Medical imaging is a complex field where interpretation of results can be difficult.
AI models can assist doctors by analyzing images that may reveal abnormalities indicative of disease.
But there’s a problem. In reality, when a medical image has multiple interpretations, these AI models typically present a single solution.
For example, if you ask five experts to outline an area of interest, such as a small lump on a lung scan, you might end up with five different pictures. For example, experts may all have their own opinions about where a chunk begins and ends.
To solve this problem, researchers from MIT, MIT Harvard Broad Institute, and Massachusetts General Hospital created Tyche, an AI system that accommodates the ambiguity of medical image segmentation.
Segmentation involves labeling specific pixels in a medical image that represent important structures, such as organs or cells.
Marianne Rakic, PhD candidate in computer science at MIT and study, explains. “Having options can help you make decisions. “Just the presence of uncertainty in a medical image can influence someone’s decision, so it’s important to take that uncertainty into account.”
Tyche, named after the Greek goddess of opportunity, captures ambiguity by generating multiple possible segmentations for a single medical image.
Each segmentation highlights a slightly different region, allowing users to choose the region that best suits their needs.
Rakik says MIT News“If you can print out multiple candidates and see if they are different from each other, you can really get an advantage.”
So how does Tyche work? Let’s break this down into four simple steps.
- Learn through examples: Users provide Tyche with a small set of example images, called “context sets,” that demonstrate the segmentation operation they want to perform. These examples may include images segmented by different human experts, helping the model understand the task and the potential for ambiguity.
- Neural Network Tuning: Researchers modified the standard neural network architecture to enable Tyche to handle uncertainty. They coordinated the layers of the network so that the potential splits generated at each step could “communicate” with each other and with the context set of examples.
- various possibilities: Tyche is designed to output multiple predictions based on a single medical image input and set of contexts.
- rewarding quality: The training process has been adjusted to reward Tyche for making the best predictions possible. If a user requests 5 predictions, he or she will see all 5 medical image segmentations generated by Tyche. One might be better.
One of Tyche’s greatest strengths is its adaptability. You can perform new segmentation tasks without having to retrain from scratch.
Typically, AI models for medical image segmentation use neural networks that require extensive training on large datasets and machine learning expertise.
In contrast, Tyche can be used “out of the box” for a variety of tasks, from spotting lung lesions on X-rays to identifying brain abnormalities on MRIs.
Numerous studies have been conducted in the field of AI medical imaging, including breakthroughs. breast cancer screening And AI diagnosis is matches or beat the doctor When interpreting images
In the future, the research team plans to explore using a more flexible set of contexts, including text or multiple types of images.
They also want to improve Tyche’s worst-case predictions and develop ways for the system to recommend the best split candidates.