Analysing Variables for 90-Day Functional-Outcome Prediction of Endovascular Thrombectomy

Rodriguez Salas D, Rieß C, Martín Vicario C, Taubmann O, Ditt H, Schwab S, Dörfler A (2024)

Publication Language: English

Publication Type: Conference contribution, Original article

Publication year: 2024

Publisher: Springer

Series: Lecture Notes in Computer Science

City/Town: Cham

Book Volume: 14860

Pages Range: 202–215

Conference Proceedings Title: Medical Image Understanding and Analysis. 28th Annual Conference, MIUA 2024, Manchester, UK, July 24–26, 2024, Proceedings, Part II

Event location: Manchester

ISBN: 978-3-031-66958-3

DOI: 10.1007/978-3-031-66958-3_15

Abstract

Multiple studies report the impact of variables on predicting long-term functional outcomes following Endovascular Thrombectomy (EVT). Often, the variables are analyzed separately with univariate statistical analysis or combined in a linear regression or logistic regression prediction. Although such linear combinations are easy to interpret, they do not consider nonlinear relationships. Other studies do consider nonlinearities, most notably in deep learning. However, if any, they use separate methods to assess the variable importance, which might not capture their actual contribution to the prediction. In this work, we propose to use an algorithm that combines the predictive power of neural networks and the interpretation of decision trees in a single method. ForestNet is a special artificial neural network that defines its partially connected architecture based on the hierarchical variable importance given by an ensemble of decision trees, providing a means of interpretation in a more generalized fashion. On predicting 90-day favorable functional outcomes, using 10-fold cross-validation, we reach an AUC of 0.82 and 0.86 before and after EVT, respectively. A 90-day Modified Rankin Scale (mRS)<=2 defines a favorable outcome. The analysis of the variable’s contribution shows that the pre-stroke mRS, age, and the National Institutes of Health Stroke Scale (NIHSS) play the most important role in predictions before EVT. After EVT, the infarct volume from the 24h follow-up CT scan, measured directly or in terms of affected brain regions according to the ASPECT score, is the most important variable in providing a more accurate prediction than before EVT.

Authors with CRIS profile

Involved external institutions

Siemens Healthineers Germany (DE)

How to cite

APA:

Rodriguez Salas, D., Rieß, C., Martín Vicario, C., Taubmann, O., Ditt, H., Schwab, S., & Dörfler, A. (2024). Analysing Variables for 90-Day Functional-Outcome Prediction of Endovascular Thrombectomy. In Moi Hoon Yap, Connah Kendrick, Ardhendu Behera, Timothy Cootes, Reyer Zwiggelaar (Eds.), Medical Image Understanding and Analysis. 28th Annual Conference, MIUA 2024, Manchester, UK, July 24–26, 2024, Proceedings, Part II (pp. 202–215). Manchester, GB: Cham: Springer.

MLA:

Rodriguez Salas, Dalia, et al. "Analysing Variables for 90-Day Functional-Outcome Prediction of Endovascular Thrombectomy." Proceedings of the 28th Annual Conference on Medical Image Understanding and Analysis, Manchester Ed. Moi Hoon Yap, Connah Kendrick, Ardhendu Behera, Timothy Cootes, Reyer Zwiggelaar, Cham: Springer, 2024. 202&#8211;215.

BibTeX: Download