New MIT AI System Makes Image Segmentation Faster And Easier

MIT researchers have developed an AI system that helps medical experts to accelerate their research through rapid image analysis of medical data.

The tool, called MultiverSeg, allows scientists to mark specific image areas, by simply clicking or scribbling, and the system uses this information to generate predictions for upcoming results.

MIT explains that the initial and most labor-intensive process in clinical research requires medical image annotation, also known as segmentation. For example, to study how the hippocampus in the brain changes with age, researchers must manually trace it across various scans.

“Many scientists might only have time to segment a few images per day for their research because manual image segmentation is so time-consuming. Our hope is that this system will enable new science by allowing clinical researchers to conduct studies they were prohibited from doing before because of the lack of an efficient tool,” said Hallee Wong, lead author and graduate student in electrical engineering and computer science.

Unlike previous systems, MultiverSeg does not require researchers to train it with large presegmented datasets. The system creates a “context set” from past segmented images and uses them to improve future predictions. The researchers explain that the system requires almost no user interaction as time progresses.

The researchers tested MultiverSeg against state-of-the-art tools, and found it required fewer clicks and scribbles, and produced more accurate results. Indeed, the AI system required only one or two manual segmentations of X-rays before it could make accurate predictions for the remaining areas.

“With MultiverSeg, users can always provide more interactions to refine the AI predictions. This still dramatically accelerates the process because it is usually faster to correct something that exists than to start from scratch,” Wong explained.

The team plans to test the system in clinical settings, with hopes that it could also improve efficiency in areas such as radiation treatment planning.