基于深度学习的阴道镜及磁共振图像子宫识别研究

The Capability to Detect Uterine in Images of Colposcopy and MRI basedon Deep Learning System

目的:探讨深度学习方法在阴道镜及磁共振图像识别子宫中的应用,通过对阴道镜图像中子宫颈转化区的分类、磁共振图像中子宫位置的判识,判断该算法的准确性。方法:通过对8000幅阴道镜及200例磁共振图像进行裁剪、粗分割等预处理操作,最后引入深度学习模型进行分类,并将机器分类结果与专家标注结果进行比对。结果:本文提出方法的分类结果和专家进行标注的结果进行准确率验证,准确率达到84%,算法判识效果良好。结论:深度学习等算法的引用可以部分代替医生的工作量,节省不少人力,但是在准确度上可能需要进一步的提升。

Purpose： To explore the capability of deep learning system to detect uterine in images of colposcopy and MRI. Methods： Eight thousand images of colposcopy and 200 cases of MRI were selected. All of these images were pre-progressed by cutting, denoising, et al. ResNet50 deep learning system was introduced for image classification. The results of classification were compared with the labels noted by the specialists. Results： The accuracy of the deep learning system was 84%, which indicated some application value of this algorithm. Conclusions： The deep learning system could do some basic clinical work. But larger sample size study is needed to improve the accuracy,