The Value of TruScreen (An Artificial Intelligence Cervical Cancer Screening System) in High-Risk HPV Positive Patients

Abstract

Background: To investigate the value of the artificial intelligence cervical cancer screening system TruScreen (TS) in high-risk human papillomavirus (HPV)-positive patients in a clinical setting. Methods: Three hundred and eighteen patients positive for high-risk HPV in the gynecological clinic of our hospital from May 2020 to June 2021 were analyzed retrospectively. Colposcopy was performed when it was clinically indicated. Results: Among the 318 patients, 203 were TS negative and 115 were TS positive, of whom 84 were referred for colposcopy and possible biopsy. Among the 318 patents, 74.53% (237/318) had a single type of HPV infection, and 25.47% (81/318) had more than two types of HPV infection. In terms of HPV types, the top 5 types were 52, 58, 51, 56 and 16. HPV52 accounted for 27.4% (87/318), followed by HPV58, accounting for 17.30% (55/318). A total of 84 patients underwent colposcopy. The negative predictive values of TS and thinprep cytology test (TCT) screening for cervical cancer and precancerous lesions were 33.33% and 16.90%, respectively. The positive predictive values were 88.41% and 92.31%, respectively. The sensitivity was 85.92% and 16.90%, respectively. The specificity was 38.46% and 92.31%, respectively. Among 251 non-16/18 high-risk HPV-positive patients with TCT negative intraepithelial lesion or malignancy (NILM) 49 underwent colposcopy. The positive predictive value of TS for cervical cancer and precancerous lesions was 84.78% and the sensitivity was 92.86%. Conclusions: This study demonstrated that in a clinical setting, TS had a better sensitivity than TCT in cervical cancer screening, but less specificity than TCT. In the non-16/18 HPV-positive population with TCT NILM, TS screening can be considered as having potential for clinical management. However, the current research sample size was small, and further research needs to be performed with a larger sample size.