Symmetric Perception and Ordinal Regression for Detecting Scoliosis Natural Image

TL;DR

A dual-path scoliosis detection network with two main modules: symmetric feature matching module (SFMM) and ordinal regression head (ORH) that outperforms state-of-the-art methods as well as human performance, which provides a promising and economic solution to wide-range scoliosis screening.

Abstract

Scoliosis is one of the most common diseases in adolescents. Traditional screening methods for the scoliosis usually use radiographic examination, which requires certified experts with medical instruments and brings the radiation risk. Considering such requirement and inconvenience, we propose to use natural images of the human back for wide-range scoliosis screening, which is a challenging problem. In this paper, we notice that the human back has a certain degree of symmetry, and asymmetrical human backs are usually caused by spinal lesions. Besides, scoliosis severity levels have ordinal relationships. Taking inspiration from this, we propose a dual-path scoliosis detection network with two main modules: symmetric feature matching module (SFMM) and ordinal regression head (ORH). Specifically, we first adopt a backbone to extract features from both the input image and its horizontally flipped image. Then, we feed the two extracted features into the SFMM to capture symmetric relationships. Finally, we use the ORH to transform the ordinal regression problem into a series of binary classification sub-problems. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods as well as human performance, which provides a promising and economic solution to wide-range scoliosis screening. In particular, our method achieves accuracies of 95.11% and 81.46% in estimation of general severity level and fine-grained severity level of the scoliosis, respectively.

Figures (9)

• Figure 1: Example images with different Cobb angles cobb1948outline at different general severity levels of scoliosis. There are four general severity levels: normal, minor, moderate, and severezhang2015principlesyang2019developmentchen2022computerized. By comparing the images in the upper and lower rows, we can find that the more severe the scoliosis, the more asymmetrical the back shape will be.
• Figure 2: The architecture of our network. We use the visual attention network (VAN) guo2022visual as the backbone. The input of the dual-path network consists of two images, one is the original back image and the other is the horizontally flipped image. After being fed to the weight-sharing backbone, the features $\mathbf{F}$ and $\mathbf{F}^{f}$ are obtained. Then, $\mathbf{F}$ and $\mathbf{F}^{f}$ are fed to symmetric feature matching module (SFMM) for symmetric relationship perception and feature fusion. Finally, we use ordinal regression head (ORH) to transform the multi-class classification task into an ordinal regression task and obtain the final prediction results.
• Figure 3: The number of samples for different fine-grained scoliosis severity levels in our constructed dataset USTC&SYSU-Scoliosis. Each fine-grained severity level contains a range of $5$ Cobb angle degrees.
• Figure 4: Comparison with experts' results in general scoliosis severity level estimation on the fifth fold of USTC&SYSU-Scoliosis. The left two confusion matrices are the results of two experts, while the rightmost confusion matrix is the result of our method.
• Figure 5: Comparison with experts' results in fine-grained scoliosis severity level estimation on the fifth fold of USTC&SYSU-Scoliosis. The left two confusion matrices are the results of two experts, while the rightmost confusion matrix is the result of our method.