COMPUTER VSISION BASED DERMA MONITORING APPLICATION FOR PATIENT CARE AND TREATMENT

Abstract

Skin diseases are a very common occurrence in a person's life. Millions of people around the world are affected by various kinds ofskin disorders. These diseases require a lot of experience to identify with a naked eye, human judgment is often considered unreliable due to the similarity ofskin diseases. A slight judgment in error can result in a life-threatening situation. For a skin cancer the process beginning with a visual screening followed by dermoscopic analysis, biopsy and histopathological examination is a very time consuming and costly task for a patient in the later stages of a skin cancer. Here we need such a system which can assist in identifying these skin diseases and serve as a reliable computer aided diagnostics system. To perform such tasks many experiments have been conducted worldwide in previous years in Artificial Intelligence; specifically, deep learning. The field of machine learning has evolved rapidly from discriminating cats and dogs to predicting the chemical reactions in organic chemistry. Deep learning has been applied to fields including computer vision, speech recognition, natural language processing, audio recognition, social network filtering, machine translation etc. and have yielded near accurate results. One of the most prominent machine learning architecture when it comes to tasks belonging to computer vision is CNN (Convolutional Neural Network). CNN’s have been very successful due to their ability in extracting features of images. CNN’s are similar to ANN’s but are more efficient and light which makes them easy to implement in a mobile device. When matched with the power of tensor flow and android these neural networks can potentially enhance a dermatologist's task by providing a near accurate and low-cost universal access device which can be used in diagnostic care.We took the dataset from dermnet and ISIC and as well as from PNS SHIFA HOSPITAL web-sites for our project. The overall accuracy of our skin lesion application is approximately 68%.