✨ ScolioVis ✨
Automated Cobb Angle Measurement on Anterior-Posterior Spine X-Rays using Multi-Instance Keypoint Detection with Keypoint RCNN
📖 Summary
This repository serves as the compiled package of our undergraduate research for West Visayas State University - College of Information and Communications Technology entitled: "ScolioVis: Automated Cobb Angle Measurement on Anterior-Posterior Spine X-Rays using Multi-Instance Keypoint Detection with Keypoint RCNN"
In this repo, you can:
- [x] Read our research manuscript.
- [x] Understand our project.
- [x] Try our live, deployed demo on scoliovis.app
- [x] Try running our project locally.
- [x] Recreate our research/project.
📑 Contents
- :book: About
- :toolbox: Setup Instructions
- :ledger: Colab Notebooks
- :brain: Models
- :scroll: Important References
- :trophy: Acknowledgements
- :writing_hand: Cite our Project
:book: About
ScolioVis is a tool for automatically measuring the Cobb Angle—the standard measurement to assess Scoliosis. We harness the power of computer vision and machine learning to extract the cobb angles of an anterior-posterior Spine x-ray image. We built this application from the ground-up to an actual implementation in a usable web app.
We trained a Keypoint RCNN model on the SpineWeb Dataset 16. Boasting a performance of 93% AP at IoU=0.50 on object detections and 57% AP at IoU=0.50 on keypoint detections. The dataset is also part of the Accurate Automated Spinal Curvature Estimation (AASCE) 2019 Grand Challenge. Atlhough we aren't competing, using the performance metric of the challenge, we have achieved an SMAPE of 8.97 in cobb angle calculation which means ScolioVis as a whole is able to predict cobb angles at 91.03% accuracy.
A live deployed version of the application is available at scoliovis.app.
:toolbox: Setup Instructions
👉 Go to /src for detailed instructions on how to setup this project on your machine.
Source Repositories:
🎨scoliovis-web - Front End Repo⚡scoliovis-api - Back End Repo🏋️♂️scoliovis-training - Model Training Repository
:ledger: Colab Notebooks
:brain: Models
:scroll: Important References
-
Any Paper that uses the SpineWeb Dataset 16, must cite the following:
Wu, H., Bailey, Chris., Rasoulinejad, Parham., and Li, S., 2017.Automatic landmark estimation for adolescent idiopathic scoliosis assessment using boostnet. Medical Image Computing and Computer Assisted Intervention:127-135. Retrieved from http://www.digitalimaginggroup.ca/members/Shuo/MICCAIAutomatic.pdf
-
Our
📄 Thesis Manuscriptand📘 User Manualare available on /doc.
:trophy: Acknowledgements
| Name | Contributions |
|---|---|
| 👨🏫 Dr. Frank I. Elijorde | Our ever-supportive Thesis Adviser. |
| 🤵 Dr. Bobby D. Gerardo | Our ever-supportive Thesis Co-Adviser. |
| 👨🔬 Dr. Shuo Li | For giving us access to the SpineWeb Dataset 16. |
| 👩💼 Dr. Julie Ann Salido | For her expertise in computer vision research. |
| 👨💼 Mr. Paolo Hilado | For his expertise in data science research. |
| 👩⚕️ Dra. Jocelyn F. Villanueva | For her expertise in radiology. |
| 👨⚕️ Dr. Christopher Barrera | For his expertise in radiology. |
:writing_hand: Cite Our Project
Convert the following bibtex to
APA | MLA
(Credits to bibtex.online)
@article{article,
type={Bachelor's Thesis},
author = {Taleon, Carlo Antonio and Elizalde, Glecy and Rubinos, Christopher Joseph},
title = {ScolioVis: Automated Cobb Angle Measurement on Anterior-Posterior Spine X-Rays using Multi-Instance Keypoint Detection with Keypoint RCNN},
journal = {West Visayas State University College of Information and Communications Technology},
address = {La Paz, Iloilo City, Iloilo, Philippines},
year = {2023}
}
2023 © Taleon, Elizalde, Rubinos (BSCS4A) - West Visayas State University - College of Information and Communications Technology. All Rights Reserved.