Stroke is the third leading cause of death worldwide after cancer and heart disease and a leading cause of long-term adult disability, affecting 15 million people worldwide each year. Over the past decade, we have witnessed increasing interest in applying machine learning technologies for stroke image analysis. The reason for this is that deep learning has shown substantial success in the field of machine learning. However, there are still challenges in this field. The machine learning models are unable to generalize from one center's data to the data from other centers due to overfitting and underfitting issues. Developing reliable stroke imaging real-time applications is difficult due to the need for well-defined training, validation, and testing datasets. Given the heterogeneity in both injury and recovery, it is challenging to predict outcomes reliably soon after a stroke. Moreover, deep learning's "black-box" nature precludes clinicians from identifying and addressing biases within the algorithms. Artificial intelligence (AI) in stroke image analysis with clinical applications still needs more research and investigation.
This Research Topic focuses on new developments in AI for stroke image analysis in acute and chronic phases. The types of studies focusing on the following areas are particularly of interest:
• Research focusing on new methods for stroke image analysis, with particular emphasis on efforts to apply explainable artificial intelligence, generalization, weakly supervised learning, and computer vision to real-time stroke applications.
• Studies using novel methods to contribute to the basic science of processing and analyzing using stroke images.
• Novel approaches utilizing stroke imaging datasets at all spatial scales;
• The typical stroke imaging datasets of interest are those obtained by Magnetic Resonance Imaging (MRI) and Computed Tomography (CT);
• New developments and implementations of artificial intelligence algorithms and strategies based on the use of various deep learning models to solve the following types of problems using stroke imaging datasets: early diagnosis, prediction, treatment decisions, generalization, adaptation, visualization, feature extraction, segmentation, registration, ischemic stroke, radiological outcomes, clinical mortality outcomes, functional outcomes, and digital anatomical atlases.
Stroke is the third leading cause of death worldwide after cancer and heart disease and a leading cause of long-term adult disability, affecting 15 million people worldwide each year. Over the past decade, we have witnessed increasing interest in applying machine learning technologies for stroke image analysis. The reason for this is that deep learning has shown substantial success in the field of machine learning. However, there are still challenges in this field. The machine learning models are unable to generalize from one center's data to the data from other centers due to overfitting and underfitting issues. Developing reliable stroke imaging real-time applications is difficult due to the need for well-defined training, validation, and testing datasets. Given the heterogeneity in both injury and recovery, it is challenging to predict outcomes reliably soon after a stroke. Moreover, deep learning's "black-box" nature precludes clinicians from identifying and addressing biases within the algorithms. Artificial intelligence (AI) in stroke image analysis with clinical applications still needs more research and investigation.
This Research Topic focuses on new developments in AI for stroke image analysis in acute and chronic phases. The types of studies focusing on the following areas are particularly of interest:
• Research focusing on new methods for stroke image analysis, with particular emphasis on efforts to apply explainable artificial intelligence, generalization, weakly supervised learning, and computer vision to real-time stroke applications.
• Studies using novel methods to contribute to the basic science of processing and analyzing using stroke images.
• Novel approaches utilizing stroke imaging datasets at all spatial scales;
• The typical stroke imaging datasets of interest are those obtained by Magnetic Resonance Imaging (MRI) and Computed Tomography (CT);
• New developments and implementations of artificial intelligence algorithms and strategies based on the use of various deep learning models to solve the following types of problems using stroke imaging datasets: early diagnosis, prediction, treatment decisions, generalization, adaptation, visualization, feature extraction, segmentation, registration, ischemic stroke, radiological outcomes, clinical mortality outcomes, functional outcomes, and digital anatomical atlases.