An Ensemble Learning for Automatic Stroke Lesion Segmentation Using Compressive Sensing and Multi-Resolution U-Net.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Stroke is a leading cause of death, necessitating accurate brain lesion segmentation for diagnosis and treatment.
• Computed Tomography (CT) scans are vital for detecting abnormal brain tissue.
• Existing medical image segmentation methods often overlook patient privacy concerns.

Purpose of the Study:

• To propose a deep learning network that integrates compressive sensing and ensemble learning for efficient and privacy-preserving brain lesion segmentation.
• To enhance the accuracy and reliability of stroke lesion segmentation in CT images.

Main Methods:

• A novel deep network, CS-Ensemble Net, utilizing compressive sensing for data compression and privacy.
• An ensemble of two multi-resolution modified U-shaped networks for segmentation.
• Application to the ISLES 2018 challenge dataset for evaluation.

Main Results:

• Achieved high performance metrics: 92.43% accuracy, 91.3% specificity, and 91.83% dice coefficient.
• Demonstrated superior efficiency compared to state-of-the-art methods.
• Confirmed the effectiveness of compressive sensing for information privacy and ensemble learning for improved results.

Conclusions:

• The proposed CS-Ensemble Net effectively segments brain lesions in CT scans while preserving patient privacy.
• This approach offers a significant advancement in automated stroke lesion segmentation.
• The combination of compressive sensing and ensemble learning presents a promising direction for medical image analysis.