Deep Learning–Derived High-Level Neuroimaging Features Predict Clinical Outcomes for Large Vessel Occlusion

Abstract

Background and Purpose— For patients with large vessel occlusion, neuroimaging biomarkers that evaluate the changes in brain tissue are important for determining the indications for mechanical thrombectomy. In this study, we applied deep learning to derive imaging features from pretreatment diffusion-weighted image data and evaluated the ability of these features in predicting clinical outcomes for patients with large vessel occlusion. Methods— This multicenter retrospective study included patients with anterior circulation large vessel occlusion treated with mechanical thrombectomy between 2013 and 2018. We designed a 2-output deep learning model based on convolutional neural networks (the convolutional neural network model). This model employed encoder-decoder architecture for the ischemic lesion segmentation, which automatically extracted high-level feature maps in its middle layers, and used its information to predict the clinical outcome. Its performance was internally validated with 5-fold cross-validation, externally validated, and the results compared with those from the standard neuroimaging biomarkers Alberta Stroke Program Early CT Score and ischemic core volume. The prediction target was a good clinical outcome, defined as a modified Rankin Scale score at 90-day follow-up of 0 to 2. Results— The derivation cohort included 250 patients, and the validation cohort included 74 patients. The convolutional neural network model showed the highest area under the receiver operating characteristic curve: 0.81±0.06 compared with 0.63±0.05 and 0.64±0.05 for the Alberta Stroke Program Early CT Score and ischemic core volume models, respectively. In the external validation, the area under the curve for the convolutional neural network model was significantly superior to those for the other 2 models. Conclusions— Compared with the standard neuroimaging biomarkers, our deep learning model derived a greater amount of prognostic information from pretreatment neuroimaging data. Although a confirmatory prospective evaluation is needed, the high-level imaging features derived by deep learning may offer an effective prognostic imaging biomarker.

Precision medicine in stroke: towards personalized outcome predictions using artificial intelligence

“…age, sex, stroke severity and comorbidities) to generate predictions for 204 patients, and achieved a test set accuracy of 74%. Nishi and colleagues 147 relied on diffusion-weighted MRI (DWI) of 324 patients to predict favourable outcomes, and demonstrated superior test set performance of their deep learning model compared to simpler baseline models (deep learning model: AUC = 0.81 versus ASPECTS: AUC = 0.63). Several further studies considered imaging data from a database of a 132 first-time ischaemic and haemorrhagic stroke patients.…”

Machine Learning-Based Model for Predicting Incidence and Severity of Acute Ischemic Stroke in Anterior Circulation Large Vessel Occlusion

“…In this study, for model derivation, we adopted 5-fold cross-validation, which is a standard way of optimizing the model with inner test data and has been used in a previous study ( 16 ). During modeling, the grid search algorithm which is a greedy algorithm was combined to tune and optimize the model hyperparameters.…”

Artificial Intelligence and Deep Learning in Neuroradiology: Exploring the New Frontier

“…Nishi et al 30 developed a U-Net model to predict clinical posttreatment outcomes using pretreatment diffusion-weighted imaging on patients who underwent mechanical thrombectomy. Clinical outcome was defined using the modified Rankin Scale (mRS) at 90 days after the stroke.…”