Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transformer-based Time-Series Biomarker Discovery for COPD Diagnosis

Soham Gadgil, Joshua Galanter, Mohammadreza Negahdar

TL;DR

A transformer-based deep learning technique is designed to process the raw spirogram values along with demographic information and predict clinically-relevant endpoints related to COPD and makes the framework more interpretable by identifying parts of the spirogram that are important for the model predictions.

Abstract

Chronic Obstructive Pulmonary Disorder (COPD) is an irreversible and progressive disease which is highly heritable. Clinically, COPD is defined using the summary measures derived from a spirometry test but these are not always adequate. Here we show that using the high-dimensional raw spirogram can provide a richer signal compared to just using the summary measures. We design a transformer-based deep learning technique to process the raw spirogram values along with demographic information and predict clinically-relevant endpoints related to COPD. Our method is able to perform better than prior works while being more computationally efficient. Using the weights learned by the model, we make the framework more interpretable by identifying parts of the spirogram that are important for the model predictions. Pairing up with a board-certified pulmonologist, we also provide clinical insights into the different aspects of the spirogram and show that the explanations obtained from the model align with underlying medical knowledge.

Transformer-based Time-Series Biomarker Discovery for COPD Diagnosis

TL;DR

A transformer-based deep learning technique is designed to process the raw spirogram values along with demographic information and predict clinically-relevant endpoints related to COPD and makes the framework more interpretable by identifying parts of the spirogram that are important for the model predictions.

Abstract

Chronic Obstructive Pulmonary Disorder (COPD) is an irreversible and progressive disease which is highly heritable. Clinically, COPD is defined using the summary measures derived from a spirometry test but these are not always adequate. Here we show that using the high-dimensional raw spirogram can provide a richer signal compared to just using the summary measures. We design a transformer-based deep learning technique to process the raw spirogram values along with demographic information and predict clinically-relevant endpoints related to COPD. Our method is able to perform better than prior works while being more computationally efficient. Using the weights learned by the model, we make the framework more interpretable by identifying parts of the spirogram that are important for the model predictions. Pairing up with a board-certified pulmonologist, we also provide clinical insights into the different aspects of the spirogram and show that the explanations obtained from the model align with underlying medical knowledge.

Paper Structure

This paper contains 14 sections, 4 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Dataset and Label Generation
  4. Time-Series Transformer
  5. Results
  6. Clinical Insights
  7. Conclusion
  8. Appendix
  9. Related Work
  10. Data Preprocessing
  11. Details of Label Generation
  12. Model Selection and Training Details
  13. Additional Evaluation Metrics
  14. Results for Mortality endpoint

Figures (2)

  • Figure 1: The overall framework that we use. We start with the raw Volume-Time curve and convert it to a Flow-Volume curve. This curve is then smoothened using a Gaussian filter and patchified before passing as an input to the time-series transformer.
  • Figure 2: Overlaying the attention weights from the transformer encoder onto the Flow-Volume curve to visualize the importance of each patch for the COPD Risk and Exacerbation prediction tasks. The black rectangle in each curve represents the most important patch.