Osteoarthritis Classification using Convolutional Neural Networks to Mitigate the Digital Divide

Abstract

Osteoarthritis, the erosion of cartilage in the knee region, affects millions of individuals around the world, and, if left untreated, can severely hinder quality of life. To support diagnoses of osteoarthritis, convolutional neural networks (CNNs) can supplant doctors’ decisions when identifying the severity of osteoarthritis. However, since many CNNs require substantial computing power, this limits their utility to doctors in regions with limited infrastructure, thereby exacerbating the digital divide. In this paper, we implement a custom neural network that uses X-ray images to classify the severity of osteoarthritis that only requires minimal computational resources. We show that through careful hyperparameter tuning, our model can achieve high levels of accuracy even though it has far fewer parameters than many CNNs. Therefore, we anticipate that medical professionals could use our work in resource-limited areas as an aid to diagnosis.