Uncertainty-aware automatic TNM staging classification for [18F] Fluorodeoxyglucose PET-CT reports for lung cancer utilising transformer-based language models and multi-task learning
2024

Automatic TNM Staging Classification for Lung Cancer PET-CT Reports

Sample size: 2498 publication 10 minutes Evidence: high

Author Information

Author(s): Barlow Stephen H., Chicklore Sugama, He Yulan, Ourselin Sebastien, Wagner Thomas, Barnes Anna, Cook Gary J.R.

Primary Institution: King’s College London

Hypothesis

Can a multi-task transformer-based classifier effectively extract TNM staging information from PET-CT reports for lung cancer?

Conclusion

The TNMu classifier successfully extracts TNM staging information from PET-CT reports, demonstrating that multi-task approaches yield better performance and efficiency.

Supporting Evidence

  • A multi-task approach with GatorTron achieved an overall accuracy of 84% on the internal test dataset.
  • Performance on individual TNM tasks approached expert performance with macro average F1 scores of 0.91, 0.95, and 0.90 respectively on external data.
  • The model's uncertainty classification task achieved an F1 score of 0.77.
  • External validation demonstrated the model's ability to generalize across different reporting styles.

Takeaway

This study created a smart computer program that can read lung cancer PET-CT reports and tell if there are tumors, nodes, or metastasis, helping doctors make better decisions.

Methodology

The study involved annotating 2498 PET-CT reports and training a multi-task classifier using eleven pre-trained language models.

Potential Biases

Potential bias exists due to the single annotator's influence on the model's training data.

Limitations

The study was limited by the reliance on a single annotator for internal data, which may introduce bias.

Statistical Information

P-Value

0.05

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.1186/s12911-024-02814-7

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication