To search, Click below search items.


All Published Papers Search Service


A Hybrid ResNet50-UNet Model for Ischemic Stroke brain Segmentation from MRI Images


Fathia ABOUDI, Cyrine DRISSI, Tarek KRAIEM


Vol. 22  No. 11  pp. 467-476


Ischemic brain stroke is the most common cere brovascular disease and one of the leading causes of death and long-term disabilities worldwide. Early detection of ischemic brain stroke helps physicians to take a precocious diagnosis which significantly reduces possible cases of death or disabilities. Several modalities are used to detect Ischemic brain stroke in medical research; though, magnetic resonance imaging (MRI) remains the most effective modality in this field. Recently, many researchers used deep learning models to detect ischemic brain stroke in MRI images and have proven encouraging results. In this paper, we present an automated approach for segmenting ischemic stroke lesions (ISL) from MRI images using a deep learning model. The UNet used model is used as a hybrid framework with a pre-trained ResNet50 architecture. Data augmentation technique has been applied to outperform the model’s accuracy. The proposed workflow has been trained and tested on a public Ischemic Stroke Lesion Segmentation challenge (ISLES) 2015 datasets. The experimental findings demonstrate the efficiency of the performance of our approach, it achieves a 99.43% average accuracy, and a 64.14% Dice Coefficient(DC). Our approach outperforms other state-of-the-art methods, more specifically, for the accuracy values.


Medical image segmentation, Ischemic stroke dis- ease, UNet, ResNet50, MRI, Transfer learning, data augmentation