Construction and Clinical Validation of a Deep Learning-Based Automatic Measurement Model for Palmar Tilt and Radial Inclination in Distal Radius Fractures
10.13288/j.11-2166/r.2026.10.011
- VernacularTitle:基于深度学习的桡骨远端骨折掌倾角与尺偏角自动测量模型构建及临床验证
- Author:
Guoda DAI
1
;
Jianwei WANG
1
;
Mao WU
1
;
Bin KANG
2
;
Yang SHAO
1
;
Hengyan CUI
1
;
Shaoshuo LI
1
;
Tingchen ZHU
1
;
Zhen HUA
1
;
Zhongming SHEN
1
;
Jintao LIU
3
;
Ming ZHOU
4
Author Information
1. Wuxi Hospital Affiliated to Nanjing University of Chinese Medicine，Wuxi，214071
2. Nanjing University of Posts and Telecommunications
3. Suzhou Hospital of Traditional Chinese Medicine
4. Changzhou Hospital of Traditional Chinese Medicine
- Publication Type:Journal Article
- Keywords:
distal radius fracture;
manipulative reduction;
palmar tilt;
radial inclination;
deep learning;
artificial intelligence
- From:
Journal of Traditional Chinese Medicine
2026;67(10):1093-1100
- CountryChina
- Language:Chinese
-
Abstract:
ObjectiveTo construct an automatic measurement model for palmar tilt and radial inclination suitable for traditional Chinese medicine （TCM） clinical scenarios， and to validate its accuracy and efficiency in TCM manipulative reduction settings. MethodsData on anteroposterior （AP） and lateral X-rays of distal radius fractures were collected from patients admitted to 18 TCM/ integrated TCM and western medicine hospitals in Jiangsu province between September 1st， 2023， and September 1st， 2024， via the Jiangsu Diagnosis and Treatment Big Data Platform for TCM Dominant Diseases. A medical image segmentation framework based on multi-scale feature fusion and edge-awareness was employed， combined with anatomical knowledge specific to TCM orthopedics， to optimize the feature extraction strategy of an artificial intelligence （AI） model. This framework enabled automatic segmentation of fracture regions and measurement of distal radius palmar tilt and radial inclination. The accuracy of the AI model in measuring radial inclination and volar tilt was validated， and the measurement time and average time gain rate of the AI model were compared to those of manual measurement. ResultsA total of 15,444 AP and lateral X-ray images of distal radius fractures were collected， and were divided into a training set （11,144 images， 5066 AP and 6078 lateral）， a validation set （3700 images， 1840 AP and 1860 lateral）， and an independent test set （600 images， 300 AP and 300 lateral） after preprocessing. In the measurement of 300 AP X-rays in the independent test set for radial inclination， when the degree error between AI measurement and manual measurement was ＜3° and ＜5°， AI measurement accuracy was 83% and 93%， respectively. In 300 lateral X-rays in the test set for palmar tilt， when AI measurements had an error of ＜3° and ＜5° compared to manual measurements， corresponding accuracy rate was 78% and 90%， respectively. For 50 X-ray images， AI measurement time was （1.37±0.05） min for radial inclination while manual measurement time was （22.57±2.52） min （P＜0.001）； in terms of palmar tilt， the AI measurement time was （1.33±0.14） min， shorter than （23.70±2.80） min for manual measurement time （P＜0.001）. Average time gain rates for manual and AI measurements were 93.93% and 94.39% respectively. ConclusionAn automatic measurement model for palmar tilt and radial inclination in distal radius fractures has been established， enabling more accurate and efficient assessment as well as providing a tool to support the quantitative evaluation of the efficacy of TCM manipulative reduction and large-sample clinical research.
