Fluids and barriers of the CNS · 2026
CT-based automatic segmentation of key CSF regions for detecting disproportionately enlarged subarachnoid space hydrocephalus.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Yamada Shigeki, Ito Hirotaka, Hagiwara Keisuke, Kawata Yasuo, Iseki Chifumi, Tanikawa Motoki, Otani Tomohiro, Ii Satoshi, Watanabe Yoshiyuki, Wada Shigeo, Oshima Marie, Mase Mitsuhito
Аннотация
BACKGROUND: Disproportionately enlarged subarachnoid space hydrocephalus (DESH) is a characteristic neuroimaging feature of idiopathic normal pressure hydrocephalus (iNPH), a treatable but frequently underdiagnosed condition. While brain MRI enables detailed region-based segmentation, CT is faster and widely used in daily clinical practice. However, objective and automated evaluation of DESH-related features on plain CT remains challenging. This study aimed to develop and validate an artificial intelligence-based segmentation model for automatically extracting four cerebrospinal fluid (CSF) regions from plain CT.
METHODS: Synthetic CT images were generated from annotated 3D T1-weighted MRI using a cycle-generative adversarial network (Cycle-GAN) to create training data. However, segmentation accuracy based on synthetic CT was insufficient. therefore, the final model was trained using manually annotated real CT images. A deep learning model was developed to segment four CSF regions (total ventricles, total subarachnoid space [SAS], high-convexity SAS, and Sylvian fissures with basal cisterns). Model performance was evaluated using an external validation dataset (30 DESH and 30 non-DESH). In addition, a second independent validation set consisting of 115 consecutive patients was used to assess clinical utility and determine optimal cutoff values using receiver operating characteristic analysis.
RESULTS: Based on external validation Dice scores, Version 10 was adopted as the final model. In internal validation, the final model achieved Dice scores > 0.9 for total ventricles and > 0.8 for total intracranial CSF space, with lower scores for high-convexity SAS and Sylvian fissures with basal cisterns. External validation yielded Dice scores of 0.92, 0.85, 0.60, and 0.94 for the four regions, respectively. In the second validation set, the DESH index demonstrated excellent diagnostic performance, with an area under the curve of 1.00 and an optimal cutoff value of 10 (sensitivity 100%, specificity 100%). Venthi and Sylhi indices also showed high diagnostic performance.
CONCLUSIONS: The proposed artificial intelligence model enables fully automated segmentation and quantitative assessment of DESH-related CSF regions on plain CT with high reliability. This approach may improve diagnostic accuracy, facilitate earlier detection, and support objective evaluation of postoperative changes in iNPH. Its applicability to routine CT makes it particularly valuable in settings where MRI is unavailable or contraindicated.
METHODS: Synthetic CT images were generated from annotated 3D T1-weighted MRI using a cycle-generative adversarial network (Cycle-GAN) to create training data. However, segmentation accuracy based on synthetic CT was insufficient. therefore, the final model was trained using manually annotated real CT images. A deep learning model was developed to segment four CSF regions (total ventricles, total subarachnoid space [SAS], high-convexity SAS, and Sylvian fissures with basal cisterns). Model performance was evaluated using an external validation dataset (30 DESH and 30 non-DESH). In addition, a second independent validation set consisting of 115 consecutive patients was used to assess clinical utility and determine optimal cutoff values using receiver operating characteristic analysis.
RESULTS: Based on external validation Dice scores, Version 10 was adopted as the final model. In internal validation, the final model achieved Dice scores > 0.9 for total ventricles and > 0.8 for total intracranial CSF space, with lower scores for high-convexity SAS and Sylvian fissures with basal cisterns. External validation yielded Dice scores of 0.92, 0.85, 0.60, and 0.94 for the four regions, respectively. In the second validation set, the DESH index demonstrated excellent diagnostic performance, with an area under the curve of 1.00 and an optimal cutoff value of 10 (sensitivity 100%, specificity 100%). Venthi and Sylhi indices also showed high diagnostic performance.
CONCLUSIONS: The proposed artificial intelligence model enables fully automated segmentation and quantitative assessment of DESH-related CSF regions on plain CT with high reliability. This approach may improve diagnostic accuracy, facilitate earlier detection, and support objective evaluation of postoperative changes in iNPH. Its applicability to routine CT makes it particularly valuable in settings where MRI is unavailable or contraindicated.