A Deep Learning Interpretable Classifier for Diabetic Retinopathy Disease Grading

Deep neural network models have been proven to be very successful in image classification tasks, also for medical diagnosis, but their main concern is its lack of interpretability. They use to work as intuition machines with high statistical confidence but unable to give interpretable explanations about the reported results. The vast amount of parameters of these models make difficult to infer a rationale interpretation from them. In this paper we present a diabetic retinopathy interpretable classifier able to classify retine images into the different levels of disease severity and of explaining its results by assigning a score for every point in the hidden and input space, evaluating its contribution to the final classification in a linear way. The generated visual maps can be interpreted by an expert in order to compare its own knowledge with the interpretation given by the model. Keywords: deep learning, classification, explanations, diabetic retinopathy, model interpretation 2010 MSC: 68T10 1. Introduction Deep Learning methods have been used extensively in the last years for many automatic classification tasks. For the case of image analysis, the usual procedure consists on extracting the important features with a set of convolutional layers and, after that, make a final classification with these features using a set of fully connected layers. Finally, a soft-max output layer gives as a result the predicted output probabilities of the set of classes predefined in the model. Once the classifier has been trained (i.e. the parameters of the different layers of the model have been fixed), the quality of the classification outputs predicted is compared against the correct "true" values stored on a labeled dataset. This data is considered as the gold standard, ideally coming from the consensus of the knowledge of a human experts committee. This mapping allows the classification of multidimensional objects into a small number of categories. The model is composed by many neurons that are organized in layers and blocks of layers, piled together in a hierarchical way.