Computer Vision News 36 Nature Communications Paper The CVPR 2024 challenge seeks universal medical image segmentation models that are deployable on laptops or other edge devices without reliance on GPUs. standard manual workflow. “That’s a huge boost of efficiency,” he points out. “You can imagine how we could use MedSAM to help doctors do a much faster job in segmentations. This is the biggest pain point for physicians, radiologists, and CT scan readers because they spend significant time doing manual contouring.” Looking at broader applications of MedSAM in clinical practice, the team points to its potential for revolutionizing tumor measurement - the RECIST criteria. RECIST is a standard approved by the FDA to assess tumor progression. They choose two points representing the longest diameter of tumors in MRI or CT scans and then use this length as the metric to indicate tumor progression before and after treatment. Outside of this paper, Jun says his research is focused on employing advanced AI models in clinical practice for more personalized treatment and to improve patient outcomes. Plans are underway to organize challenges at CVPR and MICCAI this year, hoping to optimize MedSAM for use on a laptop and further develop it for generalized cancer segmentation in whole-body CT scans. “I really appreciated the support of Professor Bo Wang, my collaborators and our institutions for their computing and resources for me to do this work,” he adds. “Now, we want to gather the efforts from the whole community to make the model deployable in clinical practice!”
RkJQdWJsaXNoZXIy NTc3NzU=