Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

In-context learning for medical image segmentation

Eichi Takaya, Shinnosuke Yamamoto

TL;DR

This work tackles the annotation burden in sequential medical image segmentation by introducing In-context Cascade Segmentation (ICS), a method built on the UniverSeg few-shot framework. ICS propagates segmentation across a volume by iteratively updating the support set with each predicted slice, enabling bidirectional information flow and improved boundary consistency without retraining. Evaluated on HVSMR MRI data, ICS shows significant gains in several cardiac regions and demonstrates the importance of initial support size and placement, while highlighting trade-offs in computation and potential over-segmentation. Overall, ICS offers a practical approach to scalable, accurate sequential medical image segmentation with reduced labeling effort, suitable for clinical and research deployment after broader validation.

Abstract

Annotation of medical images, such as MRI and CT scans, is crucial for evaluating treatment efficacy and planning radiotherapy. However, the extensive workload of medical professionals limits their ability to annotate large image datasets, posing a bottleneck for AI applications in medical imaging. To address this, we propose In-context Cascade Segmentation (ICS), a novel method that minimizes annotation requirements while achieving high segmentation accuracy for sequential medical images. ICS builds on the UniverSeg framework, which performs few-shot segmentation using support images without additional training. By iteratively adding the inference results of each slice to the support set, ICS propagates information forward and backward through the sequence, ensuring inter-slice consistency. We evaluate the proposed method on the HVSMR dataset, which includes segmentation tasks for eight cardiac regions. Experimental results demonstrate that ICS significantly improves segmentation performance in complex anatomical regions, particularly in maintaining boundary consistency across slices, compared to baseline methods. The study also highlights the impact of the number and position of initial support slices on segmentation accuracy. ICS offers a promising solution for reducing annotation burdens while delivering robust segmentation results, paving the way for its broader adoption in clinical and research applications.

In-context learning for medical image segmentation

TL;DR

This work tackles the annotation burden in sequential medical image segmentation by introducing In-context Cascade Segmentation (ICS), a method built on the UniverSeg few-shot framework. ICS propagates segmentation across a volume by iteratively updating the support set with each predicted slice, enabling bidirectional information flow and improved boundary consistency without retraining. Evaluated on HVSMR MRI data, ICS shows significant gains in several cardiac regions and demonstrates the importance of initial support size and placement, while highlighting trade-offs in computation and potential over-segmentation. Overall, ICS offers a practical approach to scalable, accurate sequential medical image segmentation with reduced labeling effort, suitable for clinical and research deployment after broader validation.

Abstract

Annotation of medical images, such as MRI and CT scans, is crucial for evaluating treatment efficacy and planning radiotherapy. However, the extensive workload of medical professionals limits their ability to annotate large image datasets, posing a bottleneck for AI applications in medical imaging. To address this, we propose In-context Cascade Segmentation (ICS), a novel method that minimizes annotation requirements while achieving high segmentation accuracy for sequential medical images. ICS builds on the UniverSeg framework, which performs few-shot segmentation using support images without additional training. By iteratively adding the inference results of each slice to the support set, ICS propagates information forward and backward through the sequence, ensuring inter-slice consistency. We evaluate the proposed method on the HVSMR dataset, which includes segmentation tasks for eight cardiac regions. Experimental results demonstrate that ICS significantly improves segmentation performance in complex anatomical regions, particularly in maintaining boundary consistency across slices, compared to baseline methods. The study also highlights the impact of the number and position of initial support slices on segmentation accuracy. ICS offers a promising solution for reducing annotation burdens while delivering robust segmentation results, paving the way for its broader adoption in clinical and research applications.

Paper Structure

This paper contains 17 sections, 6 equations, 8 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Medical Image Segmentation
  4. In-context Learning
  5. Semi-supervised Learning
  6. Methods
  7. Problem Definition
  8. UniverSeg
  9. In-context Cascade Segmentation (ICS)
  10. Experiments
  11. Dataset
  12. Experimental Settings
  13. Evaluation Metrics
  14. Implementation
  15. Results
  16. ...and 2 more sections

Figures (8)

  • Figure 1: UniverSeg for sequential inference
  • Figure 2: An overview of in-contest cascade segmentation
  • Figure 3: Box plots of the DSC for the baseline method (green) and ICS (orange) across each anatomical region. The boxes represent the interquartile range and the horizontal line within each box indicates the median.
  • Figure 4: Segmentation results for the PA region in selected slices of a patient’s volume. From top to bottom: the raw image, the prediction by the baseline method, the prediction by ICS, and the ground truth label. Columns show different slice positions.
  • Figure 5: Segmentation results for the LV region in selected slices of a patient’s volume. From top to bottom: the raw image, the prediction by the baseline method, the prediction by ICS, and the ground truth label. Columns show different slice positions.
  • ...and 3 more figures