目的:以临床前期糖尿病视网膜病变为例分析医学图像人工智能系统对于自发荧光图像识别的能力,为早期诊断治疗提供技术支持.方法:连续收集2017年8月至2018年5月在温州医科大学附属第三医院眼科门诊就诊患者的眼底自发荧光图像,按是否患有糖尿病标准,纳入正常组102例(200眼),糖尿病组105例(200眼).受检者均行裂隙灯显微镜、前置镜、裸眼视力或矫正视力、眼底自发荧光影像等检查.采用基于二维格子复杂性度量的医学图像特征提取和识别系统对糖尿病组及正常组图像进行分析.结果:该系统分析临床前期糖尿病视网膜病变眼底自发荧光图像与正常视网膜自发荧光图像具有可识别差异,提取出具有比较意义的25个特征.针对25个特征进行单个特征及多个特征的10折交叉检验以及5折交叉检验,准确率为82.47％.结论:复杂性分析医学图像人工智能系统可用于识别临床前期糖尿病视网膜病变的眼底自发荧光改变,准确率高.%Objective: To analyze the application of a medical imaging artificial intelligence system for spontaneous fluorescence imaging recognition using pre-clinical diabetic retinopathy as an example so as to provide a technical exploration for early diagnosis and treatment. Methods: The fundus autofluorescence images of 102 patients (200 eyes) in a control group and 105 patients (200 eyes) in a study group were collected from August 2017 to May 2018. All patients were examined by a slit lamp microscope, preview lens, naked eye or corrected visual acuity and fundus autofluorescence images. The images from the control and study groups were used for analysis. The medical image extraction and recognition system is based on a two-dimensional lattice complexity measurement and was used to analyze the discernible differencesbetween the fundus autofluorescence image of pre-clinical diabetic retinopathy and the normal retinal autofluorescence image. Results: Twenty-five features with comparative significance were extracted. The single and multiple features were tested by 10-fold and 5-fold cross tests for 25 features, and the accuracy rate was 82.47％. Conclusions: Complex analysis of a medical imaging artificial intelligence system can be used to identify the spontaneous fluorescence changes on the fundus of pre-clinical diabetic retinopathy with high accuracy.
展开▼
