Efficient Cerebral Infarction Segmentation Using U-Net and U-Net3 + Models.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Cerebral infarction is a major cause of death and disability worldwide.
• Early diagnosis and intervention are crucial for improving patient outcomes.
• Accurate segmentation of infarction regions is vital for treatment planning.

Purpose of the Study:

• To introduce a novel deep learning approach for cerebral infarction segmentation.
• To compare the performance of U-Net and U-Net3+ architectures for this task.
• To evaluate the effectiveness of these models in supporting medical decision-making.

Main Methods:

• A dataset of 110 patient MRI scans was used, augmented to 6732 images.
• Two convolutional neural network architectures, U-Net and U-Net3+, were employed.
• Performance was evaluated using Dice score, IoU, pixel accuracy, and specificity.

Main Results:

• The basic U-Net achieved a Dice score of 0.8947 and IoU of 0.8798.
• U-Net outperformed U-Net3+ in segmentation accuracy.
• Both models demonstrated high pixel accuracy (0.9963) and specificity (0.9984).

Conclusions:

• Deep learning, particularly the U-Net architecture, is effective for precise cerebral infarction segmentation.
• The findings support the use of AI in enhancing stroke diagnosis and treatment planning.
• Model complexity may not always correlate with superior performance in medical imaging segmentation.