LKCA-Net: a segmentation model for automated diagnosis in cervical precancerous lesions with colposcopy

Résumé fourni par la source

Cervical cancer poses a serious threat to global women's health. Early detection of precancerous cervical lesions and timely treatment are crucial for reducing the incidence and mortality of cervical cancer. Clinically, colposcopy is an effective method for screening cervical lesions. However, the accuracy of manual observation heavily relies on clinicians' subjective experience, often leading to misdiagnoses and undiagnosed cases in underdeveloped regions. To address these issues, we propose a large kernel convolutional attention network (LKCA-Net), which improves the diagnostic accuracy and segmentation precision of lesions via attention-based enhancement and feature fusion processing. Our LKCA-Net exhibits an outstanding performance improvement over existing state-of-the-art methods, with the mIoU increased by 2.09%. For instance, LKCA-Net achieves the smallest IoU difference between HSIL and LSIL. Moreover, the model's generalization ability is validated on the ISIC2017 segmentation dataset, exhibiting superior performance on varying datasets. From a clinical standpoint, the results of the clinical comparison trial demonstrate the efficacy of the model in enhancing the detection efficiency of colposcopy in underdeveloped regions.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.