https://github.com/mariannerakic/tyche

https://github.com/mariannerakic/tyche

Last synced: about 1 month ago
JSON representation

• Host: GitHub
• URL: https://github.com/mariannerakic/tyche
• Owner: mariannerakic
• License: apache-2.0
• Created: 2024-01-22T21:47:21.000Z (12 months ago)
• Default Branch: main
• Last Pushed: 2024-04-11T17:14:32.000Z (9 months ago)
• Last Synced: 2024-04-12T02:55:02.807Z (9 months ago)
• Language: Python
• Size: 389 KB
• Stars: 12
• Watchers: 4
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE
  • Citation: CITATION.bib

Awesome Lists containing this project

• Awesome-Segment-Anything - [code

README

        
# Tyche: Stochastic In-Context Learning for Medical Image Segmentation

Official website of: [_Tyche: Stochastic In-Context Learning for Medical Image Segmentation_](http://arxiv.org/abs/2401.13650).  

[Marianne Rakic](https://mariannerakic.github.io/), [Hallee E. Wong](https://halleewong.github.io/), [Jose Javier Gonzalez Ortiz](https://josejg.com/),  

[Beth Cimini](https://www.broadinstitute.org/bios/beth-cimini), [John V. Guttag](https://people.csail.mit.edu/guttag/) \& [Adrian V. Dalca](https://www.mit.edu/~adalca/)

More code coming soon ... :)

## Abstract

Existing learning-based solutions to medical image segmentation have two important shortcomings. First, for each new segmentation task, usually a new model has to be trained or fine-tuned. This requires extensive resources and machine-learning expertise, and is therefore often infeasible for medical researchers and clinicians. Second, most existing segmentation methods produce a single deterministic segmentation mask for a given image. However, in practice, there is often considerable uncertainty about what constitutes the _correct_ segmentation, and different expert annotators will often segment the same image differently. We tackle both of these problems with _Tyche_, a model that uses a context set to generate stochastic predictions for previously unseen tasks without the need to retrain. Tyche differs from other in-context segmentation methods in two important ways.  

1. We introduce a novel convolution block architecture that enables interactions among predictions.

2. We introduce in-context test-time augmentation, a new mechanism to provide prediction stochasticity.

When combined with appropriate model design and loss functions, Tyche can predict a set of plausible diverse segmentation candidates for new or unseen medical images and segmentation tasks without the need to retrain.