Enhancing semi-supervised learning for fine-grained 3D cerebrovascular segmentation with cross-consistency and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Accurate segmentation of cerebral blood vessels from time-of-flight magnetic resonance angiography (TOF-MRA) is crucial for diagnosing and treating cerebrovascular diseases.
• Supervised deep learning methods face limitations due to high annotation costs and limited applicability.
• There is a need for alternative approaches to improve automatic 3D cerebrovascular segmentation efficiency and clinical deployment.

Purpose of the Study:

• To develop a semi-supervised learning method that leverages unlabeled TOF-MRA data for improved 3D cerebrovascular segmentation.
• To address the challenge of limited labeled data by exploiting vessel structures and assessing pseudo-label reliability.
• To enhance the utilization of unlabeled data and boost segmentation accuracy.

Main Methods:

• Introduced a cross-consistency dual uncertainty quantification mean teacher method for semi-supervised learning.
• Employed a dual-consistency learning approach incorporating pixel-image transformation consistency and feature perturbation invariance.
• Evaluated segmentation uncertainty using student and teacher model predictions for guiding consistency regularization and boosted performance with a region-specific supervised loss.

Main Results:

• The proposed method outperformed state-of-the-art semi-supervised learning techniques on two public datasets.
• Achieved a Dice similarity coefficient of 83.3% and intersection-over-union of 71.5% on the IXI dataset.
• Demonstrated superior performance compared to the baseline uncertainty-aware mean teacher method by 1.7% and 2.8% respectively.

Conclusions:

• The developed framework shows potential in reducing manual annotation efforts for cerebrovascular segmentation.
• Its effectiveness in handling unlabeled data offers significant advantages for accurate cerebrovascular extraction tasks.
• The method achieves competitive performance across various metrics, indicating its clinical applicability.