Attention-enhanced SAM with PBFO tuning: advancing glioma MRI segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Brain tumor segmentation from MRI is complex due to tissue variability.
• Accurate segmentation is crucial for diagnosis and treatment planning.
• Existing methods face challenges with intricate tumor structures.

Purpose of the Study:

• To introduce PoSAM-ULTRA, an enhanced framework for brain tumor MRI segmentation.
• To improve the accuracy and robustness of automated segmentation.
• To address the challenges posed by complex tumor tissues.

Main Methods:

• PoSAM-ULTRA utilizes an improved Segment Anything Model backbone and a ResNet-34 encoder.
• Hyperparameter tuning is performed using the Polar-Bear Foraging Optimisation (PBFO) algorithm.
• Multi-scale feature extraction, attention mechanisms (CBAM, Attention Gates), and a multistage decoder are employed.

Main Results:

• PoSAM-ULTRA achieved a Dice score of 91.4%, IoU of 88.9%, Accuracy of 99.8%, Precision of 95.2%, and Recall of 93.3%.
• The framework demonstrated superior performance compared to UNet, UNet++, and nnUNet.
• The results highlight the model's effectiveness on the Lower Grade Gliomas (LGG) dataset.

Conclusions:

• PoSAM-ULTRA shows significant robustness and reliability in segmenting brain tumors.
• The framework is effective for challenging medical image segmentation tasks.
• This advancement offers a promising tool for clinical applications in neuro-oncology.