ORIGINAL RESEARCH article
Front. Med.
Sec. Precision Medicine
Attention-Enhanced SAM with PBFO Tuning: Advancing Glioma MRI Segmentation
Salem Alhatamleh 1
Hamad Yahia Abu Mhanna 2
Mohammad Amin 1
Amal Alishwait 1
Mohammad Latayfeh 1
Qutaiba Mohammad 1
Ghada A. Khouqeer 3
Abdullah Alrefai 4
Sitah Alanazi 3
Kholoud Sandougah 3
1. Yarmouk University, Irbid, Jordan
2. Isra University, Amman, Jordan
3. Imam Muhammad Ibn Saud Islamic University, Riyadh, Saudi Arabia
4. Jordan University of Science and Technology, Irbid, Jordan
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Abstract
The segmentation of brain tumor MRI images is one of the most challenging tasks because of the variability and complexity associated with tumor tissues. The study introduces PoSAM-ULTRA, a improved segmentation framework that is all set to employ the Polar-Bear Foraging Optimization PBFO algorithm for hyperparameter tuning along with an improved Segment Anything Model as its backbone. It is based on a ResNet-34 encoder that has been modified to accept a four-channel input (RGB + prior information) which is then subjected to multi-scale feature extraction via DownBlocks, reinforcing discriminative power through the use of Channel and Spatial Attention (CBAM) and ensuring quality skip connections through the use of Attention Gates while a multistage decoder takes care of the upsampling and feature integration in a robust manner. The application of deep supervision, dropout regularization, and a composite loss function improves the stability and generalization of training. The model was tested on a dataset that came from the Integrative Genomic Analysis of Diffuse Lower Grade Gliomas (LGG) and was compared with UNet, UNet++, and nnUNet. The results showed that the proposed PoSAM-ULTRA model outperformed the baseline models, achieving high performance in Dice (91.4%), IoU (88.9%), Accuracy (99.8), Precision (95.2%), and Recall (93.3%), which in turn was supported by its robustness and reliability in challenging medical image segmentation tasks.
Summary
Keywords
deep learning, Glioma segmentation, lower grade gliomas, Medical Image Analysis, MRI imaging, Segment Anything Model (SAM)
Received
05 January 2026
Accepted
26 January 2026
Copyright
© 2026 Alhatamleh, Abu Mhanna, Amin, Alishwait, Latayfeh, Mohammad, Khouqeer, Alrefai, Alanazi and Sandougah. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Kholoud Sandougah
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.