Stroke patients may show early neurological deterioration and relapse in the acute phase, which may critically impair their prognosis. Therefore, strategies to prevent early neurological deterioration and stroke recurrence in stroke patients are necessitated.

In this study, we aimed to develop a deep learning-based artificial intelligence algorithm using clinical and neuroimaging data, which predicts early neurological deterioration in the acute phases. This deep learning-based artificial intelligence algorithm could be applied for developing effective treatment strategies in biomedicine.

The aim of the study is to assess whether baseline functional connectivity within the visual cortices predicts visual perceptual learning-induced recovery from visual field defect in chronic ischemic stroke.

We hypothesized that higher functional connectivity within the interhemispheric visual regions at baseline could predict greater visual field defect recovery induced by visual perceptual learning related to chronic stroke.

The identified brain biomarker for predicting visual field defect recovery induced by visual perceptual learning may potentially facilitate development of individualized neurorehabilitation protocols.

Automatic detection and quantification of ischemic stroke lesions using brain magnetic resonance imaging can be helpful for timely and efficient clinical practice.

This study is aimed 1) to improve the performance of U-net for the lesion detection by incrementally increasing the amount of learning data through active learning, and 2) to develop a review system that can efficiently give feedback to clinicians based on the prediction results of the model that has been developed.

Therefore, we ultimately aimed to develop a methodology which could improve the reliability of data.

Intravenous thrombolysis, which can be administered only to patients within 4.5 hours of the ischemic stroke, may improve the patient's neurological symptoms by dissolving blood clots blocking the blood vessels in the brain.

However, approximately 20-25% of all cerebral infarctions has unclear onset time, making it difficult to prescribe thrombolysis. The mismatch between the visualized lesions on diffusion-weighted imaging (DWI) and the absent lesions on fluid-attenuated inversion recovery imaging (FLAIR) has been suggested to indicate the stroke onset within 4.5 hours.

We aimed to develop a machine learning model that can predict thrombolysis and prognosis of each patient by combining the DWI/FLAIR mismatch information based on machine learning on clinical imaging and the clinical variables.

As rehabilitation training involves repetitive training of simple tasks, it is significant to create an environment where patients can consistently focus on training tasks.

Through this study, we developed a patient-specific difficulty adjustment system based on the real-time training data, in order to improve the concentration and compliance with the rehabilitation training.

This study is based on the flow theory that positive concentration on the training task is possible when the task is appropriate for the difficulty level.

Predicting the prognosis after stroke onset is critical for the strategical approaches for the long-term prognosis as well as the short-term treatment of the patients.

We aimed to develop a machine learning algorithm that predicts a patient's long-term global functional outcome using the clinical and medical image data collected during the acute phases, which may improve effective decision-making in the clinical field.