Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Computing | en_US |
| dc.contributor.advisor | Guo, Song (COMP) | en_US |
| dc.contributor.advisor | Liu, Yan (COMP) | en_US |
| dc.creator | Xu, Zhenda | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13506 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Development and evaluation of edge devices for scoliosis analysis | en_US |
| dcterms.abstract | Adolescent idiopathic scoliosis(AIS) has become a common spinal disorder among teenagers. The traditional evaluation method for scoliosis is mainly using X-ray equipment. However, X-ray equipment produces radiation, which can elevate the risk of cancer with long-term exposure(especially for adolescents), making it unsuitable for long-term follow-up of scoliosis. Additionally, traditional scoliosis screening methods are time-consuming and dependent on experienced personnel, leading to low positive predictive values and potentially unnecessary referrals and radiation exposure. Furthermore, early stage scoliosis is often accompanied by abnormal body posture, making it difficult to distinguish and monitor. To achieve radiation-free early screening and monitoring of scoliosis and abnormal posture, we propose multiple methods for automated screening and monitoring of scoliosis and abnormal posture using edge devices. These edge devices consist of three-dimensional(3D) imaging devices(infrared RGB-D camera) and two-dimensional(2D) imaging device (mobile phone). | en_US |
| dcterms.abstract | However, assessing and screening for scoliosis based on 2D or 3D data of the human back is not straightforward in real life. It requires an understanding of the spine’s anatomical structure, the medical anatomical features of the human back, biomechanics, and the correlation between 2D and 3D data of the human back and above features. To build a system for radiation-free assessment and screening of scoliosis based on 2D or 3D data of the human back, we summarize three core challenges in system construction and our main contributions to addressing these challenges in section 1.3. More precisely, by conducting a comprehensive background review of scoliosis and related analysis system in chapter 2, we intend to build the systems in three aspects: (1) In chapter 3, curvature is calculated from 3D point clouds of the human back to locate anatomical landmarks, and a correlation model integrating spinal anatomy and biomechanics is established to precisely infer the spine’s 3D curve. Utilizing the industry-standard for full-spine X-ray imaging, we validated the accuracy of spinal Cobb angle estimation from 3D back point clouds, leading to the development of a commercial system; (2) In chapter 4, a database of 2D back images and 3D point clouds is created, confirming their high correlation on key metric Axial Trunk Rotation(ATR); and (3) In chapter 5, the relationship between 2D postural features and scoliosis X-ray imaging is explored and validated, with a deep learning-based framework for landmark points and topological structures addressing the challenge of accurately screening scoliosis and abnormal posture using 2D images. | en_US |
| dcterms.abstract | In chapter 3, to explore the correlation between the 3D point cloud of the human back and scoliosis and abnormal posture, we established a database of human 3D back point clouds and the corresponding X-ray images. By identifying anatomical landmarks on 3D back point clouds and correlating them with the spine’s midline, we derived a 3D model of the spine’s midline. The results show that the proposed method can extract anatomic landmark points and evaluate scoliosis accurately (average Root Mean Square Error of anatomic landmark points extraction is around 5mm and Cobb angle estimation is around 3° ), which is feasible and promising. In this chapter, we validate the strong correlation between 3D back point clouds and scoliosis, and we proposed a Kinect based low cost, easy to use, non radiation, and high accuracy method to quickly reconstruct the 3D shape of the spine, which can be used to evaluate spinal deformation. This method is effective for scoliosis assessment, but 3D point cloud images require additional RGB-D equipment to accurately evaluate scoliosis and posture, which will limit usage scenarios. If the correlation between 2D human back images and 3D back point cloud images can be verified, then 2D back images can be used to evaluate scoliosis and abnormal body posture. | en_US |
| dcterms.abstract | In chapter 4, to better explore the correlation of features between 2D back images and corresponding 3D back point cloud images, we develop an efficient deep learning-based framework to allow a large-scale screening for scoliosis using only one 2D unclothed human back image without any X-radiation equipment. We classify the normal individual and abnormal scoliosis using the ATR value as the classification label, calculated from the human back 3D point cloud. Our accuracy in the task of AIS classification reaches 81.3%, far exceeding the accuracy of visual observation by an experienced doctor (65.1%), which can be used as a remote preliminary scoliosis screening method. This chapter verifies that 2D and 3D images of the human back on concave convex features (such as ATR) have strong correlation, which can lay the foundation for inferring the features of 3D point cloud images based on 2D images of the back in the future and it also validates the feasibility of screening for scoliosis using an unclothed back image, thus empowering users. However, while using a single back 2D image can help distinguish whether a person suffers from scoliosis, it is not capable of distinguishing the severity of scoliosis, especially considering that scoliosis often accompanies abnormal posture, yet abnormal posture does not necessarily indicate scoliosis. Furthermore, the inability to quantify mild scoliosis or abnormal posture may hinder regular home monitoring, potentially leading to missed opportunities for optimal intervention. | en_US |
| dcterms.abstract | In chapter 5, to better distinguish the severity of scoliosis and identify the correlation between 2D human body posture features and the 2D curve of scoliosis, we propose a novel approach. we propose a set of back feature points and network topology based on deep learning algorithms. We establish a database with labeled 2D back images and corresponding whole-spine standing posterior-anterior X-ray images and propose a new network topology of the 2D back image to localize the back landmarks. With only an unclothed back image, this system can automatically classify normal and abnormal posture and scoliosis with an overall classification accuracy of 88.1%. | en_US |
| dcterms.abstract | In order to improve the classification rate of scoliosis and posture abnormalities, as well as to better quantify the risk severity of scoliosis and achieve quantifiable risk progression monitoring, we propose the use of three back images for analysis. We have developed an online mini-program where users only need to upload three images to achieve precise screening and monitoring of scoliosis and abnormal posture at home. Simultaneously, we calculate parameters of the human back and the ATR angle for quantifiable daily monitoring. The optimized system has a sensitivity of 96% and a specificity of 89% for scoliosis, far exceeding the accuracy of experienced doctors (sensitivity of 81% and specificity of approximately 86%). The probability of misjudging abnormal posture as scoliosis is 8%, and the probability of misjudging scoliosis as abnormal posture is 5%. In this chapter, we verify that a new set of feature points and network topology based on deep learning algorithms can effectively describe the correlation between 2D body features and scoliosis. Furthermore, based on this structure, we have developed a mobile, cost-effective, accurate, and radiation-free screening and monitoring system for the screening and daily monitoring of scoliosis. | en_US |
| dcterms.abstract | To date, we have successfully deployed and commercialized the aforementioned systems, garnering accolades from numerous international innovation awards. We have catered to over 20 hospitals, rehabilitation institutions, and insurance agencies, facilitating the screening of more than 300,000 adolescents. | en_US |
| dcterms.extent | xix, 129 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2024 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Scoliosis in children -- Diagnosis -- Data processing | en_US |
| dcterms.LCSH | Scoliosis -- Diagnosis -- Data processing | en_US |
| dcterms.LCSH | Artificial intelligence -- Medical applications | en_US |
| dcterms.LCSH | Medical screening | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13506