Automatic Grading of Knee Osteoarthritis on the Kellgren-Lawrence Scale from Radiographs Using Convolutional Neural Networks

Kondal, Sudeep Kulkarni, Viraj Gaikwad, Ashrika Kharat, Amit Pant, AniruddhaThe severity of knee osteoarthritis is graded using the 5-point Kellgren-Lawrence scale where healthy knees are assigned grade 0, and subsequent grades 1–4 represent increasing affliction. Although several methods have been proposed in recent years to develop models that can automatically predict from a given radiograph, most developed evaluated on datasets not sourced India. These fail perform well radiographs Indian patients. In this paper, we propose novel method convolutional neural networks scale. Our works two connected stages: first stage, an object detection model segments individual rest image; second regression each separately We train our publicly available Osteoarthritis Initiative dataset demonstrate fine-tuning before evaluating it private hospital significantly lowers corresponding mean absolute error. Additionally, compare classification built for same task outperforms classification.