The value of deep learning-based X-ray techniques in detecting and classifying K-L grades of knee osteoarthritis: a systematic review and meta-analysis.

Abstract

Objectives: Knee osteoarthritis (KOA), a prevalent degenerative joint disease, is primarily diagnosed through X-ray imaging. The Kellgren-Lawrence grading system (K-L) is the gold standard for evaluating KOA severity through X-ray analysis. However, this method is highly subjective and non-quantifiable, limiting its effectiveness in detecting subtle joint changes on X-rays. Recent researchers have been directed towards developing deep-learning (DL) techniques for a more accurate diagnosis of KOA using X-ray images. Despite advancements in these intelligent methods, the debate over their diagnostic sensitivity continues. Hence, we conducted the current meta-analysis.

Methods: A comprehensive search was conducted in PubMed, Cochrane, Embase, Web of Science, and IEEE up to July 11, 2023. The QUADAS-2 tool was employed to assess the risk of bias in the included studies. Given the multi-classification nature of DL tasks, the sensitivity of DL across different K-L grades was meta-analyzed.

Results: A total of 19 studies were included, encompassing 62,158 images. These images consisted of 22,388 for K-L₀, 13,415 for K-L₁, 15,597 for K-L₂, 7768 for K-L₃, and 2990 for K-L₄. The meta-analysis demonstrated that the sensitivity of DL was 86.74% for K-L₀ (95% CI: 80.01%-92.28%), 64.00% for K-L₁ (95% CI: 51.81%-75.35%), 75.03% for K-L₂ (95% CI: 66.00%-83.09%), 84.76% for K-L₃ (95% CI: 78.34%-90.25%), and 90.32% for K-L₄ (95% CI: 85.39%-94.40%).

Conclusions: The DL multi-classification methods based on X-ray imaging generally demonstrate a favorable sensitivity rate (over 50%) in distinguishing between K-L₀-K-L₄. Specifically, for K-L₄, the sensitivity is highly satisfactory at 90.32%. In contrast, the sensitivity rates for K-L_1-2 still need improvement.

Clinical relevance statement: Deep-learning methods have been useful to some extent in assessing the effectiveness of X-rays for osteoarthritis of the knee. However, this requires further research and reliable data to provide specific recommendations for clinical practice.

Key points: X-ray deep-learning (DL) methods are debatable for evaluating knee osteoarthritis (KOA) under The Kellgren-Lawrence system (K-L). Multi-classification deep-learning methods are more clinically relevant for assessing K-L grading than dichotomous results. For K-L3 and K-L4, X-ray-based DL has high diagnostic performance; early KOA needs to be further improved.