Related Experiment Video
Updated: Aug 3, 2025
08:41
Patient-Specific Polyvinyl Alcohol Phantom Fabrication with Ultrasound and X-Ray Contrast for Brain Tumor Surgery Planning
Published on: July 14, 2020
8.6K
A Foreground Prototype-Based One-Shot Segmentation of Brain Tumors.
Ananthakrishnan Balasundaram1, Muthu Subash Kavitha2, Yogarajah Pratheepan3
1School of Computer Science and Engineering, Vellore Institute of Technology, Chennai 600127, Tamil Nadu, India.
Diagnostics (Basel, Switzerland)
|April 13, 2023
Summary
This study introduces a novel one-shot learning model for brain tumor segmentation in MRI scans. The model achieves high accuracy with minimal data, outperforming traditional methods in efficiency and performance.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Biology
Background:
- Deep learning networks (DNNs) show promise for brain tumor segmentation but require extensive training data.
- A significant challenge for DNNs is their performance on unseen classes, limiting their clinical applicability.
- Few-shot learning offers a potential solution to data scarcity in medical image analysis.
Purpose of the Study:
- To develop and evaluate a one-shot learning model for accurate brain tumor segmentation using minimal data.
- To address the limitations of conventional DNNs in handling limited and unseen data for brain tumor segmentation.
- To improve the efficiency and effectiveness of brain tumor segmentation in MRI by leveraging few-shot learning techniques.
Main Methods:
- Proposed a one-shot learning model utilizing a single prototype similarity score for brain tumor segmentation.
- Employed few-shot learning techniques with support and query image sets, focusing on foreground slices.
- Utilized a metric learning-based approach with non-parametric thresholds for differentiating query images from class prototypes.
- Trained the model iteratively using random foreground slices from the multimodal Brain Tumor Image Segmentation (BraTS) 2021 dataset.
Main Results:
- Achieved a mean dice score of 83.42%, outperforming existing literature benchmarks.
- Demonstrated superior performance compared to conventional methods in terms of computational time and memory usage.
- Successfully segmented brain tumors using a single prototype similarity score with minimal training data.
Conclusions:
- The proposed one-shot learning model offers a highly effective and efficient solution for brain tumor segmentation in MRI.
- This approach significantly reduces the need for large annotated datasets, making it valuable for clinical applications.
- The model's ability to perform well with limited data and its computational efficiency represent a significant advancement in medical image analysis.
