OSTEOARTHRITIS GRADING: A SYNTHESIZED MAGNETIC RESONANCE IMAGES TECHNIQUE

Abstract

Osteoarthritis (OA) in the knee is a major cause of decreased activity and physical limitations among older people. Identifying and treating knee osteoarthritis in its early stages can help patients delay the progression of the condition. Currently, early detection of knee osteoarthritis involves the use of X-ray images and assessment using the Kellgren-Lawrence (KL) grading system. Doctors' evaluations can be subjective and may differ among different doctors. Similar to a computer systems analyst, the automatic knee OA grading and diagnosis can be a valuable tool for doctors, enabling them to streamline their workload and provide more efficient care. An innovative network named OA_GAN_ViT has been developed to autonomously detect knee OA. The network is a ViT architecture consisting of two branches: one branch utilizes the synthesized MR image derived from X-ray images for data processing before classification operations via the GAN network, while the other branch employs a histogram-equalized X-ray image. The OA_GAN_ViT network demonstrated superior performance in terms of accuracy and MAE compared to well-known neural networks such as ResNet, DenseNet, VGG, Inception, and ViT. It achieved an impressive accuracy of 79.2 and an MAE of 0.492, highlighting its effectiveness.

Keywords

• deep learning
• multimodal synthesis
• oa grading
• pre-process