Automatic Measuring of Finger Joint Space Width on Hand Radiograph using Deep Learning and Conventional Computer...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Hand osteoarthritis (OA) severity assessment via radiographs is often subjective.
• Current semi-quantitative grading systems lack precision for minor differences.
• Joint space width (JSW) measurement quantifies OA severity but manual methods are time-consuming.

Purpose of the Study:

• To develop and evaluate automated methods for measuring JSW in hand radiographs.
• To compare a deep learning approach against traditional computer vision techniques for JSW quantification.
• To improve the efficiency and robustness of hand OA severity assessment.

Main Methods:

• Two novel methods were proposed: a segmentation-based (SEG) method using computer vision and a regression-based (REG) deep learning method (modified VGG-19).
• A dataset of 3,591 hand radiographs with 10,845 DIP joints was used.
• A U-Net model generated bone masks, and ground truth JSW was manually labeled.

Main Results:

• The REG method achieved a correlation coefficient of 0.88 and MSE of 0.02 mm.
• The SEG method achieved a correlation coefficient of 0.42 and MSE of 0.15 mm.
• The REG deep learning method demonstrated significantly higher accuracy and efficiency.

Conclusions:

• Deep learning approaches, specifically the REG method, show promising performance for automatic JSW measurement.
• Automated JSW quantification can facilitate more accurate and efficient assessment of hand OA.
• This technology can advance the quantitative analysis of distance features in medical imaging.