A new review of 32 studies suggests that deep-learning algorithms, a form of artificial intelligence, can help grade knee osteoarthritis from X-rays. The analysis found that these AI models performed well at identifying moderate to severe cases, but they were less reliable at catching the earliest signs of the disease.
Researchers looked at studies that used deep learning to grade knee osteoarthritis according to the Kellgren-Lawrence scale, which ranges from grade 0 (no arthritis) to grade 4 (severe). Across the studies, the AI models correctly identified about 90% of cases with no arthritis (grade 0) and 88% of cases with severe arthritis (grade 4). However, for grade 1, which indicates doubtful or minimal arthritis, the models only caught about 66% of cases.
The review also found that the AI models were better at correctly identifying which knees had a particular grade when they were moderate to severe. For example, precision was 91% for grade 4 but only 71% for grade 1. The authors noted high variability among the studies, which makes the results less certain.
While these findings are encouraging, the technology is not yet ready for everyday clinical use, especially for detecting early-stage knee osteoarthritis. The studies varied widely in their methods, and many lacked external validation, meaning the models were not tested on diverse patient groups. More research is needed to improve accuracy for early stages and to ensure the models work reliably across different populations.
For now, patients and doctors should view AI as a potential future tool, not a replacement for a doctor's judgment. If you have concerns about knee pain or osteoarthritis, talk to your healthcare provider about the best ways to diagnose and manage your condition.
