A Multi-dataset Approach for DME Risk Detection in Eye Fundus Images Catarina Carvalho 1(B ) , Jo˜ ao Pedrosa 1 , Carolina Maia 2 , Susana Penas 2,3 , ˆ Angela Carneiro 2,3 , Lu´ ıs Mendon¸ ca 4 , Ana Maria Mendon¸ ca 1,5 , and Aur´ elio Campilho 1,5 1 Institute for Systems and Computer Engineering, Technology and Science (INESC TEC), Porto, Portugal catarina.b.carvalho@inesctec.pt 2 Centro Hospitalar Universit´ario S˜ao Jo˜ao (CHUSJ), Porto, Portugal 3 Faculdade de Medicina da Universidade do Porto (FMUP), Porto, Portugal 4 Hospital de Braga, Braga, Portugal 5 Faculdade de Engenharia da Universidade do Porto (FEUP), Porto, Portugal Abstract. Diabetic macular edema is a leading cause of visual loss for patients with diabetes. While diagnosis can only be performed by opti- cal coherence tomography, diabetic macular edema risk assessment is often performed in eye fundus images in screening scenarios through the detection of hard exudates. Such screening scenarios are often associated with large amounts of data, high costs and high burden on specialists, motivating then the development of methodologies for automatic dia- betic macular edema risk prediction. Nevertheless, significant dataset domain bias, due to different acquisition equipment, protocols and/or different populations can have significantly detrimental impact on the performance of automatic methods when transitioning to a new dataset, center or scenario. As such, in this study, a method based on residual neu- ral networks is proposed for the classification of diabetic macular edema risk. This method is then validated across multiple public datasets, simu- lating the deployment in a multi-center setting and thereby studying the method’s generalization capability and existing dataset domain bias. Fur- thermore, the method is tested on a private dataset which more closely represents a realistic screening scenario. An average area under the curve across all public datasets of 0.891 ± 0.013 was obtained with a ResNet50 architecture trained on a limited amount of images from a single public dataset (IDRiD). It is also shown that screening scenarios are signifi- cantly more challenging and that training across multiple datasets leads to an improvement of performance (area under the curve of 0.911 ± 0.009). Keywords: Diabetic macular edema · Eye fundus · Screening · Classification C. Carvalho and J. Pedrosa—Equal contribution. c  Springer Nature Switzerland AG 2020 A. Campilho et al. (Eds.): ICIAR 2020, LNCS 12132, pp. 285–298, 2020. https://doi.org/10.1007/978-3-030-50516-5_25