Automated Detection of the Kyphosis Angle Using a Deep Learning Approach: A Cross-Sectional Study on Young Adults

Area of Science:

• Medical imaging
• Deep learning in healthcare
• Biomechanical analysis

Background:

• Thoracic kyphosis is influenced by lifestyle and stress, impacting posture.
• Current diagnostic methods include costly radiological and varied non-radiological techniques.
• There's a need for efficient, non-invasive methods for kyphosis assessment.

Purpose of the Study:

• To develop a decision support system for automatic thoracic kyphosis angle measurement.
• To eliminate radiation exposure associated with radiological measurements.
• To reduce the time and resources required for kyphosis evaluation.

Main Methods:

• Utilized a convolutional neural network (CNN) for image segmentation.
• Automatically identified T1 and T12 vertebrae for angle calculation.
• Features were based on expert-marked thoracic kyphosis measurements.

Main Results:

• Achieved high intra-class consistency (ICC > 0.95, p < 0.05).
• Demonstrated high internal consistency reliability (Cronbach's α = 0.947).
• The automated system provided accurate and reliable measurements.

Conclusions:

• The developed algorithm offers a reliable automated solution for thoracic kyphosis measurement.
• The system minimizes radiation exposure and measurement duration.
• This approach enhances clinical diagnosis and evaluation of posture.