|
--- |
|
license: apache-2.0 |
|
metrics: |
|
- mse |
|
- mae |
|
- r_squared |
|
tags: |
|
- medical |
|
model-index: |
|
- name: Vit-COVID-19-severity |
|
results: |
|
- task: |
|
name: Image Classification |
|
type: image-classification |
|
metrics: |
|
- name: MSE |
|
type: mse |
|
value: 0.5135470628738403 |
|
- name: MAE |
|
type: mae |
|
value: 0.5676153898239136 |
|
- name: R^ 2 |
|
type: r_squared |
|
value: 0.5378701686859131 |
|
pipeline_tag: image-classification |
|
--- |
|
|
|
## Diagnosing Medical Images for COVID-19 Severity (Regression task) |
|
|
|
The COVID-19 pandemic has strained healthcare resources and prompted discussion about how machine learning can alleviate physician burdens and contribute to diagnosis. Chest x-rays (CXRs) are used for diagnosis of COVID-19, but few studies predict the severity of a patient’s condition from CXRs. |
|
In this study, we produce a large COVID severity dataset by merging three sources and investigate the efficacy of transfer learning vision transformers (ViTs) in severity regression task. |
|
The ViT had the best regression results, with an MSE of 0.5135. Code developed in this project is available at https://github.com/stwhitfield/ift6759_project. |
|
|
|
## Model description |
|
|
|
We were fine-tuned the Vision Transformer (ViT) from HuggingFaces on our training set for 300 epochs using an AdamW optimizer with a learning rate of 0.00001 and a batch size of 10. |
|
Early stopping was applied to the model to prevent over-fitting. |
|
For ViT, training stopped after 70 epochs. The only difference between the preprocessed data in the ViT and the other models was that it normalized to a tensor image with mean and standard deviation during data augmentation. |
