NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·1h ago
AI Research
arXiv CS.AI·Jun 29·5 min read
Distribution-based deep multiple instance learning for tumor proportion scoring in NSCLC
TL;DR
Non-small cell lung cancer (NSCLC) diagnosis often lacks precise tumor proportion scoring, which is crucial for treatment decisions. A novel distribution-based deep multiple instance learning (MIL) method was developed to enhance scoring accuracy.
✦ Why It Matters
Engineers can leverage this MIL approach to enhance automated tumor analysis in clinical settings.
Key Takeaways
- Adopt the distribution-based MIL framework for TPS scoring.
- Utilize ZIBeta modeling for improved prediction accuracy.
- Focus on histopathological feature extraction from image patches.
- Leverage slide-level TPS scores for training models.
How It Works
The framework consists of two models: the first extracts features from individual image patches using a multiclass classification network, while the second aggregates these features through a MIL approach to predict a ZIBeta distribution. This distribution captures the overall TPS probability for the slide, addressing challenges posed by zero class images.