Deep Transfer Learning for interpretable Chest X-Ray diagnosis C. Lago, * I. Lopez-Gazpio, and E. Onieva Faculty of Engineering, University of Deusto (UD) Av. Universidades 24, 48007 Bilbao, Spain carloslago@opendeusto.es, {inigo.lopezgazpio,enrique.onieva}@deusto.es Abstract. This work presents an application of different deep learning related paradigms to the diagnosis of multiple chest pathologies. Within the article, the application of a well-known deep Convolutional Neural Network (DenseNet ) is used and fine-tuned for different chest X-Ray medical diagnosis tasks. Different image augmentation methods are ap- plied over the training images to improve the performance of the resulting model as well as the incorporation of an explainability layer to highlight zones of the X-Ray picture supporting the diagnosis. The model is finally deployed in a web server, which can be used to upload X-Ray images and get a real-time analysis. The proposal demonstrates the possibilities of deep transfer learning and convolutional neural networks in the field of medicine, enabling fast and reliable diagnosis. The code is made publicly available 1 . Keywords: X-Ray diagnosis, Deep Learning, Convolutional Neural Net- works, Model interpretability, Transfer Learning, Image classification 1 Introduction Artificial Intelligence is poised to play an increasingly prominent role in medicine and healthcare due to advances in computing power, learning algorithms, and the availability of large datasets sourced from medical records and wearable health monitors [2]. In recent literature, deep learning shows promising results in medical specialities such as radiology [15], cancer detection [5], detection of referable diabetic retinopathy, age-related macular degeneration and glaucoma [19], and cardiology, in a wide range of problems involving cardiovascular dis- eases, performing diagnosis, predictions and helping in interventions [3]. In this article, we investigate the application of deep learning models for multiple chest pathology diagnoses with the objective of designing a fast and re- liable method for diagnosing various pathologies by analyzing X-Ray images. For the task, we employ, tune, and train a deep learning model using Tensorflow[1], as well as evaluate it on various medical state-of-the-art benchmarks through transfer learning. The used base model, DenseNet [11], is a state-of-the-art deep 1 https://github.com/carloslago/IntelligentXray - for the model training https://github.com/carloslago/IntelligentXray Server - for the server demo