AI-assisted cervical lesions assessment and classification based on optical colposcopy

Cervical cancer is a malignant tumor with a clear etiology and is preventable and controllable. Early screening and detection are essential for effective treatment. In the current cervical cancer screening process, abnormalities detected by liquid-based cytology or HPV screening usually require colposcopic guidance for further biopsy, with histologic diagnosis as the gold standard for confirming the diagnosis. However, there is a shortage of experienced and well-trained colposcopists, especially in less developed areas such as counties and communities. In this study, a real clinical database was constructed based on 602 cases of cervical lesions provided by Fujian Maternity and Child Health Hospital. Then, the CNN network was used to construct an AI-assisted diagnosis model for the optical colposcopy, which realized a fourclassification diagnosis of cervical lesions. The evaluation by quantitative metrics showed that the average accuracy, precision, recall and F1 scores for the four-classification of the proposed model were 81%, 84%, 79% and 92%, respectively. The results show that the developed AI-assisted colposcopy model has good lesion classification effect, which improved the accuracy of the diagnosis of cervical lesions, and reduced the risk of misdiagnosis and omission of diagnosis due to the inexperience or subjective judgment of doctors, so as to provide powerful technical support for the precise diagnosis and treatment of cervical lesions in the early stage.