Automated Brain Tumor Segmentation via YOLOv8-Derived Spatial Prompts for the Segment Anything Model
Computational Discovery and Intelligent Systems (CDIS) · 2026
DOI: 10.66279/vv8n5p41 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computational Discovery and Intelligent Systems (CDIS), 2026. Verfasst von Maha Alabsi und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 25 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Brain tumor segmentation from magnetic resonance imaging~(MRI) is a critical step in the diagnosis and treatment planning of intracranial malignancies. Although supervised convolutional networks achieve strong benchmark performance within their training distribution, they exhibit limited transferability across acquisition protocols. Conversely, foundation models such as the Segment Anything Model~(SAM) encode rich visual representations but produce unreliable masks in the absence of accurate spatial guidance. The present work introduces a fully automated, end-to-end pipeline that couples YOLOv8 object detection with SAM-based segmentation without modifying the parameters of either network. A lightweight preprocessing stage comprising skull stripping and Contrast Limited Adaptive Histogram Equalization~(CLAHE) conditions each MRI slice; the resulting image is forwarded to a trained YOLOv8 detector whose highest confidence bounding box is passed directly to SAM's prompt encoder as the sole spatial cue.Evaluation on 1,226 held-out images from the publicly available Cheng et. al. benchmark, partitioned by patient identity to prevent data leakage, yields a mean Dice Similarity Coefficient~(DSC) of 0.8153 pm 0.032$ and a mean Intersection over Union (IoU) of 0.7136 pm 0.028, with a total inference latency of 473.76 ms per image on an NVIDIA~T4 GPU. An ablation study confirms that each pipeline stage contributes positively to segmentation performance. YOLOv8 detection achieves a mean Average Precision~(mAP@0.5) of 0.91, precision of 0.88, and recall of 0.86. The results demonstrate that high-quality, automatically generated spatial prompts can substitute for costly parameter adaptation of general-purpose foundation models in specialized medical imaging tasks.
Zitieren
APA
Alabsi, M., AQashlan, A., Mohamed, R., & Taha, A. (2026). Automated Brain Tumor Segmentation via YOLOv8-Derived Spatial Prompts for the Segment Anything Model. Computational Discovery and Intelligent Systems (CDIS). https://doi.org/10.66279/vv8n5p41
Vancouver
Alabsi M, AQashlan A, Mohamed R, Taha A. Automated Brain Tumor Segmentation via YOLOv8-Derived Spatial Prompts for the Segment Anything Model. Computational Discovery and Intelligent Systems (CDIS). 2026. doi:10.66279/vv8n5p41
BibTeX
@article{Alabsi2026,
author = {Alabsi, Maha and AQashlan, Amjad and Mohamed, Rehab and Taha, Ayat},
title = {Automated Brain Tumor Segmentation via YOLOv8-Derived Spatial Prompts for the Segment Anything Model},
journal = {Computational Discovery and Intelligent Systems (CDIS)},
year = {2026},
doi = {10.66279/vv8n5p41}
}
EUVIMED