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

TL;DR

This study presents a deep learning system to automatically measure thoracic kyphosis in young adults, avoiding radiation and saving time.

Contribution

A deep learning-based automated system for measuring thoracic kyphosis with high reliability is introduced.

Findings

The system achieved intra-class consistency with ICC > 0.95 (p < 0.05).

Internal consistency reliability was measured at Cronbach’s α = 0.947.

The method avoids radiation exposure and reduces measurement time.

Abstract

Objectives: In healthy young adults, thoracic kyphosis can be attributed to a number of factors, including a sedentary lifestyle, stress, poor posture, activity and daily habits, muscle pain, fatigue, and anxiety. In regard to clinical diagnosis and evaluation methods, high-cost radiological measurements and a variety of non-radiological clinical methods are employed. In this study, a decision support system that performs automatic thoracic kyphosis angle measurements has been developed with the objective of avoiding exposure of the human body to radiation and reducing the time required for measurements. Methods: The features were determined with reference to the thoracic kyphosis measurements that were manually marked by the expert on the subjects. The kyphosis angle was calculated by automatically identifying the T1 and T12 points through image segmentation using a convolutional neural network (CNN), which is a type of deep learning algorithm. Results: Intra-class consistency of ICC > 0.95 (p < 0.05) and internal consistency reliability of Cronbach’s α = 0.947 are obtained. Conclusions: The results demonstrate that the proposed algorithm exhibits high intra-class consistency and high internal consistency reliability to provide an automated thoracic kyphosis angle measurement system.

Figures