Automated morphometric segmentation analysis of hand X-ray image using deep learning network
- 1Department of Thoracic and Cardiovascular Surgery, The Affiliated Drum Tower Hospital of Nanjing University Medical School, Kuang Yaming Honors School, Nanjing University
- 2National Laboratory of Solid State Microstructure, Department of Physics, Nanjing University
Abstract
Background The rapid development of deep learning in computer vision has led to increasing interest in its applications to medical image analysis. While many studies have focused on morphometric measurements from MR and CT images, comparatively few works have applied deep learning to extract morphometric features from hand X-ray images. Objectives This research aims to establish a scheme for deep learning based automatic segmentation and morphology feature extraction of hand X-ray images with accuracy and efficiency. Methods A total of 668 hand X-ray images from patients were obtained and were randomly classified as 498 training images(74.6%), 102 validation images(15.2%) and 68 testing images(10.2%). A combination of nnU-net and YOLO deep learning model is applied to perform the automatic segmentation. The outcomes were benchmarked against manual segmentations performed by experienced clinicians, enabling an assessment of the model’s accuracy and overall performance. An in-house program was applied to the segmented part to extract the morphology features, such as Finger Bone Length(FBL) and Finger Bone Width(FBW). Results For nnU-net training, the IoU score on the training dataset is 0.9596 and the IoU score on the validation dataset is 0.9584. For YOLOv8 training, we got an accuracy of 0.9816 and a loss of 0.0627. For feature extraction, the Pearson and Spearmann coefficient between automatic and manual results along with ICC shows the consistency of our automatic feature extraction method and human extracted method at a 95% confidence level, and the mean difference is close to zero. As a result, our scheme proved both accurate and precise. Conclusions We present a two-stage, clinically interpretable workflow for hand X-ray analysis that combines nnU-Net-based hand ROI segmentation with YOLO bone-wise instance segmentation, followed by automated extraction of finger-bone morphometric indicators (FBL/FBW). Beyond reporting segmentation/detection metrics, we quantitatively validate the extracted morphometrics against expert measurements using correlation, ICC, and Bland-Altman analysis, demonstrating consistent bone-wise agreement and supporting the potential utility of the pipeline for standardized clinical measurements.
Related
- Automated correction of training sets for biological image segmentation by convolutional networks
- Contrastive adversarial mix-up learning for unsupervised cross-modality image segmentation
- Enhancing medical image segmentation with HCE-Net: A heterogeneous feature collaborative extraction network
- An attention-guided dense convolutional network for automated diabetic foot ulcer segmentation
- Spatially-adaptive dynamic dictionary learning for fine-grained remote sensing image segmentation
BibTeX
@article{zhang2026automated,
title = {{Automated morphometric segmentation analysis of hand X-ray image using deep learning network}},
author = {Yiyang Zhang and Zhukai Zhuang and Pengli Yu and Yachong Guo and Xiaocheng Shi and Wei Wang},
journal = {BMC Medical Imaging},
year = {2026},
month = jul,
doi = {10.1186/s12880-026-02511-z},
url = {https://doi.org/10.1186/s12880-026-02511-z},
note = {CC BY},
}
RIS
TY - JOUR
TI - Automated morphometric segmentation analysis of hand X-ray image using deep learning network
AU - Yiyang Zhang
AU - Zhukai Zhuang
AU - Pengli Yu
AU - Yachong Guo
AU - Xiaocheng Shi
AU - Wei Wang
PY - 2026
DA - 2026/07/01
JO - BMC Medical Imaging
DO - 10.1186/s12880-026-02511-z
UR - https://doi.org/10.1186/s12880-026-02511-z
AB - Background The rapid development of deep learning in computer vision has led to increasing interest in its applications to medical image analysis. While many studies have focused on morphometric measurements from MR and CT images, comparatively few works have applied deep learning to extract morphometric features from hand X-ray images. Objectives This research aims to establish a scheme for deep learning based automatic segmentation and morphology feature extraction of hand X-ray images with accuracy and efficiency. Methods A total of 668 hand X-ray images from patients were obtained and were randomly classified as 498 training images(74.6%), 102 validation images(15.2%) and 68 testing images(10.2%). A combination of nnU-net and YOLO deep learning model is applied to perform the automatic segmentation. The outcomes were benchmarked against manual segmentations performed by experienced clinicians, enabling an assessment of the model’s accuracy and overall performance. An in-house program was applied to the segmented part to extract the morphology features, such as Finger Bone Length(FBL) and Finger Bone Width(FBW). Results For nnU-net training, the IoU score on the training dataset is 0.9596 and the IoU score on the validation dataset is 0.9584. For YOLOv8 training, we got an accuracy of 0.9816 and a loss of 0.0627. For feature extraction, the Pearson and Spearmann coefficient between automatic and manual results along with ICC shows the consistency of our automatic feature extraction method and human extracted method at a 95% confidence level, and the mean difference is close to zero. As a result, our scheme proved both accurate and precise. Conclusions We present a two-stage, clinically interpretable workflow for hand X-ray analysis that combines nnU-Net-based hand ROI segmentation with YOLO bone-wise instance segmentation, followed by automated extraction of finger-bone morphometric indicators (FBL/FBW). Beyond reporting segmentation/detection metrics, we quantitatively validate the extracted morphometrics against expert measurements using correlation, ICC, and Bland-Altman analysis, demonstrating consistent bone-wise agreement and supporting the potential utility of the pipeline for standardized clinical measurements.
SN - 1471-2342
ER -