Original Article Application of deep learning image assessment software VeriSeeä for diabetic retinopathy screening Yi-Ting Hsieh a, *, Lee-Ming Chuang b,c , Yi-Der Jiang b , Tien-Jyun Chang b , Chung-May Yang a,c , Chang-Hao Yang a,c , Li-Wei Chan a,d , Tzu-Yun Kao a,e , Ta-Ching Chen a , Hsuan-Chieh Lin f , Chin-Han Tsai g , Mingke Chen g a Department of Ophthalmology, National Taiwan University Hospital, Taipei, Taiwan b Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan c College of Medicine, National Taiwan University, Taipei, Taiwan d Department of Ophthalmology, Taipei Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, New Taipei, Taiwan e Department of Ophthalmology, Shin Kong Wu Ho-Su Memorial Hospital, Taipei, Taiwan f Department of Ophthalmology, National Taiwan University Hospital, Hsinchu Branch, Hsinchu, Taiwan g Acer Inc., New Taipei, Taiwan Received 1 September 2019; received in revised form 9 December 2019; accepted 30 March 2020 KEYWORDS Artificial intelligence; Deep learning; Diabetic retinopathy; Convolutional neural network; Retinal fundus photography Purpose: To develop a deep learning image assessment software VeriSeeä and to validate its accuracy in grading the severity of diabetic retinopathy (DR). Methods: Diabetic patients who underwent single-field, nonmydriatic, 45-degree color retinal fundus photography at National Taiwan University Hospital between July 2007 and June 2017 were retrospectively recruited. A total of 7524 judgeable color fundus images were collected and were graded for the severity of DR by ophthalmologists. Among these pictures, 5649 along with another 31,612 color fundus images from the EyePACS dataset were used for model training of VeriSeeä. The other 1875 images were used for validation and were graded for the severity of DR by VeriSeeä, ophthalmologists, and internal physicians. Area under the receiver operating characteristic curve (AUC) for VeriSeeä, and the sensitivities and specific- ities for VeriSeeä, ophthalmologists, and internal physicians in diagnosing DR were calculated. Results: The AUCs for VeriSeeä in diagnosing any DR, referable DR and proliferative diabetic retinopathy (PDR) were 0.955, 0.955 and 0.984, respectively. VeriSeeä had better sensitivities * Corresponding author. Department of Ophthalmology, National Taiwan University Hospital, 7 Zhongshan S. Rd., Zhongzheng Dist., Taipei 10002, Taiwan. Fax: þ886 2 23934420. E-mail address: ythyth@gmail.com (Y.-T. Hsieh). + MODEL Please cite this article as: Hsieh Y-T et al., Application of deep learning image assessment software VeriSeeä for diabetic retinopathy screening, Journal of the Formosan Medical Association, https://doi.org/10.1016/j.jfma.2020.03.024 https://doi.org/10.1016/j.jfma.2020.03.024 0929-6646/Copyright ª 2020, Formosan Medical Association. Published by Elsevier Taiwan LLC. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Available online at www.sciencedirect.com ScienceDirect journal homepage: www.jfma-online.com Journal of the Formosan Medical Association xxx (xxxx) xxx