https://doi.org/10.1140/epjs/s11734-024-01423-9
Regular Article
A new deep learning-based GUI design and implementation for automatic detection of brain strokes with CT images
1
Department of Computer Engineering, Faculty of Engineering, Sakarya Applied Sciences University, Sakarya, Turkey
2
Department of Computer Engineering, Graduate Education Institue, Sakarya Applied Sciences University, Sakarya, Turkey
Received:
31
August
2024
Accepted:
22
November
2024
Published online:
9
December
2024
Brain stroke is a disease that can occur in almost any age group, especially in people over 65. There are two main types of strokes: ischemic stroke and hemorrhagic stroke. Blockage of brain vessels causes ischemic stroke, while rupture of blood vessels in or around the brain causes hemorrhagic stroke. According to the World Health Organization, 5 million people die annually from this disease; 85% of these 5 million are paralyzed by ischemic stroke and 15% by hemorrhagic stroke. For these reasons, especially in the early diagnosis and treatment of ischemic stroke, patients can lead to more comfortable lives. In this study, we evaluate the effectiveness of real-time object detection algorithms for the detection of brain strokes in computed tomography images and propose an artificial intelligence-supported system for busy physicians to quickly analyze computed tomography images. The aim of this study is to compare the performance of YOLOv7, YOLOv8, and YOLOv9 models in the detection of ischemic and hemorrhagic strokes from brain computed tomography images, to compare the performance of YOLO-based algorithms known as real-time object detection networks in segmentation with other segmentation networks known as U-Net and U-Net variants and Mask-RCNN algorithms, and to develop a system that doctors can use to analyze stroke computed tomography images. In this study, 6951 anonymized brain computed tomography slices obtained from the Turkish Ministry of Health were used. The YOLOv7, YOLOv8, and YOLOv9 models are trained using deep learning algorithms. Model training and testing processes were performed with the PyTorch deep learning framework and CUDA acceleration. The dataset consists of anonymized brain computed tomography images collected between 2019 and 2020. Experimentally, three different studies were conducted using ischemic stroke-health images, hemorrhagic stroke-health images, and ischemic stroke–hemorrhagic stroke-health images together for comparison with the literature. In all studies, the YOLOv9-Seg model is successful. In the ischemic stroke-health images study, 99.50% segment mAP@0.5 success was achieved; in the hemorrhagic stroke-health images study, 99.49% segment mAP@0.5 success; and in the ischemic stroke–hemorrhagic stroke-health images study, %99.71 segment mAP@0.5 success was achieved. The YOLOv7 and YOLOv8 models also exhibited high accuracy rates but lagged behind the YOLOv9-Seg model. These findings suggest that the YOLOv9-Seg model is the most suitable model for the detection and segmentation of ischemic and hemorrhagic strokes in brain CT images. The real-time processing capability of the model will help to make fast and accurate decisions in emergency medical interventions. In addition, this study has shown that YOLO-based models can be used effectively in health data. It is thought that the use of the model in clinical applications will make significant contributions to early diagnosis and treatment processes.
Copyright comment Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
© The Author(s), under exclusive licence to EDP Sciences, Springer-Verlag GmbH Germany, part of Springer Nature 2024
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
