A segmentation model to detect cevical lesions based on machine learning of colposcopic images

• DOI: 10.1016/j.heliyon.2023.e21043
• Publication Date: 2023-10-20T07:20:32Z
• Authors: Zhen Li; Chu-Mei Zeng; Yan-Gang Dong; Ying Cao; Li-Yao Yu; Hui-Ying Liu; Xun Tian; Rui Tian; Chao-Yue Zhong; Ting-Ting Zhao; Jia-Shuo Liu; Ye Chen; Li-Fang Li; Zhe-Ying Huang; Yu-Yan Wang; Zheng Hu; Jingjing Zhang; Jiu-Xing Liang; Ping Zhou; Yi-Qin Lu
• Keywords: Social sciences (General); H1-99; Artificial intelligence; Q1-390; Segmentation; Science (General); HSIL; Colposcopy; Diagnosis; 3. Good health; Research Article

Semantic segmentation is crucial in medical image diagnosis. Traditional deep convolutional neural networks excel classification and object detection but fall short tasks. Enhancing the accuracy efficiency of detecting high-level cervical lesions invasive cancer poses a primary challenge model development.Between 2018 2022, we retrospectively studied total 777 patients, comprising 339 patients with 313 microinvasive or cancer. Overall, 1554 colposcopic images were put into DeepLabv3+ for learning. Accuracy, Precision, Specificity, mIoU employed to evaluate performance prediction cancer.Experiments showed that our had better diagnosis than experts other artificial intelligence models, reached Accuracy 93.29 %, Precision 87.2 Specificity 90.1 80.27 respectively.The good post-acetic-acid can assist colposcopists improving