Early Detection of Rheumatoid Arthritis Using Image Classification

Abstract

Rheumatoid arthritis (RA) is a form of chronic inflammatory disorder that affects areas such as the hands and feet. According to MedicalNewsToday, roughly 1.3 million people in the US have RA, representing 0.6 to 1% of the population. Artificial intelligence (AI) is the ability of machines to perform tasks that typically require human intelligence and is becoming more widespread in areas such as healthcare. The detection of the more common osteoarthritis has been performed using AI before, and RA detections are starting to emerge too. However, these detection methods use X-rays and protein scans, which take time and money. Since arthritis is a disorder that happens in the joints, automating its detection using images could be done in a new, revolutionary way. To get this, two image datasets were used, the first being healthy hands with no arthritis symptoms. The second data set would contain images of nodules which are bumps on the hand for RA symptoms. The model would be created using Jupyter Notebook, TensorFlow, and Keras, as well as Python 3.9, where the data would then go through preprocessing, scaling, and splitting for faster training. The deep learning model known as a convolutional neural network is used along with model.fit for training. The accuracy yielded 99.48%, and overall, it was able to classify between the two data sets. The conclusion is that classifying RA from just a scan of someone's hand could, in the future, allow for a faster diagnosis of any arthritis when it is perfected.