistance of structural segmentation accuracy, specificity, sensitivity, and positive predictive value. The results of our tests indicate that the approach has an accuracy of 89.35%, a specificity of 84%, a sensitivity of 97.45%, and a precision of 81.25% accordingly. The effectiveness of the method in clinical segmentation has been demonstrated. According to the findings of this study, the use of the K-means algorithm for cervical regional segmentation of colposcopic photos based on HSV colour space has significant clinical value and can assist medical professionals in the detection of cervical cancer.