A Segment Anything Model for Automated Intracranial Aneurysm Segmentation in Computed Tomography Angiography Images
Issued Date
2024-01-01
Resource Type
Scopus ID
2-s2.0-85215105679
Journal Title
2024 17th International Convention on Rehabilitation Engineering and Assistive Technology, i-CREATe 2024 and World Rehabilitation Robot Convention, WRRC 2024 - Proceedings
Rights Holder(s)
SCOPUS
Bibliographic Citation
2024 17th International Convention on Rehabilitation Engineering and Assistive Technology, i-CREATe 2024 and World Rehabilitation Robot Convention, WRRC 2024 - Proceedings (2024)
Suggested Citation
Nakhawatchana S., Apirakkan M., Ritthipravat P. A Segment Anything Model for Automated Intracranial Aneurysm Segmentation in Computed Tomography Angiography Images. 2024 17th International Convention on Rehabilitation Engineering and Assistive Technology, i-CREATe 2024 and World Rehabilitation Robot Convention, WRRC 2024 - Proceedings (2024). doi:10.1109/i-CREATe62067.2024.10776105 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/102993
Title
A Segment Anything Model for Automated Intracranial Aneurysm Segmentation in Computed Tomography Angiography Images
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Early Intracranial Aneurysm (IA) detection is crucial for preventing life-threatening ruptures and enhancing patient outcomes. This retrospective study introduces a novel automated segmentation deep learning model to detect and segment IAs from computed tomographic angiography (CTA). Utilizing the Large IA Segmentation Dataset by Bo et al. (2021), which includes 1,338 cases from six medical institutions for training and 150 cases from two external institutions for validation, we developed a Segment Anything Model for 3D medical images (SAM-Med3D) for IA segmentation. This model integrates multi-dimensional spatial data to enhance anatomical accuracy in volumetric data analysis. Our model demonstrated superior performance compared to existing state-of-the-art models. It achieved high precision (External A: 0.89, External B: 0.77), high recall (External A: 0.89, External B: 0.87), and a high Dice's Similarity Coefficient (DSC) (External A: 0.87, External B: 0.82). The automated segmentation model reliably detects and segments IAs in CTA scans across a diverse population. These findings highlight the capability of our model to detect IAs early in clinical settings and promise a viable path for further research and technological advancement.