Deep learning-based personalised outcome prediction after acute ischaemic stroke

Abstract

BackgroundWhether deep learning models using clinical data and brain imaging can predict the long-term risk of major adverse cerebro/cardiovascular events (MACE) after acute ischaemic stroke (AIS) at the individual level has not yet been studied.MethodsA total of 8590 patients with AIS admitted within 5 days of symptom onset were enrolled. The primary outcome was the occurrence of MACEs (a composite of stroke, acute myocardial infarction or death) over 12 months. The performance of deep learning models (DeepSurv and Deep-Survival-Machines (DeepSM)) and traditional survival models (Cox proportional hazards (CoxPH) and random survival forest (RSF)) were compared using the time-dependent concordance index (Ctdindex).ResultsGiven the top 1 to all 60 clinical factors according to feature importance, CoxPH and RSF yieldedCtdindex of 0.7236–0.8222 and 0.7279–0.8335, respectively. Adding image features improved the performance of deep learning models and traditional models assisted by deep learning models. DeepSurv and DeepSM yielded the bestCtdindex of 0.8496 and 0.8531 when images were added to all 39 relevant clinical factors, respectively. In feature importance, brain image was consistently ranked highly. Deep learning models automatically extracted the image features directly from personalised brain images and predicted the risk and date of future MACEs at the individual level.ConclusionsDeep learning models using clinical data and brain images could improve the prediction of MACEs and provide personalised outcome prediction for patients with AIS. Deep learning models will allow us to develop more accurate and tailored prognostic prediction systems that outperform traditional models.

Knowledge structure and global trends of machine learning in stroke over the past decade: A scientometric analysis

“…Therefore, strengthening the early prediction, diagnosis, and treatment of stroke is particularly important. ML models can help doctors predict [ 43 , 44 ], diagnose [ 45 ], and treat [ 46 ] stroke more accurately and quickly, thus improving the treatment outcomes and reducing morbidity and disability. In a word, the need for early detection, accurate diagnosis, and timely treatment has promoted the increasing application of ML in stroke care.…”

Cox proportional hazards deep neural network identifies peripheral blood complete remission to be at least equivalent to morphologic complete remission in predicting outcomes of patients treated with azacitidine—A prospective cohort study by the AGMT

“…DeepSurv has been shown to be superior to several machine learning and canonical regression survival models and to have the best discriminative performance and calibration at providing accurate predictions of individual survival and at predicting prognosis and risk stratification 36 . Since publication of the method in 2018, 70 publications have applied this technique to data from patients, mostly with solid tumours (e.g., References 35–46 ). DeepSurv has the potential to supplement traditional survival analysis and become a standard method for medical practitioners to study and recommend personalized treatment options, 34–36 which is why we chose this method to strengthen and validate our results.…”

“…This could be observed for all three time-to-event endpoints analyzed. [Color figure can be viewed at wileyonlinelibrary.com] from patients, mostly with solid tumours (e.g., References [35][36][37][38][39][40][41][42][43][44][45][46] ). Deep-Surv has the potential to supplement traditional survival analysis and become a standard method for medical practitioners to study and recommend personalized treatment options, [34][35][36] which is why we chose this method to strengthen and validate our results.…”

Uncertainty-aware deep learning for trustworthy prediction of long-term outcome after endovascular thrombectomy

“…Early approaches utilized traditional regression methods like logistic regression (LR) to predict ischemic stroke outcomes post-intravenous thrombolysis 6 . Since 2014, machine learning (ML) and deep learning (DL) have shown promising results 7 – 13 . Variables selected by experts, obtained using statistical tests, or identified automatically by ML or DL algorithms have been used to pinpoint independent predictors of good clinical outcomes 14 , 15 .…”