TDGU‐Net: A Hybrid CNN‐Transformer Model for Intracranial Aneurysm Segmentation

ABSTRACT Intracranial aneurysms are life‐threatening cerebrovascular conditions, and their accurate identification is crucial for early diagnosis and treatment planning. Automated segmentation technology plays a key role in enhancing diagnostic accuracy and enabling timely intervention. However, the segmentation task is challenging due to the diverse morphologies of aneurysms, indistinct boundaries, and their resemblance to adjacent vascular structures. This study introduces TDGU‐Net, a deep learning‐based method that combines Convolutional Neural Networks (CNNs) with Transformer architecture to improve segmentation accuracy and efficiency. The model uses CNNs for efficient local feature extraction, while Transformer blocks are employed to establish global relationships within local regions, enhancing the model's ability to capture contextual dependencies. Furthermore, a multi‐scale feature fusion module is incorporated to capture critical information across different resolutions, and the Attention Gate mechanism is used to improve the model's ability to accurately identify aneurysm regions. The proposed model was evaluated on the Large IA Segmentation dataset and further validated on the MICCAI 2020 ADAM dataset to demonstrate its adaptability to different datasets. It achieved a Dice coefficient of 76.92% and a sensitivity of 79.65%, demonstrating robust segmentation performance and accurate detection of aneurysms. The proposed method provides a promising tool for the automated diagnosis of intracranial aneurysms, with significant potential for clinical application and improving patient outcomes.