Researchers from MIT, the Broad Institute of MIT and Harvard, and Massachusetts General Hospital have developed an AI tool called Tyche that can generate multiple plausible segmentations for medical images. Unlike existing models that provide only one answer, Tyche allows users to select the most appropriate segmentation based on their purpose. It does not require retraining and can be used for various tasks like identifying lesions in lung X-rays or anomalies in brain MRIs. The researchers modified a neural network architecture to build Tyche, and testing showed that its predictions were diverse and often better than baseline models. The research was funded by the National Institutes of Health and Quanta Computer.

--- Send in a voice message: https://podcasters.spotify.com/pod/show/tonyphoang/message