Abstract
BACKGROUND: Cervical cancer remains a leading cause of mortality for women in low-income countries. Therefore, there is an urgent need for a point-of-care cervical cancer screening test for community-based settings that can be used for triaging and managing women with preinvasive and early invasive lesions of cervix. METHOD: A cross-sectional study was conducted to evaluate the diagnostic accuracy of an indigenous artificial intelligence (AI)-enabled trans-vaginal colposcopic device called Smart Scope®. Parallel screening with cervical cytology and assessment of magnified digital images was done by a trained observer to minimize false negatives. Cervical biopsy was taken for any abnormality identified on any of the three tests. Histopathology report of the biopsy was taken as the reference standard. RESULT: A total of 268 women were recruited for the study. Cervical evaluation with the Smart Scope® identified 17.5% (n = 47) cervices with pre-cancerous/cancerous lesions, 19.4% (n = 52) with high-risk lesions, 14.2% (n = 38) with inflammatory/benign lesions and 48.9% (n = 131) cervices as normal. Taking cervical biopsy as the reference standard, sensitivity, specificity, PPV and NPV were 68.4%, 65.5%, 13.1%, and 96.4% respectively. In comparison, the sensitivity, specificity, PPV and NPV of cervical cytology were 43.8%, 93.3%, 46.7%, and 92.5% respectively, while that of assessment of magnified digital images by trained observer was 89.5%, 49.1%, 22.4%, and 96.6% respectively. CONCLUSION: Smart Scope®-based evaluation in conjunction with the assessment of magnified digital images by a trained observer can be an effective tool for cervical cancer screening. Further improvement of AI-based diagnosis can make it an effective tool for community-based screening.