Functional Outcome Prediction in Acute Ischemic Stroke Using a Fused Imaging and Clinical Deep Learning Model

Abstract

Background: Predicting long-term clinical outcome based on the early acute ischemic stroke information is valuable for prognostication, resource management, clinical trials, and patient expectations. Current methods require subjective decisions about which imaging features to assess and may require time-consuming postprocessing. This study's goal was to predict ordinal 90-day modified Rankin Scale (mRS) score in acute ischemic stroke patients by fusing a Deep Learning model of diffusion-weighted imaging images and clinical information from the acute period.

Methods: A total of 640 acute ischemic stroke patients who underwent magnetic resonance imaging within 1 to 7 days poststroke and had 90-day mRS follow-up data were randomly divided into 70% (n=448) for model training, 15% (n=96) for validation, and 15% (n=96) for internal testing. Additionally, external testing on a cohort from Lausanne University Hospital (n=280) was performed to further evaluate model generalization. Accuracy for ordinal mRS, accuracy within ±1 mRS category, mean absolute prediction error, and determination of unfavorable outcome (mRS score >2) were evaluated for clinical only, imaging only, and 2 fused clinical-imaging models.

Results: The fused models demonstrated superior performance in predicting ordinal mRS score and unfavorable outcome in both internal and external test cohorts when compared with the clinical and imaging models. For the internal test cohort, the top fused model had the highest area under the curve of 0.92 for unfavorable outcome prediction and the lowest mean absolute error (0.96 [95% CI, 0.77-1.16]), with the highest proportion of mRS score predictions within ±1 category (79% [95% CI, 71%-88%]). On the external Lausanne University Hospital cohort, the best fused model had an area under the curve of 0.90 for unfavorable outcome prediction and outperformed other models with an mean absolute error of 0.90 (95% CI, 0.79-1.01), and the highest percentage of mRS score predictions within ±1 category (83% [95% CI, 78%-87%]).

Conclusions: A Deep Learning-based imaging model fused with clinical variables can be used to predict 90-day stroke outcome with reduced subjectivity and user burden.

Keywords: goal; infarction; ischemic stroke; magnetic resonance imaging; quality of life.

Figure 1:

Training and testing flowcharts for patients in the current study. The gray boxes highlight the test case patients who were completely excluded from all training related to model development.

Figure 2:

(A) The overall architecture of the fused model, which comprises a deep learning-based imaging model and a clinical model. The clinical model uses clinical variables as inputs, while the imaging model includes diffusion-weighted and B0 images. (B) Detailed compositions of the models employed in the study. Specifically, Clinical Model I incorporates the following clinical information: age, gender, race, baseline NIHSS, prior mRS, medical history (including hypertension, diabetes, atrial fibrillation, heart diseases, and previous stroke), occlusion status, and treatment regimen. Clinical Model II extends Clinical Model I by incorporating NIHSS scores obtained after 24 hours. Fused Models I and II are created by integrating the imaging model with Clinical Models I and II, respectively. All models generate continuous predictions of 90-day mRS outcomes.

Figure 3:

MR images (first and second columns are diffusion-weighted imaging (DWI) and B0 images) and corresponding saliency activation maps (rightmost column) generated by deep learning-based imaging models for three patients with varying clinical histories and 90-day mRS scores. The activations are color-coded, with red indicating higher attention. All clinical and fused models in this figure are Type I (without 24-hour NIHSS). The predictions of the model are continuous but rounded to the nearest whole number to facilitate comparison with the true mRS. Patient A, a 39-year-old male, has a history of hypertension and a 90-day mRS of 3. The clinical and fused models correctly predict his score, and the imaging model displays high activations around the lesions. Patient B, a 73-year-old female, has no history of diabetes or hypertension and a 90-day mRS of 0. The fused and clinical models predict a score of 2, while the imaging model predicts 3, with low activations around the stroke lesion. Patient C, a 71-year-old female, has histories of diabetes and hypertension and a 90-day mRS of 6. All three models predict a score of 3, and the imaging model shows high activations around the lesions.

Figure 4:

ROC Comparisons for Clinical Models, Imaging Model, and Fused Models across the two testing cohorts. A) Internal Test (IT) Cohort: Left - Clinical Model I, Imaging Model, and Fused Model I comparisons; Right - Clinical Model II, Imaging Model, and Fused Model II comparisons. B) Lausanne University Hospital (LUH) Cohort: Left - Clinical Model I, Imaging Model, and Fused Model I comparisons; Right - Clinical Model II, Imaging Model, and Fused Model II comparisons. The maximum value of the Youden index was used to determine the optimal cut-off points, shown as solid triangles on the ROC curves.