Original title: Generalizable Brain Tumor Segmentation with Self-Training and Tumor-Aware Deformations
Henrique Zan Grande, Jeovane Honorio Alves, Rayson Laroca, Andre Gustavo Hochuli · September 2026
Finding brain tumor boundaries on medical scans is hard when tumors look very different across diverse patients. Researchers tackled this by building an artificial intelligence model that learns from both labeled scans and unlabeled scans. Their system generates its own guesses, called pseudo-labelsPredictions made on unlabeled data that are treated as true labels during training to expand the dataset., for scans that lack human markings. It also uses a tumor-aware deformation technique. This strategy bends and stretches the tumor shapes to make diverse practice examples while leaving the healthy brain parts untouched. Testing showed that combining these fake labels with the tumor-stretching method worked better than learning from only labeled scans across every tumor region. The two techniques complemented each other, helping the tool more accurately identify the whole tumor and its internal sub-regions.
Imagine packing for a hike where the trail conditions change unexpectedly based on your route. If you pack only for sunny weather, you'll be unprepared for rain or mud. In this paper, the system is trained to handle variable tumor shapes by adding controlled deformations, much like packing versatile gear to adapt to different trail conditions.
Accurate brain tumor segmentation helps doctors plan treatments for cancer patients. Making it work across varied patient scans means more people get reliable results, regardless of their specific tumor characteristics.
The abstract does not state explicit limitations, but the method is evaluated only on the BraTS-GoAT validation set, so its generalization to other datasets or tumor types is not confirmed. It relies on the nnU-Net framework and specific augmentation techniques, which may not be optimal for all scenarios.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Generalizable Brain Tumor Segmentation with Self-Training and Tumor-Aware Deformations — Henrique Zan Grande, Jeovane Honorio Alves, Rayson Laroca et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.