Non-Invasive Multi-Disease Classification via Facial Image Analysis Using a Convolutional Neural Network

Abstract

Diabetes and lung disease are some of the most common medical conditions in the world. The economic costs and social burdens brought by these two diseases are considerable. Even though there are proven methodologies for diagnosing each disease individually in practice, there does not exist a single non-invasive methodology/procedure that can detect both diseases. With recent advancements made in machine learning and pattern recognition, the Convolutional Neural Network (CNN) has been widely used in many recognition applications due to its high efficiency and performance. Therefore, in this paper we propose an approach using CNN for non-invasive multi-disease classification called Multi-Disease CNN (MD-CNN). Facial images are first captured using our specially designed device. Next, four facial blocks are extracted located at specific regions on the face. Finally, the facial blocks are concatenated and used as input for our MD-CNN. Based on three datasets consisting of healthy control, diabetes and lung disease, the proposed method achieved an average accuracy of 73%. When compared to other classifiers not employing a deep learning architecture, MD-CNN produced the highest result. This show a potentially new way to perform multi-disease classification.