AUTOMATED KNEE OSTEOARTHRITIS SEVERITY DETECTION FROM X-RAY IMAGES: A DATA-CENTRIC AI APPROACH

Abstract

Knee osteoarthritis remains one of the main sources of pain and disabling conditions that affect the knees of millions of people. Consequently, the early detection of the disease can greatly impact patient care. Nevertheless, the diagnosis of KOA through X-ray is not always a simple solution. These vast datasets of medical images are typically quite inconsistent, noisy, and a variety of image qualities. We aimed to enable AI to detect KOA more accurately by connecting data preparation as the most crucial step of the process in our study. To render the images more AI-friendly, we performed a thorough cleaning of the X-ray images to remove noise and ensure consistent quality. We separated the images into different levels of severity of the disease after cleaning, which allowed our algorithm to recognize the smallest changes as the disease progressed. This well-prepared dataset allowed the AI model to obtain a high level of accuracy in detecting KOA, even being able to match the performance of human experts. We designed an AI tool that can offer stable and reliable assistance to doctors, thus, the diagnostic procedure can be potentially expedited and patient outcomes improved without the need for data of inferior quality from the very beginning. We plan to perfect these methods and also study the possibility of using them for diagnosing other joint disorders.