Transfer Learning for Hand Arthritis Prediction from X-Ray Images

Abstract

Arthritis is a bone disorder that includes swelling and pain in one or more joints. Everyone can develop osteoarthritis, but it grows more common as individuals get older. When arthritis deteriorates over time, it can lead to persistent pain, making it challenging to do daily tasks, and making activities like walking and climbing stairs painful and difficult. If arthritis is correctly identified and treated in its early stages, these consequences can be avoided. The goal of this project is to create two transfer learning models that, by spotting arthritis in its earliest stages, can lower the likelihood of acquiring chronic arthritis. For this purpose, Google served as the source of the images used in this study. After being purchased from Google, the data collection is preprocessed using three different methods. Image scaling, noise reduction, and image enhancement are a few of the pre-processing approaches. The transfer learning models are trained and assessed using this preprocessed dataset. In this work, two distinct transfer learning models are established. The models include SegNet and ENet. On a graph, the outcomes for the performances of both models are displayed. The training data from the first few epochs of the ENet model and SegNet model are also used in the analysis. The models' final accuracy and loss values are then assessed. In the end, it was discovered that the SegNet model had a lower loss value and more accuracy than the other. The model created in this study can be utilised as a preliminary test for arthritis when a person exhibits moderate arthritis symptoms because the final accuracy of the model is higher than or equal to 95%.