Automated 3D brain tumor segmentation in MRI — University Hospital Tübingen × statworx

University Hospital Tübingen and the data science consultancy statworx developed a computer vision model that automatically measures the three-dimensional extent of brain tumors in MRI images, according to a statworx case study. The work was a five-month pro bono project. "What impressed me most was that good segmentation accuracy was achieved despite the heterogeneity of the data due to anatomical variants, different tumor sizes and different imaging modalities," said PD Dr. med. Florian Grimm (translated from German). He added that the hospital would very much like to continue the collaboration, ideally as a third-party-funded project. An accurate picture of how a brain tumor is growing is central to treatment decisions, but doctors usually have to draw the tumor's extent by hand in MRI images, which takes a long time. The interdisciplinary team at the hospital worked with statworx's computer vision and biomedicine specialists to build a segmentation model, using PyTorch, Python and TensorBoard, that recognizes the spatial extent of tumors in MRI scans. Clinicians can now measure tumors in three dimensions automatically and use the result to assess growth over time and choose therapy. statworx says this improves accuracy and, above all, speeds up correct diagnosis and treatment, but the case study gives no segmentation scores or time savings.