This project aims to classify patients with dysphagia with various machine learning and signal processing methods.

Extract Features from Hyoid Bone Location Annotations:

• ARMA (code)
• HMM + EM (code)

Classfication with HB Features (code):

• Logistic Regression
• SVM
• K-Neighbours
• GMM

Geniohyoid Muscle Classification (code):

• CNN to extract features from GH mask annotations
• LSTM for classification

---

Dataset can be found in GDrive, and metadata can be found in info_summary, data cleaning code segments can be found in utils.

• ~ 240 swallow ultrasound clips from healthy adults, healthy elderly, dysphagia patients;

• each from 100 - 600 frames, with a size of 1068 * 800 pixels^2;

• genihyoid muscle annotations, hyoid bone annotations, and swallowing event timestamp included.

Well-trained models can be found in GDrive

• File name indicates the number of BCU's used in CNNs-LSTM models

• Please feel free to contact Xinhui Yu via xinhuiyu at student dot ubc dot ca if you have any questions regarding this part.

Data authorship acknowledgments

• collected by Huberta Chan;

• annotated by Man Fung, Chan Chan, Audrey Cheung, Hugo;

• cleaned by Suri Feng.