3D dissimilar-siamese-u-net for hyperdense Middle cerebral artery sign segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• The hyperdense middle cerebral artery sign (HMCAS) is a critical indicator of thromboembolism in acute ischemic stroke diagnosis.
• Accurate and timely identification of HMCAS is crucial for effective patient triage and treatment selection, including thrombolysis and thrombectomy.

Purpose of the Study:

• To develop and validate a deep learning model for precise automatic segmentation of HMCAS on non-contrast-enhanced CT (NCCT) images.
• To evaluate the performance of the proposed model against existing methods for HMCAS detection.

Main Methods:

• A novel deep Dissimilar-Siamese-U-Net (DSU-Net) architecture was developed, integrating Siamese and U-Net networks.
• The DSU-Net utilizes twin sub-networks to process left and right cerebral hemispheres separately, with a Dissimilar block to analyze inter-hemispheric differences.
• The model was trained and validated on a dataset of 624 retrospectively collected head NCCT scans.

Main Results:

• The DSU-Net demonstrated precise segmentation capabilities for HMCAS.
• Ablation studies confirmed the effectiveness of individual DSU-Net components.
• The proposed DSU-Net outperformed baseline U-Net and other state-of-the-art models in HMCAS segmentation accuracy.

Conclusions:

• The DSU-Net offers a novel and effective automated approach for HMCAS segmentation in clinical practice.
• This AI-driven tool has the potential to improve the efficiency and accuracy of acute ischemic stroke diagnosis.