Related Experiment Video
Updated: May 11, 2026
Intact Histological Characterization of Brain-implanted Microdevices and Surrounding Tissue
Published on: February 11, 2013
TriDeNT : Triple deep network training for privileged knowledge distillation in histopathology
Lucas Farndale1, Robert Insall2, Ke Yuan3
1School of Cancer Sciences, University of Glasgow, Scotland, UK; Cancer Research UK Scotland Institute, Scotland, UK; School of Computing Science, University of Glasgow, Scotland, UK; School of Mathematics and Statistics, University of Glasgow, Scotland, UK.
We developed TriDeNT, a new self-supervised method to improve computational pathology models by using extra data during training. This approach significantly boosts model performance on routine inputs, outperforming existing methods.
Area of Science:
- Computational pathology
- Machine learning in digital pathology
- Biomedical image analysis
Background:
- Computational pathology models often fail to leverage all available data.
- Highly informative data like immunohistochemistry (IHC) and spatial transcriptomics are typically excluded during inference.
- This limits the performance and generalizability of existing models.
Purpose of the Study:
- To introduce TriDeNT, a novel self-supervised method for computational pathology.
- To enable models to utilize 'privileged' data (available during training but not inference).
- To enhance the performance of computational pathology models on downstream tasks.
Main Methods:
- Developed TriDeNT, a self-supervised learning framework.
- Utilized privileged data, including IHC, spatial transcriptomics, and expert nuclei annotations, during model training.
- Evaluated TriDeNT against state-of-the-art methods on various paired data settings.
Main Results:
- TriDeNT significantly outperformed baseline and state-of-the-art methods across all tested downstream tasks.
- Observed performance improvements of up to 101% using TriDeNT.
- Demonstrated effective knowledge distillation from scarce or costly data into models trained for routine inputs.
Conclusions:
- TriDeNT offers a powerful new approach to incorporate privileged information in computational pathology.
- This method enhances model performance by effectively learning from data not available at inference.
- TriDeNT paves the way for more robust and accurate computational pathology tools.
