Hou Lijie, Zhang Jian, Zhao Lihui, Meng Ke, Feng Xin
School of Life Science and Technology, Changchun University of Science and Technology, ChangChun City, 130000, China.
The Third Bethune Hospital of JiLin University, Neurosurgery, ChangChun City, 130000, China.
Sci Rep. 2025 Mar 27;15(1):10593. doi: 10.1038/s41598-025-95143-2.
Accurately segmenting the aneurysm area from CTA data can reconstruct the three-dimensional morphology of the aneurysm, effectively evaluating the type, size, and risk of rupture of the aneurysm. However, accurate separation of the aneurysm is limited by the accuracy of image segmentation algorithms. Currently, the segmentation methods for intracranial aneurysms using CTA big data and deep learning lack universality. When faced with a new hospital acquired imaging modality, it is usually necessary to redesign and train the segmentation network. In response to this issue, this article proposes a more universal segmentation model and develops the GLIA Net algorithm (MGLIA Net model) based on MoblieNet, which can perform adaptive target segmentation on aneurysm images collected under different conditions. To verify the effectiveness of the algorithm in intracranial aneurysm segmentation, performance tests were conducted on an open-source dataset. The results showed that the proposed algorithm achieved segmentation accuracy of 55.9% and 73.1% on two datasets, respectively, significantly better than the original GLIA-Net algorithm.
从CTA数据中准确分割出动脉瘤区域，可以重建动脉瘤的三维形态，有效评估动脉瘤的类型、大小和破裂风险。然而，动脉瘤的准确分割受到图像分割算法精度的限制。目前，利用CTA大数据和深度学习进行颅内动脉瘤分割的方法缺乏通用性。当面对新的医院获取的成像模态时，通常需要重新设计和训练分割网络。针对这一问题，本文提出了一种更通用的分割模型，并基于MoblieNet开发了GLIA Net算法（MGLIA Net模型），该模型可以对在不同条件下采集的动脉瘤图像进行自适应目标分割。为了验证该算法在颅内动脉瘤分割中的有效性，在一个开源数据集上进行了性能测试。结果表明，所提算法在两个数据集上的分割准确率分别达到了55.9%和73.1%，明显优于原始的GLIA-Net算法。
