Deep Learning Based Image Classification for Remote Medical Diagnosis

In this paper, we describe research on Convolutional Neural Networks to advance access to medical diagnosis. Our research focuses on classifying skin cancer images as Benign or Malignant as a starting base to build on to develop an app that allows easily-accessible medical diagnosis in underdeveloped countries. The methodology of our research is based on Convolutional Neural Networks: AlexNet [3] and GoogleNet [4]. We hope to expand our work to be able to classify additional physically visible diseases in addition to Skin Cancer, since a camera only captures exterior physically visible diseases [5]. All our training and testing presented in this paper has been run on the NVIDIA Jetson TX2 GPU. Results are promising, showing accuracy rates up to 74 percent depending on how neural network parameters are changed. Down the road, we intend to incorporate this technology with a previously developed Vital Signs Multi-sensor kit [1]. The kit will be a compact and affordable device equipped with sensors that can be used to take a patients vital signs, such as blood pressure, heart rate, blood glucose concentration, and blood oxygen saturation. Combined, the tools will provide a complete system for remote medical diagnosis.