Researchers in Japan developed an artificial intelligence (AI)-based diagnostic tool for colposcopy examinations that can accurately identify cervical intraepithelial neoplasia (CIN) - abnormal cells found on the surface of the cervix that may become cancer and spread to nearby normal tissue - and suggest appropriate biopsy sites.

The research will be presented at the 2023 American Society of Clinical Oncology (ASCO) Breakthrough Meeting, taking place August 3-5 in Yokohama, Japan.

“Currently, there is no certification system for performing colposcopies in Japan, and the quality and interpretation of these examinations varies. Our study aimed to develop an artificial intelligence (AI)-based tool that reproduced colposcopy examination techniques of specialists to be used as a diagnostic aid by accurately identifying CIN lesions and guiding tissue sampling locations,” said Akihiko Ueda, MD, a physician in the Department of Gynecology and Obstetrics at Kyoto University.

To validate the accuracy of this technology, the researchers performed a retrospective analysis of 8,341 patients who had a colposcopy examination for secondary screening of abnormal cervical cytology or follow-up of CIN between 2013-2019.

Patients in the study were a median age of 41 years and were diagnosed with 7 early-stage cervical cancer cases, 203 CIN3 cases, 276 CIN2 cases, and 456 CIN1 cases. 

Researchers constructed the AI-based tool to detect lesions by annotating abnormal colposcopy findings after acetic acid processing in cervical cancer and CIN3 cases for which diagnoses were confirmed by biopsies.

The resulting detection model was applied to CIN1 and CIN2 cases, and the diagnostic accuracies of these lesions were evaluated by sensitivity, specificity, and area under the curve (AUC: an evaluation metric that takes a value between 0 and 1, with 0 indicating low accuracy and 1 indicating high accuracy), as well as the number of lesions identified. 

The model identified severe lesions in CIN3 cases with a sensitivity of 85%, a specificity of 73%, an AUC of 0.89 for lesion area, and an accuracy of 95% for the number of lesions identified.

The model predicted abnormal colposcopy findings in CIN1 cases with a sensitivity of 87%, a specificity of 70%, an AUC of 0.81 for lesion area, and an accuracy of 97% for the number of lesions identified.

The model predicted abnormal colposcopy findings in CIN2 cases with a sensitivity of 86%, a specificity of 67%, an AUC of 0.81 for lesion area, and an accuracy of 93% for the number of lesions identified.

“Colposcopy plays an important role in cervical cancer screening. This study showed that harnessing the power of artificial intelligence in cancer screening could pave the way for a potentially more effective and improved diagnostic performance in cervical cancer care,” said Roselle B. De Guzman, MD, ASCO Expert.

According to the authors, there is room for improvement in the application’s ability to accurately predict histopathological diagnosis, and the relationship between chronological changes in abnormal colposcopy findings and histopathological diagnosis needs to be investigated.

Source: ASCO