In Retina (Philadelphia, Pa.)
PURPOSE : To develop a deep learning (DL) model to detect morphologic patterns of diabetic macular edema (DME) based on optical coherence tomography (OCT) images.
METHODS : In the training set, 12,365 OCT images were extracted from a public dataset and an ophthalmic center. A total of 656 OCT images were extracted from another ophthalmic center for external validation. The presence or absence of three OCT patterns of DME, including diffused retinal thickening (DRT), cystoid macular edema (CME), and serous retinal detachment (SRD) were labeled with 1 or 0, respectively. A DL model was trained to detect three OCT patterns of DME. Occlusion test was applied for visualization of the DL model.
RESULTS : Applying five-fold cross-validation method in internal validation, the area under the receiver operating characteristic curve (AUC) for detection of three OCT patterns (i.e., DRT, CME, and SRD) were 0.971, 0.974, and 0.994, respectively, with accuracy of 93.0%, 95.1%, and 98.8%, respectively, sensitivity of 93.5%, 94.5%, and 96.7%, respectively, and specificity of 92.3%, 95.6%, and 99.3%, respectively. In external validation, the AUC were 0.970, 0.997, and 0.997, respectively, with accuracy of 90.2%, 95.4%, and 95.9%, respectively, sensitivity of 80.1%, 93.4%, and 94.9%, respectively, and specificity of 97.6%, 97.2%, and 96.5%, respectively. Occlusion test showed that the DL model could successfully identify the pathologic regions most critical for detection.
CONCLUSIONS : Our DL model demonstrated high accuracy and transparency in detection of OCT patterns of DME. These results emphasized the potential of artificial intelligence in assisting clinical decision-making processes in DME patients.
Wu Qiaowei, Zhang Bin, Hu Yijun, Liu Baoyi, Cao Dan, Yang Dawei, Peng Qingsheng, Zhong Pingting, Zeng Xiaomin, Xiao Y U, Li Cong, Fang Ying, Feng Songfu, Huang Manqing, Cai Hongmin, Yang Xiaohong, Yu Honghua