A deep learning‐based model for prediction of hemorrhagic transformation after stroke

Abstract

Hemorrhagic transformation (HT) is one of the most serious complications after endovascular thrombectomy (EVT) in acute ischemic stroke (AIS) patients. The purpose of this study is to develop and validate deep‐learning (DL) models based on multiparametric magnetic resonance imaging (MRI) to automatically predict HT in AIS patients. Multiparametric MRI and clinical data of AIS patients with EVT from two centers (data set 1 for training and testing: n = 338; data set 2 for validating: n = 54) were used in the DL models. The acute infarction area of diffusion‐weighted imaging (DWI) and hypoperfusion of perfusion‐weighted imaging (PWI) was labeled manually. Two forms of data sets (volume of interest [VOI] data sets and slice data sets) were analyzed, respectively. The models based on single parameter and multiparameter models were developed and validated to predict HT in AIS patients after EVT. Performance was evaluated by area under the receiver‐operating characteristic curve (AUC), accuracy (ACC), sensitivity, specificity, negative predictive value, and positive predictive value. The results showed that the performance of single parameter model based on MTT (VOI data set: AUC = 0.933, ACC = 0.843; slice data set: AUC = 0.945, ACC = 0.833) and TTP (VOI data set: AUC = 0.916, ACC = 0.873; slice data set: AUC = 0.889, ACC = 0.818) were better than the other single parameter model. The multiparameter model based on DWI & MTT & TTP & Clinical (DMTC) had the best performance for predicting HT (VOI data set: AUC = 0.948, ACC = 0.892; slice data set: AUC = 0.932, ACC = 0.873). The DMTC model in the external validation set achieved similar performance with the testing set (VOI data set: AUC = 0.939, ACC = 0.884; slice data set: AUC = 0.927, ACC = 0.871) (p > 0.05). The proposed clinical, DWI, and PWI multiparameter DL model has great potential for assisting the periprocedural management in the early prediction HT of the AIS patients with EVT.

Pre-thrombectomy prognostic prediction of large-vessel ischemic stroke using machine learning: A systematic review and meta-analysis

“…The mean or median ages of the study participants ranged from 64.0 to 86.0 years, and the proportion of male participants ranged from 35.0 to 65.9%. Only one US study ( 24 ) specifically described the self-reported ethnicity of the patients (63.0–69.0% European ancestry); the other studies reported the place of patient recruitment [USA: 1 ( 32 ); Europe: 10 ( 22 , 23 , 26 – 29 , 31 , 33 – 35 ); Asia: 4 ( 25 , 30 , 36 , 37 )]. The training sample sizes ranged widely, from 109 to 1,401.…”

“…Regarding the testing sample, two studies used hold-out test sets, respectively containing 208 patients ( 30 ) and 100 patients ( 35 ). The remaining studies performed cross-validation ( 23 – 26 , 28 , 29 , 31 – 34 , 36 , 37 ) or bootstrap approach ( 22 , 27 ). The five studies ( 23 , 29 , 31 , 34 , 35 ) used data obtained from MR CLEAN Registry ( 38 ).…”

“…The authors utilized structured receptive field kernels (as opposed to learned convolutional kernels) to help prevent overfitting. Samak et al ( 35 ) and Jiang et al ( 37 ) both used a 3-D CNN feature encoder and incorporated imaging and clinical data using metadata fusion technique. The former additionally used self-attention technique (squeeze and excitation modules) in their encoders, while the latter is based on pre-trained Inception V3 encoders.…”

Prediction Model of Hemorrhage Transformation in Patient with Acute Ischemic Stroke Based on Multiparametric MRI Radiomics and Machine Learning

“…Previous studies failed to fully utilize the MRI images, resulting in low prediction accuracy [ 21 , 22 , 32 ]. Besides, although deep learning has become a powerful tool for classification tasks in recent years, its training process often requires a large amount of data, which is often difficult to meet [ 23 , 24 ]. In addition, the poor interpretability of deep learning limits its wide application in clinical practice.…”

“…Besides, deep learning (DL) has recently emerged as a powerful tool for medical image analysis. As for HT prediction, Jiang et al [ 23 ] built a DL model based on multiparametric MRI images by using diffusion-weighted imaging (DWI), mean transit time (MTT) and TTP sequences combined with clinical data. They first extracted the deep features of each slice using inception V3, then fused the features of all slices of each sequence, and finally spliced the features of all sequences with clinical features to realize HT prediction.…”