Sign in Subscribe
Machine learning

MIT Researchers Develop AI-Powered Tool for Rapid Segmentation of Medical Images

Strategic Insights

— 2 min read

MIT researchers have created an artificial intelligence-powered tool that enables medical researchers to rapidly segment new biomedical imaging datasets by clicking, scribbling, and drawing boxes on the images. This new AI model, called MultiverSeg, uses user interactions to predict segmentation and can accurately segment each new image without user input. The tool has the potential to accelerate studies of new treatment methods, reduce the cost of clinical trials and medical research, and improve the efficiency of clinical applications such as radiation treatment planning.

Key Takeaways:

  • The new AI model, MultiverSeg, combines the best of interactive segmentation and machine-learning-based approaches to segment medical images.
  • Users can rapidly segment new biomedical imaging datasets by clicking, scribbling, and drawing boxes on the images.
  • The model uses user interactions to predict segmentation and can accurately segment each new image without user input.
  • MultiverSeg requires less user input with each image, with the number of clicks decreasing to zero as the model improves its predictions.
  • The tool can segment an entire dataset without repeating work for each image, reducing the need for manual segmentation.
  • Users can also correct the model's predictions and iterate until it reaches the desired level of accuracy.
  • Unlike other tools, MultiverSeg does not require a presegmented image dataset for training, reducing the need for machine-learning expertise and extensive computational resources.
  • The researchers designed the model's architecture to use a context set of any size, giving it flexibility to be used in a range of applications.
  • The tool outperformed state-of-the-art tools for in-context and interactive image segmentation, requiring fewer clicks and achieving higher accuracy.
  • MultiverSeg has the potential to accelerate studies of new treatment methods, reduce the cost of clinical trials and medical research, and improve the efficiency of clinical applications such as radiation treatment planning.

Statistics:

  • By the 9th new image, MultiverSeg needed only 2 clicks from the user to generate a segmentation more accurate than a model designed specifically for the task.
  • The tool required roughly 2/3 the number of scribbles and 3/4 the number of clicks compared to the researchers' previous system, while achieving 90% accuracy.
  • The model's interactivity enables users to make corrections to the model's prediction, iterating until it reaches the desired level of accuracy.
  • MultiverSeg has the potential to accelerate studies of new treatment methods by enabling researchers to conduct studies they were prohibited from doing before due to the lack of an efficient tool.

Sources:

  • [Hallee Wong, et al., "MultiverSeg: A Context Set-Based Interactive Image Segmentation Model" (MIT CSAIL, 2025)] (no date provided)
  • [International Conference on Computer Vision (2025)] (no date provided)
  • [Cambridge, MA - NewsRx LLC, "MIT Researchers Develop AI-Powered Tool for Rapid Segmentation of Medical Images" (2025)] (no date provided)

Read more