A Portable System for Screening of Cervical Cancer
Loading...
Authors
Cepeda Andrade, Paloma
Issue Date
2023
Type
Dissertation
Language
Keywords
Cervical Cancer , Colpocsopy , K-Means , Portable Colposcope , Segmentation , Specular Reflection
Alternative Title
Abstract
Cervical cancer is one of the most common cancers that affect women, with the highest incidence and mortality rates occurring in low- and middle-income countries. Early detection is crucial for successful treatment, but the need for expensive equipment, trained colposcopists, and clinical infrastructure has made it difficult to eradicate this disease. Accurately determining the size and location of a precancerous lesion involves specialized and costly equipment, making it difficult to track the progression of the disease or the efficacy of treatment. Imaging and machine learning techniques have been attempted by several researchers to overcome these limitations, but the subjective nature of diagnosis and other challenges persist. Therefore, there is a need to develop a system to automatically segment lesions on the cervix and quantify their size in relation to the cervical region of interest. Challenges to the automated detection of cervical cancer include:• Low quality of the devices used, which impair the image resolution; lighting conditions, which can make shadows appear, hindering the ability to find the cervical region of interest (ROI); distortion of the images due to the presence of glare or specular reflections (SR) from the light source; and the appearance of artifacts such as the speculum and surrounding tissue. The limitations that exist in selecting or designing a device to acquire cervical images (cervigrams) have been investigated. • The acquisition of cervical images requires access to sensitive patient information, which raises concerns about patient privacy and data security. Ensuring that patient data is protected and used only for diagnostic purposes is critical to building patient trust and ensuring widespread adoption of automated screening technologies. A pilot study to capture cervigrams from women that present early signs of cervical cancer was designed. Relevant data would be collected to further understand the progression of this disease, while maintaining privacy and confidentiality of the participants in the study. • The early detection of cervical cancer requires analyzing complex data, including images, pathology reports, and medical records. Automating the analysis of this data requires machine learning algorithms or image processing techniques capable of interpreting such information. Image processing methods based on traditional and machine learning techniques were leveraged to identify the cervical region of interest and remove light reflections from the cervical epithelium. Lesions present on the cervix were detected and their size, invariant with respect to the orientation of the camera or its distance from the cervix, was calculated. • Finally, variability and subjectivity are involved when acquiring and analyzing cervigrams. A graphical user interface was developed to facilitate data collection and analysis throughout the pilot study and future clinical trials. Results indicate that it is possible to segment images of the cervix, reduce the effect of glare from light sources, remove specular reflections and other artifacts, and successfully detect and quantify lesions through the proposed methods. The above approaches are demonstrated throughout this dissertation to show that a low-cost bioinformatics-based tool for early detection of cervical cancer can be achieved for screening patients in a clinical setting. While the algorithms used for analysis were validated using sample images from public databases, it is crucial to conduct small-scale clinical trials to further validate these methods. Furthermore, the use of more advanced image processing techniques or machine learning algorithms to improve the accuracy and speed of lesion detection is under review.
Description
Citation
Publisher
License
Creative Commons Attribution 4.0 United States