Hey, we built MedSeg because medical image segmentation has been personally slow and inapproachable: download multi-GB software, fight CUDA drivers, hope the model runs on your scan, data management...etc. Having experience in helping research groups achieve their goals by using AI models iteratively, we wanted a full workspace where the data is stored in a tidy way and where you can train and run models. So that's what this is. Drag a CT or MRI into a project, run one of several segmentation runs on a GPU on our side, and get a 3D result in a few minutes. You can refine the masks in an in-browser editor, share with a colleague, or even open it in VR. For research groups, they can work on the manual segmentation part in an editor created by radiologists. Then, once a few cases are done, they can train a model and run it on new cases, then correct them, start training again in an iterative loop, using state of the art nnUnet v2 architecture. What it's not: a regulatory-cleared clinical tool. This is for research, education, and exploration — not diagnosis. The stack is TotalSegmentator + FastSurfer + around 6 other open models + a custom web-based editor, with WebGL, running on a small GPU box. Happy to answer anything technical or walk through specific use cases.

MedSeg

AI segmentation for CT and MRI, by radiologists.

AI segmentation for CT and MRI, by radiologists.