ACE-V1.1 (MRI brain tumor detection)
Disclaimer: Metrics mentioned below are author-reported from the upstream model card and are not independently verified by this site. This is not medical advice.
ACE-V1.1 is a YOLOv11 (ultralytics) fine-tune intended for brain tumor detection in MRI scans. The model card links to an associated BRISC 2025 dataset/paper (arXiv:2506.14318) and includes practical “how to run it” details: it’s trained for 640x640 inputs and the author recommends a confidence threshold of 0.466.
The card also reports evaluation numbers (e.g., mAP50 ~0.9 on a held-out test, plus a claimed 1.00 “background specificity” focused on reducing false positives in healthy tissue). These are author-reported values from the upstream model card, not independently verified by this site, and they may not generalize beyond the referenced dataset/paper. These reported figures, especially the claimed 1.00 background specificity, should be treated as dataset-specific and not indicative of guaranteed performance on other scanners, institutions, or populations.
This model and this page do not provide medical advice and are not a substitute for professional diagnosis or treatment. This is not a clinically validated device; any real-world use should be treated strictly as research and validated through the usual institutional and regulatory processes. Any deployment in a clinical or patient-facing context will typically require separate regulatory clearance and legal review; this model card does not imply such approval. The weights are shared under CC-BY-NC-4.0, so usage is constrained to non-commercial contexts, including for any services or products built on top of this model.
What to try first: run it on a small labeled MRI set you trust, tune the confidence threshold for your desired false-positive rate, and inspect failure cases manually (missed small lesions vs spurious detections). If you’re considering any clinical workflow integration, treat this as research and validate it through your institution’s normal review process.
Quick stats from the listing feed at the time of writing: pipeline: object-detection · 1 like · 289 downloads.
Source listing: https://huggingface.co/models?sort=modified