Abstract
Thrombus volume in posterior circulation stroke (PCS) has been associated with outcome, through recanalization. Manual thrombus segmentation is impractical for large scale analysis of image characteristics. Hence, in this study we develop the first automatic method for thrombus localization and segmentation on CT in patients with PCS. In this multi-center retrospective study, 187 patients with PCS from the MR CLEAN Registry were included. We developed a convolutional neural network (CNN) that segments thrombi and restricts the volume-of-interest (VOI) to the brainstem (Polar-UNet). Furthermore, we reduced false positive localization by removing small-volume objects, referred to as volume-based removal (VBR). Polar-UNet is benchmarked against a CNN that does not restrict the VOI (BL-UNet). Performance metrics included the intra-class correlation coefficient (ICC) between automated and manually segmented thrombus volumes, the thrombus localization precision and recall, and the Dice coefficient. The majority of the thrombi were localized. Without VBR, Polar-UNet achieved a thrombus localization recall of 0.82, versus 0.78 achieved by BL-UNet. This high recall was accompanied by a low precision of 0.14 and 0.09. VBR improved precision to 0.65 and 0.56 for Polar-UNet and BL-UNet, respectively, with a small reduction in recall to 0.75 and 0.69. The Dice coefficient achieved by Polar-UNet was 0.44, versus 0.38 achieved by BL-UNet with VBR. Both methods achieved ICCs of 0.41 (95% CI: 0.27–0.54). Restricting the VOI to the brainstem improved the thrombus localization precision, recall, and segmentation overlap compared to the benchmark. VBR improved thrombus localization precision but lowered recall.
Original language | English |
---|---|
Article number | 1400 |
Journal | Diagnostics |
Volume | 12 |
Issue number | 6 |
DOIs | |
Publication status | Published - Jun 2022 |
Bibliographical note
Publisher Copyright:
© 2022 by the authors.
Access to Document
10.3390/diagnostics12061400Licence: CC BY
Deep-Learning-Based Thrombus Localization and Segmentation in Patients with Posterior Circulation StrokeFinal published version, 1.59 MBLicence: CC BY
Other files and links
Fingerprint
Dive into the research topics of 'Deep-Learning-Based Thrombus Localization and Segmentation in Patients with Posterior Circulation Stroke'. Together they form a unique fingerprint.
View full fingerprint
Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver
Zoetmulder, R., Bruggeman, A. A. E., Išgum, I., Gavves, E., Majoie, C. B. L. M., Beenen, L. F. M., Dippel, D. W. J., Boodt, N., Den Hartog, S. J., Van Doormaal, P. J., Cornelissen, S. A. P., Roos, Y. B. W. E. M., Brouwer, J., Schonewille, W. J., Pirson, A. F. V., Van Zwam, W. H., Leij, C. V. D., Brans, R. J. B., Adriaan, A. C. G., & Marquering, H. A. (2022). Deep-Learning-Based Thrombus Localization and Segmentation in Patients with Posterior Circulation Stroke. Diagnostics, 12(6), Article 1400. https://doi.org/10.3390/diagnostics12061400
Zoetmulder, Riaan ; Bruggeman, Agnetha A.E. ; Išgum, Ivana et al. / Deep-Learning-Based Thrombus Localization and Segmentation in Patients with Posterior Circulation Stroke. In: Diagnostics. 2022 ; Vol. 12, No. 6.
@article{ef859b91b25c45d184452f771ecfa143,
title = "Deep-Learning-Based Thrombus Localization and Segmentation in Patients with Posterior Circulation Stroke",
abstract = "Thrombus volume in posterior circulation stroke (PCS) has been associated with outcome, through recanalization. Manual thrombus segmentation is impractical for large scale analysis of image characteristics. Hence, in this study we develop the first automatic method for thrombus localization and segmentation on CT in patients with PCS. In this multi-center retrospective study, 187 patients with PCS from the MR CLEAN Registry were included. We developed a convolutional neural network (CNN) that segments thrombi and restricts the volume-of-interest (VOI) to the brainstem (Polar-UNet). Furthermore, we reduced false positive localization by removing small-volume objects, referred to as volume-based removal (VBR). Polar-UNet is benchmarked against a CNN that does not restrict the VOI (BL-UNet). Performance metrics included the intra-class correlation coefficient (ICC) between automated and manually segmented thrombus volumes, the thrombus localization precision and recall, and the Dice coefficient. The majority of the thrombi were localized. Without VBR, Polar-UNet achieved a thrombus localization recall of 0.82, versus 0.78 achieved by BL-UNet. This high recall was accompanied by a low precision of 0.14 and 0.09. VBR improved precision to 0.65 and 0.56 for Polar-UNet and BL-UNet, respectively, with a small reduction in recall to 0.75 and 0.69. The Dice coefficient achieved by Polar-UNet was 0.44, versus 0.38 achieved by BL-UNet with VBR. Both methods achieved ICCs of 0.41 (95% CI: 0.27–0.54). Restricting the VOI to the brainstem improved the thrombus localization precision, recall, and segmentation overlap compared to the benchmark. VBR improved thrombus localization precision but lowered recall.",
author = "Riaan Zoetmulder and Bruggeman, {Agnetha A.E.} and Ivana I{\v s}gum and Efstratios Gavves and Majoie, {Charles B.L.M.} and Beenen, {Ludo F.M.} and Dippel, {Diederik W.J.} and Nikkie Boodt and {Den Hartog}, {Sanne J.} and {Van Doormaal}, {Pieter J.} and Cornelissen, {Sandra A.P.} and Roos, {Yvo B.W.E.M.} and Josje Brouwer and Schonewille, {Wouter J.} and Pirson, {Anne F.V.} and {Van Zwam}, {Wim H.} and Leij, {Christiaan Van Der} and Brans, {Rutger J.B.} and Adriaan, {Adriaan C.G.} and Marquering, {Henk A.}",
note = "Publisher Copyright: {\textcopyright} 2022 by the authors.",
year = "2022",
month = jun,
doi = "10.3390/diagnostics12061400",
language = "English",
volume = "12",
number = "6",
}
Zoetmulder, R, Bruggeman, AAE, Išgum, I, Gavves, E, Majoie, CBLM, Beenen, LFM, Dippel, DWJ, Boodt, N, Den Hartog, SJ, Van Doormaal, PJ, Cornelissen, SAP, Roos, YBWEM, Brouwer, J, Schonewille, WJ, Pirson, AFV, Van Zwam, WH, Leij, CVD, Brans, RJB, Adriaan, ACG & Marquering, HA 2022, 'Deep-Learning-Based Thrombus Localization and Segmentation in Patients with Posterior Circulation Stroke', Diagnostics, vol. 12, no. 6, 1400. https://doi.org/10.3390/diagnostics12061400
Deep-Learning-Based Thrombus Localization and Segmentation in Patients with Posterior Circulation Stroke. / Zoetmulder, Riaan; Bruggeman, Agnetha A.E.; Išgum, Ivana et al.
In: Diagnostics, Vol. 12, No. 6, 1400, 06.2022.
Research output: Contribution to journal › Article › Academic › peer-review
TY - JOUR
T1 - Deep-Learning-Based Thrombus Localization and Segmentation in Patients with Posterior Circulation Stroke
AU - Zoetmulder, Riaan
AU - Bruggeman, Agnetha A.E.
AU - Išgum, Ivana
AU - Gavves, Efstratios
AU - Majoie, Charles B.L.M.
AU - Beenen, Ludo F.M.
AU - Dippel, Diederik W.J.
AU - Boodt, Nikkie
AU - Den Hartog, Sanne J.
AU - Van Doormaal, Pieter J.
AU - Cornelissen, Sandra A.P.
AU - Roos, Yvo B.W.E.M.
AU - Brouwer, Josje
AU - Schonewille, Wouter J.
AU - Pirson, Anne F.V.
AU - Van Zwam, Wim H.
AU - Leij, Christiaan Van Der
AU - Brans, Rutger J.B.
AU - Adriaan, Adriaan C.G.
AU - Marquering, Henk A.
N1 - Publisher Copyright:© 2022 by the authors.
PY - 2022/6
Y1 - 2022/6
N2 - Thrombus volume in posterior circulation stroke (PCS) has been associated with outcome, through recanalization. Manual thrombus segmentation is impractical for large scale analysis of image characteristics. Hence, in this study we develop the first automatic method for thrombus localization and segmentation on CT in patients with PCS. In this multi-center retrospective study, 187 patients with PCS from the MR CLEAN Registry were included. We developed a convolutional neural network (CNN) that segments thrombi and restricts the volume-of-interest (VOI) to the brainstem (Polar-UNet). Furthermore, we reduced false positive localization by removing small-volume objects, referred to as volume-based removal (VBR). Polar-UNet is benchmarked against a CNN that does not restrict the VOI (BL-UNet). Performance metrics included the intra-class correlation coefficient (ICC) between automated and manually segmented thrombus volumes, the thrombus localization precision and recall, and the Dice coefficient. The majority of the thrombi were localized. Without VBR, Polar-UNet achieved a thrombus localization recall of 0.82, versus 0.78 achieved by BL-UNet. This high recall was accompanied by a low precision of 0.14 and 0.09. VBR improved precision to 0.65 and 0.56 for Polar-UNet and BL-UNet, respectively, with a small reduction in recall to 0.75 and 0.69. The Dice coefficient achieved by Polar-UNet was 0.44, versus 0.38 achieved by BL-UNet with VBR. Both methods achieved ICCs of 0.41 (95% CI: 0.27–0.54). Restricting the VOI to the brainstem improved the thrombus localization precision, recall, and segmentation overlap compared to the benchmark. VBR improved thrombus localization precision but lowered recall.
AB - Thrombus volume in posterior circulation stroke (PCS) has been associated with outcome, through recanalization. Manual thrombus segmentation is impractical for large scale analysis of image characteristics. Hence, in this study we develop the first automatic method for thrombus localization and segmentation on CT in patients with PCS. In this multi-center retrospective study, 187 patients with PCS from the MR CLEAN Registry were included. We developed a convolutional neural network (CNN) that segments thrombi and restricts the volume-of-interest (VOI) to the brainstem (Polar-UNet). Furthermore, we reduced false positive localization by removing small-volume objects, referred to as volume-based removal (VBR). Polar-UNet is benchmarked against a CNN that does not restrict the VOI (BL-UNet). Performance metrics included the intra-class correlation coefficient (ICC) between automated and manually segmented thrombus volumes, the thrombus localization precision and recall, and the Dice coefficient. The majority of the thrombi were localized. Without VBR, Polar-UNet achieved a thrombus localization recall of 0.82, versus 0.78 achieved by BL-UNet. This high recall was accompanied by a low precision of 0.14 and 0.09. VBR improved precision to 0.65 and 0.56 for Polar-UNet and BL-UNet, respectively, with a small reduction in recall to 0.75 and 0.69. The Dice coefficient achieved by Polar-UNet was 0.44, versus 0.38 achieved by BL-UNet with VBR. Both methods achieved ICCs of 0.41 (95% CI: 0.27–0.54). Restricting the VOI to the brainstem improved the thrombus localization precision, recall, and segmentation overlap compared to the benchmark. VBR improved thrombus localization precision but lowered recall.
UR - http://www.scopus.com/inward/record.url?scp=85135016224&partnerID=8YFLogxK
U2 - 10.3390/diagnostics12061400
DO - 10.3390/diagnostics12061400
M3 - Article
AN - SCOPUS:85135016224
SN - 2075-4418
VL - 12
JO - Diagnostics
JF - Diagnostics
IS - 6
M1 - 1400
ER -
Zoetmulder R, Bruggeman AAE, Išgum I, Gavves E, Majoie CBLM, Beenen LFM et al. Deep-Learning-Based Thrombus Localization and Segmentation in Patients with Posterior Circulation Stroke. Diagnostics. 2022 Jun;12(6):1400. doi: 10.3390/diagnostics12061400