NCT05281939

Multi-center Application of an AI System for Diagnosis of Cervical Lesions Based on Colposcopy Images

Multi-center Application of an Artificial Intelligence System for Automatic Real-time Diagnosis of Cervical Lesions Based on Colposcopy Images

• Status: Unknown
• Phase: N/A
• Study type: Interventional
• Enrollment: 10,000 (estimated)
• Sponsor: Fujian Maternity and Child Health Hospital · Academic / Other
• Sex: Female
• Age: 18 Years
• Healthy volunteers: Accepted

Summary

The application of artificial intelligence in image recognition of cervical lesions diagnosis has become a research hotspot in recent years. The analysis and interpretation of colposcopy images play an important role in the diagnosis，prevention and treatment of cervical precancerous lesions and cervical cancer. At present, the accuracy of colposcopy detection is still affected by many factors. The research on the diagnosis system of cervical lesions based on multimodal deep learning of colposcopy images is a new and significant research topic. Based on the large database of cervical lesions diagnosis images and non-images, the research group established a multi-source heterogeneous cervical lesion diagnosis big data platform of non-image and image data. Research the lesions segmentation and classification model of colposcopy image based on convolutional neural network, explore the relevant medical data fusion network model that affects the diagnosis of cervical lesions, and realize a multi-modal self-learning artificial intelligence cervical lesion diagnosis system based on colposcopy images. The application efficiency of the artificial intelligence system in the real world was explored through the cohort, and the intelligent teaching model and method of cervical lesion diagnosis were further established based on the above intelligent system.

Detailed description

Based on previous studies and clinical practice, this study carried out a multi center application in Fujian Province, China. In this study, Fujian Maternity and Child Health Hospital and Mindong Hospital of Ningde City were included, with a total of 10000 participants who have undergone colposcopy examination were enrolled. In the first place, the investigators will build a multimodal artificial intelligence diagnostic system by combining colposcopy images with other non-image data, such as the results of HPV tests and Thinprep cytologic test (TCT) and so on. And then, use standardized colposcopy images and non-image medical data of cervical lesions in different medical institutions to verify the efficacy of the multimodal intelligent diagnostic system for cervical lesions. What's, more, the investigators will establish artificial intelligence cohorts (assisted by intelligent systems) and traditional physician cohorts (assisted by expert, senior and primary physicians) to contrast the diagnosis results of the multimodal artificial intelligence diagnostic system and different levels of colposcopy doctors. And can also bidirectionally analyse the diagnostic efficacy and differences of the system and colposcopy physicians of different levels, and evaluate the performance of this diagnostic system for real-world applications.

Conditions

• Artificial Intelligence
• Colposcopy
• Cervical Lesions
• Image

Interventions

Type	Name	Description
DIAGNOSTIC_TEST	Artificial intelligence diagnosis	Participants were divided into the intervention group and the control group using a random number table. The intervention group participants' cervical colposcopic image data and non-image data as follow：age, the infection of high-risk human papillomavirus (HR-HPV),the type of HR-HPV infection,the duration of HR-HPV infection, cervical cytology (TCT) results, HIV/sexually transmitted infection history, marriage and childbearing history,first sexual life history, sexual partner history, smoking history,oral contraceptives history,the use of immune drug and possible clinical symptoms of cervical lesions such as postcoital bleeding, abnormal vaginal secretions, vaginal bleeding symptoms, etc.

Timeline

• Start date: 2021-08-01
• Primary completion: 2024-08-01
• Completion: 2024-09-01
• First posted: 2022-03-16
• Last updated: 2023-11-18

Locations

5 sites across 1 country: China