Development and external validation of a stability machine learning model to identify wake-up stroke onset time from MRI
- 17 January 2022
- journal article
- research article
- Published by Springer Nature in European Radiology
- Vol. 32 (6) , 3661-3669
Abstract
Objectives To develop and externally validate a machine learning (ML) model based on diffusion-weighted imaging (DWI) and fluid-attenuated inversion recovery (FLAIR) to identify the onset time of wake-up stroke from MRI. Methods DWI and FLAIR images of stroke patients within 24 h of clear symptom onset in our hospital (dataset 1, n = 410) and another hospital (dataset 2, n = 177) were included. Seven ML models based on dataset 1 were developed to estimate the stroke onset time for binary classification (≤ 4.5 h or > 4.5 h): Random Forest (RF), support vector machine with kernel (svmLinear) or radial basis function kernel (svmRadial), Bayesian (Bayes), K-nearest neighbor (KNN), adaptive boosting (AdaBoost), and neural network (NNET). ROC analysis and RSD were performed to evaluate the performance and stability of the ML models, respectively, and dataset 2 was externally validated to evaluate the model generalization ability using ROC analysis. Results svmRadial achieved the best performance with the highest AUC and accuracy (AUC: 0.896, accuracy: 0.878), and was the most stable (RSD% of AUC: 0.08, RSD% of accuracy: 0.06). The svmRadial model was then selected as the final model, and the AUC of the svmRadial model for predicting the onset time external validation was 0.895, with 0.825 accuracy. Conclusions The svmRadial model using DWI + FLAIR is the most stable and generalizable for identifying the onset time of wake-up stroke patients within 4.5 h of symptom onset. Key Points • Machining learning model helps clinicians to identify wake-up stroke patients within 4.5 h of symptom onset. • A prospective study showed that svmRadial model based on DWI + FLAIR was the most stable in predicting the stroke onset time. • External validation showed that svmRadial model has good generalization ability in predicting the stroke onset time.Keywords
Funding Information
- The Natural Science Foundation of Jiangsu Province (BK20201118)
- Jiangsu Provincial Special Program of Medical Science (BE2017614)
This publication has 32 references indexed in Scilit:
- Three Dimensional Analysis of SPECT Images for Diagnosing Early Parkinson’s Disease using Radial Basis Function Kernel − Extreme Learning MachineCurrent Medical Imaging, 2019
- eDoctor: machine learning and the future of medicineJournal of Internal Medicine, 2018
- Applications of Support Vector Machine (SVM) Learning in Cancer GenomicsCancer Genomics & Proteomics, 2018
- Stroke With Unknown Time of Symptom OnsetStroke, 2017
- Unclear-onset stroke: Daytime-unwitnessed stroke vs. wake-up strokeInternational Journal of Stroke, 2016
- Wake-Up Stroke: Clinical Characteristics, Imaging Findings, and Treatment Option – an UpdateFrontiers in Neurology, 2014
- Can Diffusion-Weighted Imaging–Fluid-Attenuated Inversion Recovery Mismatch (Positive Diffusion-Weighted Imaging/Negative Fluid-Attenuated Inversion Recovery) at 3 Tesla Identify Patients With Stroke at <4.5 Hours?Stroke, 2013
- DWI-FLAIR mismatch for the identification of patients with acute ischaemic stroke within 4·5 h of symptom onset (PRE-FLAIR): a multicentre observational studyThe Lancet Neurology, 2011
- Fluid-Attenuated Inversion Recovery Evolution Within 12 Hours From Stroke OnsetStroke, 2010
- [Primary tensile strength of newer and modified tendon sutures. A comparative in-vitro study].1990