Grading of Cartilage Damage in Degenerative Knee Osteoarthritis Based on Quantitative Parameters of the Infrapatellar Fat Pad: A Cross-Sectional Study.

Abstract

Purpose: The aim of this research was to investigate the relationship between quantitative texture parameters and degenerative cartilage damage in knee osteoarthritis (KOA) by conducting a full-volume texture analysis of infrapatellar fat pad (IFP). In addition, this study also explored if the quantitative texture parameter models outperform semi-quantitative model in cartilage damage classification tasks.

Materials and methods: This retrospective study involved 202 patients who were diagnosed with KOA using imaging and clinical examinations. Texture parameters of the IFP were extracted from sagittal FSE PDWI fat-suppressed sequence images, and least absolute shrinkage and selection operator regression was used for feature selection. Spearman correlation analysis was conducted to assess the relationship between semi-quantitative parameter (Hoffa-synovitis score), quantitative parameters, and cartilage damage. Five multi-classification logistic regression models were developed to predict cartilage damage grade by using Hoffa-synovitis score, texture parameters, and clinical characteristics as independent variables. Subsequently, the performance of these models was compared.

Results: Eight texture features were screened out in this study. Correlation analysis showed that Hoffa synovitis score, texture parameters, and cartilage damage grade were significantly correlated (all P < 0.05). The strongest correlation was found between Hoffa-synovitis score and cartilage damage, demonstrating a moderate positive relationship (r = 0.62). In terms of texture features, the Correlation parameter exhibited a moderate positive correlation with cartilage damage (r = 0.49), while other texture parameters had a slight positive correlation degree of positive or negative correlation. In the task of classifying cartilage damage, the model's macro-average area under the curve (AUC) only using the Hoffa-synovitis score was 0.73 (95% confidence interval (CI): 0.64, 0.83), while the model using only selected texture parameters achieved a macro-average AUC of 0.84 (95% CI: 0.68, 0.94). Furthermore, the model that combined texture parameters and clinical features also achieved a macro-average AUC of 0.84 (95% CI: 0.72, 0.94). By integrating the Hoffa-synovitis score, texture parameters, and clinical features, the model's macro-average AUC experienced a slight improvement to 0.85 (95% CI: 0.74, 0.93). Notably, the model combining only Hoffa-synovitis score and texture parameters had the best classification performance, with a macro-average AUC of 0.88 (95% CI: 77, 0.97). The performance of the 4 models incorporating texture parameters outperformed that of the Hoffa-synovitis score alone (all P < 0.05), however with no significant statistical difference observed among the 4 models (all P > 0.05).

Conclusions: There existed a correlation between the texture parameters of the infrapatellar fat IFP and cartilage damage in KOA. The models using texture parameters demonstrated better performance in classifying cartilage damage compared to the models using only semi-quantitative parameter. Thus, we can infer that texture parameters had the potential to be valuable imaging biomarkers for evaluating cartilage damage.