A data-driven approach to referable diabetic retinopathy detection

Agradecimentos: We thank the medical team from the Department of Ophthalmology, Federal University of São Paulo, for the data collection and annotation and Dr. Alexandre Ferreira for suggestions on earlier drafts of this work. We thank José A. Stuchi, Flavio P. Vieira and Diego Lencione, co-founders of Phelcom Technologies for encouraging us in investigation of efficient solutions. We thank Kaggle, the California Healthcare Foundation, and EyePACs for sponsoring a diabetic retinopathy competition and for providing the community with a useful and wide dataset. The Messidor-2 dataset was kindly provided by the LaTIM laboratory (see http://latim.univ-brest.fr/) and the Messidor program partners (see http://messidor.crihan.fr/). We also thank the University of Iowa for providing the reference standard of Messidor-2 dataset. We gratefully acknowledge the support of NVIDIA Corporation with the donation of the GPUs used for this research. Finally, this work was supported in part by Microsoft Research, São Paulo Research Foundation (Fapesp) under the grant #2017/12646-3, the National Council for Scientific Research (CNPq) under grants #311486/2014-2, and 304472/2015-8, Amazon Web Services, CAPES DeepEyes project, CAPES/PNPD, and Google Research. MDA is Director and shareholder of IDx, LLC, Iowa City, and has patents and patent applications, assigned to the University of Iowa, that may compete with the technology that is the subject of this study. IDx, LLC is not associated with the present study and has no interest in the presented methods Abstract: Prior art on automated screening of diabetic retinopathy and direct referral decision shows promising performance; yet most methods build upon complex hand-crafted features whose performance often fails to generalize. We investigate data-driven approaches that extract powerful abstract representations directly from retinal images to provide a reliable referable diabetic retinopathy detector.We gradually build the solution based on convolutional neural networks, adding data augmentation, multi-resolution training, robust feature-extraction augmentation, and a patient-basis analysis, testing the effectiveness of each improvement. The proposed method achieved an area under the ROC curve of 98.2% (95% CI: 97.4-98.9%) under a strict cross-dataset protocol designed to test the ability to generalize - training on the Kaggle competition dataset and testing using the Messidor-2 dataset. With a 5 × 2-fold cross-validation protocol, similar results are achieved for Messidor-2 and DR2 datasets, reducing the classification error by over 44% when compared to most published studies in existing literature. Additional boost strategies can improve performance substantially, but it is important to evaluate whether the additional (computation- and implementation-) complexity of each improvement is worth its benefits. We also corroborate that novel families of data-driven methods are the state of the art for diabetic retinopathy screening. Significance: By learning powerful discriminative patterns directly from available training retinal images, it is possible to perform referral diagnostics without detecting individual lesions CONSELHO NACIONAL DE DESENVOLVIMENTO CIENTÍFICO E TECNOLÓGICO - CNPQ FUNDAÇÃO DE AMPARO À PESQUISA DO ESTADO DE SÃO PAULO - FAPESP COORDENAÇÃO DE APERFEIÇOAMENTO DE PESSOAL DE NÍVEL SUPERIOR - CAPES Fechado