Z Froozesh talab
Z Sabzavi
K Abareshi
M Robatjazi
A Rostami
E Pouretemad

Rationale: Although non-contrast computed tomography (NCCT) is the most widely used clinical imaging modality for acute ischemic stroke (AIS),but it cannot detect significant changes in early infarction. Our goal is to develop a deep learning model to identify early invisible AIS in NCCT and evaluate the diagnostic performance to help radiologists decide the type of treatment method.methodIn this study, we worked on two datasets, (۱) original data set and (۲) Independent test set.The ischemic area in early stages is not enhanced on CT images and is visible only on MRI images.The AIS lesions were confirmed based on the follow-up diffusion weighted imaging and clinical diagnosis .We decided to use our data to train Convolutional neural network (CNN). For this purpose, we performed pre-processing on the images to improve the processing time, resource consumption and model efficiency. In the first step, we registered the CT images on DWI and obtained masks of the target regions by segmenting stroke lesion. We converted the target areas into binary masks with class ۰ (healthy) and ۱ (stroke). we decided to perform the batchify operation on the images, It means to divide them into smaller images. We defined the appropriate structure for U-net ۲d network and trained the model with allocated ۲۰% of the data for testing and ۸۰% of the data for training.Results: ۱۵۰ patients (median age, ۵۰ years) were assigned to training and internal validation groups. This model has sensitivity ۸۳.۶۱% , specificity ۶۸.۹۹% , and accuracy, ۸۹.۸۷%.Conclusions:This deep learning model solves the challenge of not seeing AIS invisible lesions in NCCT and saves more time.With the help of this model, radiologists can provide more effective guidance in making patients’ treatment plan in clinic.

Acute ischemic stroke ,Deep learning, Non-contrast computed tomography, Computer-aided detection , artificial intelligence , deep-learning