Three- and two-dimensional deep neural network for acute ischemic stroke identification in T1-weighted magnetic resonance imaging

Deep neural networks (DNNs) are increasingly being utilized in both computer vision and analysis of medical images. Three-dimensional (3D) convolutional neural networks (CNNs) can extract spatiotemporal features from 3D medical images and can be used to classify anatomical structures. However, it increases the time complexity of the training process, which is why 3D CNNs are often used in combination with traditional two-dimensional (2D) CNNs. The diagnosis of stroke lesions relies critically on magnetic resonance imaging (MRI). Expert experience is required for accurate manual detection, which is time-consuming. Computational power has allowed CNNs to perform on par with or better than clinicians in many tasks. A DNN ResNet34–AlexNet combination was utilized to analyze MR images for diagnosing acute ischemic stroke (AIS). A performance comparison was made between 2D and 3D state-of-the-art CNNs. More computational resources and time are required to train the 3D CNN model than its 2D counterpart. The proposed model achieved an accuracy of 54.55% compared with the VGG16 model in the 3D MRI and 42.94% in the 2D MRI. A T1-weighted MR image was used in the study to compare the performance of 2D and 3D CNNs for identifying AIS.