A Novel, One Stop, Affordable, Point of Care and Artificial Intelligence Supported System of Screening, Triage and Treatment Selection for Cervical Cancer and Precancer in the Low-to-middle Income Countries
A Novel, One Stop, Affordable, Point of Care and Artificial Intelligence Supported System of Screening, Triage and Treatment Selection for Cervical Cancer and Precancer in the Low-to-middle Income Countries
Artificial intelligence (AI) is fast gaining reputation as a highly promising solution for cervical cancer screening. AI-based detection of cervical neoplasias is named automated visual exam (AVE) by the National Cancer Institute, USA. The investigators propose to develop and evaluate the performance characteristics of a novel AI system to both screen and triage women as well as help in treatment decision making. AI will analyse infrared spectroscopic signals derived from urine samples of unscreened women for the presence of high-risk human papillomavirus (hr-HPV). Our preliminary study has shown that spectroscopy can detect hr-HPV in urine. For screen-positive women the AI will interpret a set of cervical images captured with a high-quality devoted camera to detect high grade cervical precancers and cancers and to determine the type of transformation zone (TZ) (helps in treatment decision). The prototype device for image capture and the AI algorithms are already developed by us. The technologies will be further improved in part 1 (initial 2 years) and validated in part 2 (subsequent 3 years). During Part 1, the investigators will analyse urine samples collected from 1100 women at multiple screening clinics in Zimbabwe for the presence of hr-HPV using spectroscopy and use the signals generated to improve the AI algorithm. In this part the investigators will also assess the concordance between hr-HPV detection in urine samples using spectroscopy and cervical human papillomavirus (HPV) detection using a validated HPV test. The cervical image recognition device and the AI algorithm will be further improved during part 1 by collecting more images from hr-HPV positive and negative women. AI will also be trained to interpret the cervical images to determine the TZ type. In part 2 total 2100 women will be screened in Zimbabwe with AI-supported spectroscopic analysis of urine to detect hr-HPV and a validated HPV test to evaluate and compare their sensitivity and specificity to detect histology-proved high grade cervical precancers and cancers. The sensitivity and specificity of AI-supported detection of cervical neoplasias on cervical images will be evaluated to triage the HPV positive women. The accuracy of AI to determine TZ type will be compared with expert opinion. During the field validation part (part 2), the investigators will also conduct a cost analysis and compare cost of our approach to current standard Zimbabwean practice. The International Agency for Research on Cancer- World Health Organization WHO (IARC-WHO) has partnered with The Neo Sense Vector Company (NSV), Delaware, USA (industry), The Engineering Department, Lancaster University, Lancaster, UK and The University of Zimbabwe, College of Health Sciences, Harare, Zimbabwe to implement this study focusing on innovation that will greatly contribute to the global elimination of cervical cancer, a WHO priority.
Cervical cancer is a major public health challenge killing over 300,000 women annually at the most productive period of their lives and disproportionately affecting women in low- and middle-income countries (LMICs). Even in developed countries like the USA, the disparity between low- and high-income populations is striking. A cervical cancer death dramatically alters family and societal dynamics. In sub-Saharan Africa (SSA) for every 100 women who die from cervical cancer, 14 to 30 children die as an indirect consequence. Indeed, cervical cancer mortality is a real impediment to achieving WHO's Sustainable Development Goal of reducing premature mortality from non- communicable diseases (NCD) by a third before 2030. WHO have also recently adopted a resolution to eliminate cervical cancer globally. Whilst HPV vaccination will undoubtedly support this ambition for the next generation of girls, this vision also demands an effective screening and treatment programme. Yet current LMIC screening, investigation, and treatment regimes, especially are deeply flawed and not widely adopted.
The EASTER project aims to further develop and validate two new technologies for cervical cancer screening and diagnosis: (i) screening for human papillomavirus (HPV) in urine with spectroscopy, and (ii) diagnosis with artificial intelligence-assisted technology from the Neo Sense Vector Company (NSV), a private company. The project will recruit 3200 women and screen them for HPV.
The project will be implemented in two parts. Part 1. Technology improvement to achieve two key improvements.
Part 2. In the second part of the study, the investigators will test the functionality and effectiveness of the AI algorithms and devices developed through Part 1 in the same setting in Zimbabwe. The developed system of AI interpretation of urine samples will be evaluated as a screening test to detect cervical intraepithelial neoplasia grade 2 or worse (CIN2+) lesions and compared to a validated HPV detection test. The AI diagnostic accuracy of n-Gyn system to detect CIN2+ lesions based on captured cervical images will be evaluated as a triage test for HPV positive women in the detection of histopathologically confirmed CIN 2+ lesions.
Settings, procedures, and analysis:
The EASTER project (Part 1 and Part 2) will be implemented in two screening polyclinics (Epworth and Mbare) in Harare where women are routinely screened (with an average of 15% of women living with HIV (WLHIV) participants). Women aged 25-49 who agree to participate and sign the corresponding Institutional Review Board (IRB) approved consent forms will be requested to provide two self-collected samples, 1) a first void urine sample and 2) a self-collected vaginal sample. Recruitment specimens will be tested for HPV with Ampfire. Women HPV positive in either sample will be referred to colposcopy for disease ascertainment.
The colposcopist will examine the cervical images on the n-Gyn screen and independent of AI will document the visibility and location of the squamocolumnar junction (SCJ), type of transformation zone (TZ), Swede score, most appropriate site for taking biopsy (if any abnormalities are present) and suitability for treatment by ablation. Sequential images of the transformation zone will be obtained before and after cleaning with normal saline and then after applying 5% acetic acid for one minute. Appropriate magnifications will be used to enable delineation of the SCJ and to identify the worst area of suspected abnormality. A final image will be captured after the application of Lugol's iodine. After the images have been collected, the clinician will take at least one punch biopsy from the most abnormal site determined by him/her. If no lesion is visible, biopsies will be obtained from the 6 and 12 o'clock positions closest to the SCJ. All histopathology slides will be examined by pathologists at Lancet Laboratories in Harare. The patient will be managed based according to the local management protocol.
Women will be managed according to clinical coloscopy diagnosis: women without visible lesion will exit the study at this point. Women with visible lesions will be treated with thermal ablation if eligible or large loop excision of the transformation zone (LLETZ) if needed and exit the study. Women diagnosed with cancer will be referred to the regular system for appropriate management and exit the study.
Data management and study supervision will be the responsibility of the International Agency for Research on Cancer (IARC) and the local Principal Investigators, who are experienced HPV researchers.
The outcome of primary interest for the statical evaluation will be histologically confirmed cervical intraepithelial neoplasia grade 2 or worse (CIN2+), including CIN2 lesions positive for p16. For Part I, our analyses will focus on the agreement between hr-HPV detection by spectroscopic analysis of urine and by Ampfire HPV, which will be tested using the Cohen's kappa statistic. Spectroscopy will be deemed as good as HPV testing in defining screen results if a kappa of 0.7 (80% agreement) is achieved. For Part 2, standard formulations will be used to calculate the test performance characteristics (sensitivity/specificity). For the comparison of the performance characteristics of the screening tests, if ẟ is the hypothesized relative sensitivity (or specificity), the equivalence of the two tests will be inferred if the true relative risk to be within the interval (ẟ to 1/ẟ). The test of proportions will be used to assess if the performance characteristics of the triage tests are not different from the hypothesized value.
Inclusion Criteria:
Exclusion Criteria:
basup@iarc.who.int+33764485370
bothwellguzha@gmail.com