X-Rays Based Knee Osteoarthritis Diagnosis and Classification

Abstract

Knee osteoarthritis (OA) is a condition that affects the knee joints and is categorized by radiographic changes. Accurate and early diagnosis of through X-rays is crucial for initiating appropriate treatment and slowing the progression of the disease. This project work implements a deep learning based ordinal classification system to uniquely identify and categorize knee osteoarthritis (OA) using single knee X-ray images. The proposed approach leverages the Osteoarthritis Initiative (OAI) dataset and join in several models, including DenseNet161, Xception, VGG19, and ResNet34. Additionally, for detection purposes, we employ advanced YOLO models such as YOLOv6, YOLOv7, YOLOv5x6, and YOLOv5 GhostNet. While initial work involved using DenseNet169 with fine-tuning for classification and comparing Transfer Learning (TL) techniques, our project expands upon this by exploring additional methods to enhance classification performance and by utilizing a broader range of YOLO models for improved detection.