Deep learning-based segmentation of brain parenchyma and ventricular system in CT scans in the presence of anomalies
IntroductionThe automatic segmentation of brain parenchyma and cerebrospinal fluid-filled spaces such as the ventricular system is the first step for quantitative and qualitative analysis of brain CT data. For clinical practice and especially for diagnostics, it is crucial that such a method is robu...
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2023-08-01
|
| Series: | Frontiers in Neuroimaging |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fnimg.2023.1228255/full |
| _version_ | 1797754734159855616 |
|---|---|
| author | Annika Gerken Sina Walluscheck Peter Kohlmann Ivana Galinovic Kersten Villringer Jochen B. Fiebach Jan Klein Stefan Heldmann |
| author_facet | Annika Gerken Sina Walluscheck Peter Kohlmann Ivana Galinovic Kersten Villringer Jochen B. Fiebach Jan Klein Stefan Heldmann |
| author_sort | Annika Gerken |
| collection | DOAJ |
| description | IntroductionThe automatic segmentation of brain parenchyma and cerebrospinal fluid-filled spaces such as the ventricular system is the first step for quantitative and qualitative analysis of brain CT data. For clinical practice and especially for diagnostics, it is crucial that such a method is robust to anatomical variability and pathological changes such as (hemorrhagic or neoplastic) lesions and chronic defects. This study investigates the increase in overall robustness of a deep learning algorithm that is gained by adding hemorrhage training data to an otherwise normal training cohort.MethodsA 2D U-Net is trained on subjects with normal appearing brain anatomy. In a second experiment the training data includes additional subjects with brain hemorrhage on image data of the RSNA Brain CT Hemorrhage Challenge with custom reference segmentations. The resulting networks are evaluated on normal and hemorrhage test casesseparately, and on an independent test set of patients with brain tumors of the publicly available GLIS-RT dataset.ResultsAdding data with hemorrhage to the training set significantly improves the segmentation performance over an algorithm trained exclusively on normally appearing data, not only in the hemorrhage test set but also in the tumor test set. The performance on normally appearing data is stable. Overall, the improved algorithm achieves median Dice scores of 0.98 (parenchyma), 0.91 (left ventricle), 0.90 (right ventricle), 0.81 (third ventricle), and 0.80 (fourth ventricle) on the hemorrhage test set. On the tumor test set, the median Dice scores are 0.96 (parenchyma), 0.90 (left ventricle), 0.90 (right ventricle), 0.75 (third ventricle), and 0.73 (fourth ventricle).ConclusionTraining on an extended data set that includes pathologies is crucial and significantly increases the overall robustness of a segmentation algorithm for brain parenchyma and ventricular system in CT data, also for anomalies completely unseen during training. Extension of the training set to include other diseases may further improve the generalizability of the algorithm. |
| first_indexed | 2024-03-12T17:37:50Z |
| format | Article |
| id | doaj.art-92c8b8b5f1b8427bb499c1041eb3081a |
| institution | Directory Open Access Journal |
| issn | 2813-1193 |
| language | English |
| last_indexed | 2024-03-12T17:37:50Z |
| publishDate | 2023-08-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Neuroimaging |
| spelling | doaj.art-92c8b8b5f1b8427bb499c1041eb3081a2023-08-04T09:32:37ZengFrontiers Media S.A.Frontiers in Neuroimaging2813-11932023-08-01210.3389/fnimg.2023.12282551228255Deep learning-based segmentation of brain parenchyma and ventricular system in CT scans in the presence of anomaliesAnnika Gerken0Sina Walluscheck1Peter Kohlmann2Ivana Galinovic3Kersten Villringer4Jochen B. Fiebach5Jan Klein6Stefan Heldmann7Fraunhofer Institute for Digital Medicine MEVIS, Bremen, GermanyFraunhofer Institute for Digital Medicine MEVIS, Lübeck, GermanyFraunhofer Institute for Digital Medicine MEVIS, Berlin, GermanyCenter for Stroke Research Berlin (CSB) Charité, Universitätsmedizin, Berlin, Berlin, GermanyCenter for Stroke Research Berlin (CSB) Charité, Universitätsmedizin, Berlin, Berlin, GermanyCenter for Stroke Research Berlin (CSB) Charité, Universitätsmedizin, Berlin, Berlin, GermanyFraunhofer Institute for Digital Medicine MEVIS, Bremen, GermanyFraunhofer Institute for Digital Medicine MEVIS, Lübeck, GermanyIntroductionThe automatic segmentation of brain parenchyma and cerebrospinal fluid-filled spaces such as the ventricular system is the first step for quantitative and qualitative analysis of brain CT data. For clinical practice and especially for diagnostics, it is crucial that such a method is robust to anatomical variability and pathological changes such as (hemorrhagic or neoplastic) lesions and chronic defects. This study investigates the increase in overall robustness of a deep learning algorithm that is gained by adding hemorrhage training data to an otherwise normal training cohort.MethodsA 2D U-Net is trained on subjects with normal appearing brain anatomy. In a second experiment the training data includes additional subjects with brain hemorrhage on image data of the RSNA Brain CT Hemorrhage Challenge with custom reference segmentations. The resulting networks are evaluated on normal and hemorrhage test casesseparately, and on an independent test set of patients with brain tumors of the publicly available GLIS-RT dataset.ResultsAdding data with hemorrhage to the training set significantly improves the segmentation performance over an algorithm trained exclusively on normally appearing data, not only in the hemorrhage test set but also in the tumor test set. The performance on normally appearing data is stable. Overall, the improved algorithm achieves median Dice scores of 0.98 (parenchyma), 0.91 (left ventricle), 0.90 (right ventricle), 0.81 (third ventricle), and 0.80 (fourth ventricle) on the hemorrhage test set. On the tumor test set, the median Dice scores are 0.96 (parenchyma), 0.90 (left ventricle), 0.90 (right ventricle), 0.75 (third ventricle), and 0.73 (fourth ventricle).ConclusionTraining on an extended data set that includes pathologies is crucial and significantly increases the overall robustness of a segmentation algorithm for brain parenchyma and ventricular system in CT data, also for anomalies completely unseen during training. Extension of the training set to include other diseases may further improve the generalizability of the algorithm.https://www.frontiersin.org/articles/10.3389/fnimg.2023.1228255/fulldeep learningsegmentationhemorrhageparenchymaventricular system |
| spellingShingle | Annika Gerken Sina Walluscheck Peter Kohlmann Ivana Galinovic Kersten Villringer Jochen B. Fiebach Jan Klein Stefan Heldmann Deep learning-based segmentation of brain parenchyma and ventricular system in CT scans in the presence of anomalies Frontiers in Neuroimaging deep learning segmentation hemorrhage parenchyma ventricular system |
| title | Deep learning-based segmentation of brain parenchyma and ventricular system in CT scans in the presence of anomalies |
| title_full | Deep learning-based segmentation of brain parenchyma and ventricular system in CT scans in the presence of anomalies |
| title_fullStr | Deep learning-based segmentation of brain parenchyma and ventricular system in CT scans in the presence of anomalies |
| title_full_unstemmed | Deep learning-based segmentation of brain parenchyma and ventricular system in CT scans in the presence of anomalies |
| title_short | Deep learning-based segmentation of brain parenchyma and ventricular system in CT scans in the presence of anomalies |
| title_sort | deep learning based segmentation of brain parenchyma and ventricular system in ct scans in the presence of anomalies |
| topic | deep learning segmentation hemorrhage parenchyma ventricular system |
| url | https://www.frontiersin.org/articles/10.3389/fnimg.2023.1228255/full |
| work_keys_str_mv | AT annikagerken deeplearningbasedsegmentationofbrainparenchymaandventricularsysteminctscansinthepresenceofanomalies AT sinawalluscheck deeplearningbasedsegmentationofbrainparenchymaandventricularsysteminctscansinthepresenceofanomalies AT peterkohlmann deeplearningbasedsegmentationofbrainparenchymaandventricularsysteminctscansinthepresenceofanomalies AT ivanagalinovic deeplearningbasedsegmentationofbrainparenchymaandventricularsysteminctscansinthepresenceofanomalies AT kerstenvillringer deeplearningbasedsegmentationofbrainparenchymaandventricularsysteminctscansinthepresenceofanomalies AT jochenbfiebach deeplearningbasedsegmentationofbrainparenchymaandventricularsysteminctscansinthepresenceofanomalies AT janklein deeplearningbasedsegmentationofbrainparenchymaandventricularsysteminctscansinthepresenceofanomalies AT stefanheldmann deeplearningbasedsegmentationofbrainparenchymaandventricularsysteminctscansinthepresenceofanomalies |