Evolution Rules of Hip Magnetic Resonance Imaging Features in Non-Traumatic Osteonecrosis of the Femoral Head with Blood Stasis Syndrome：Based on Machine Learning Analysis
10.13288/j.11-2166/r.2026.22.012
- VernacularTitle:基于机器学习方法解析非创伤性股骨头坏死患者血瘀证髋关节磁共振成像影像特征的演变规律
- Author:
Jiaming LIN
1
;
Zhi LIANG
1
;
Yawei DONG
1
;
Jiachun LIU
1
;
Yanguang SHEN
1
;
Xin DENG
2
;
Lai JIANG
2
;
Yanqiong ZHANG
3
;
Na LIN
3
;
Baohong MI
1
;
Weiheng CHEN
1
Author Information
1. The Third Affiliated Hospital of Beijing University of Chinese Medicine，Beijing，100029
2. Beihang University
3. Institute of Chinese Materia Medica，China Academy of Chinese Medical Sciences
- Publication Type:Journal Article
- Keywords:
non-traumatic osteonecrosis of the femoral head;
magnetic resonance imaging;
blood stasis syndrome;
radiomics;
machine learning;
prediction model
- From:
Journal of Traditional Chinese Medicine
2026;67(22):2434-2442
- CountryChina
- Language:Chinese
-
Abstract:
ObjectiveTo construct a machine learning prediction model for grading blood stasis syndrome （BSS） in non-traumatic osteonecrosis of the femoral head （NONFH） based on magnetic resonance imaging （MRI） features， and to elucidate the dynamic evolution of blood stasis-related imaging features across the phlegm-stasis obstructing the collaterals syndrome， meridian obstruction syndrome， and liver-kidney depletion syndrome of NONFH. MethodsA total of 100 patients with NONFH were retrospectively enrolled and randomly divided into a training set （70 cases） and an internal validation set （30 cases） at a ratio of 7∶3. Patients were stratified into mild， moderate， and severe BSS groups. Macroscopic MRI features including degree of bone marrow edema， necrosis location， and necrosis area， and radiomic features including first-order， texture， and wavelet transform features were extracted. Optimal features were selected through correlation analysis and least absolute shrinkage and selection operator （LASSO） regression. Random forest algorithms were then used to construct a macroscopic imaging model， a radiomics model， and a combined model. An additional 30 NONFH patients enrolled prospectively served as an external validation set. Model performance was evaluated by the area under the receiver operating characteristic curve （AUC）， and the dynamic evolution of BSS was revealed by integrating model predictions with changes in imaging features. ResultsAmong the 100 patients， 10 had mild， 41 moderate， and 49 severe BSS. Among macroscopic features， the degree of bone marrow edema was significantly positively correlated with BSS grade （r=0.71， P＜0.05）， whereas necrosis location and area showed no correlation （P＞0.05）. Among radiomic features， wavelet-LHL gray-level co-occurrence matrix correlation， wavelet-HHL first-order mean， and wavelet-LLH gray-level run-length matrix gray-level non-uniformity were most strongly associated with BSS grade. Internal validation showed that the accuracy of the combined model （0.75） was superior to that of the macroscopic model （0.73） and the radiomics model （0.62）. External validation demonstrated that the combined model performed best， achieving an AUC of 0.884 （95%CI： 0.741-1.000）， with prediction accuracies of 0.86 and 0.90 for moderate and severe BSS， respectively， significantly outperforming the macroscopic model （AUC=0.737， 95%CI： 0.526-0.949）. Applying this model to the external validation set revealed that patients with meridian obstruction syndrome had the highest proportion of severe BSS （80.0%）， significantly higher than those with phlegm-stasis obstructing the collaterals syndrome （30.0%， 3/10） and liver-kidney depletion syndrome （10.0%， 1/10， P＜0.05）. The wavelet-LLH gray-level run-length matrix gray-level non-uniformity mean exhibited a "low-high-low" distribution （phlegm-stasis obstructing the collaterals syndrome， 1121.41±500.34； meridian obstruction syndrome， 1163.46±418.80； liver-kidney depletion syndrome， 1040.14±272.94； P＜0.05）. ConclusionThe combined macroscopic imaging-radiomics model for BSS in NONFH revealed a "mild-severe-mild" dynamic evolution across the phlegm-stasis obstructing the collaterals， meridian obstruction， and liver-kidney depletion syndromes from an imaging perspective.
