단일 병원에서의 소아뇌졸중의 원인, 임상적 양상 및 기능회복에 대한 연구

류주석,박진홍,박은하,차은혜,성인영 대한재활의학회 2009 Annals of Rehabilitation Medicine Vol.33 No.3

Objective: To investigate the changes of causes, clinical features, and functional outcomes in childhood strokes. Method: This study included 152 patients, aged from 1 to 18 years, who were diagnosed with stroke and admitted to a tertiary hospital between January 2000 and April 2004. All medical records and neurologic images of the patients were reviewed. A parental questionnaire was used to investigate patients' functional outcomes. These results were compared with those of the previous study performed in the same hospital in 2001. Results: The number of hemorrhagic stroke was 78 (51.3%) and that of ischemic stroke was 74 (48.7%). When compared to the previous study, the incidence of hemorrhagic stroke especially above the age of 10 years decreased and that of ischemic stroke below the age of 10 years increased. The causes of stroke were arteriovenous malformation (AVM, 42.8%), Moyamoya disease (37.5%), vasculitis (5.3%), cardiac disease (3.9%), hematologic disease (2.0%), and undetermined (8.5%). Common clinical features were headache (53.8%), vomiting (43.6%) and loss of consciousness (28.2%) in the hemorrhagic stroke, and hemiparesis (94.6%), headache (35.1%) and speech disorder (31.1%) in the ischemic stroke. 86.0% of the hemorrhagic and 64.8% of the ischemic stroke patients were categorized in the ‘good' outcome group. Conclusion: The incidence of ischemic stroke increased to the similar level of hemorrhagic stroke. The most common causes were AVM in the hemorrhagic and Moyamoya disease in the ischemic stroke. Most of these patients showed good functional outcome, regardless of the causes of stroke. Objective: To investigate the changes of causes, clinical features, and functional outcomes in childhood strokes. Method: This study included 152 patients, aged from 1 to 18 years, who were diagnosed with stroke and admitted to a tertiary hospital between January 2000 and April 2004. All medical records and neurologic images of the patients were reviewed. A parental questionnaire was used to investigate patients' functional outcomes. These results were compared with those of the previous study performed in the same hospital in 2001. Results: The number of hemorrhagic stroke was 78 (51.3%) and that of ischemic stroke was 74 (48.7%). When compared to the previous study, the incidence of hemorrhagic stroke especially above the age of 10 years decreased and that of ischemic stroke below the age of 10 years increased. The causes of stroke were arteriovenous malformation (AVM, 42.8%), Moyamoya disease (37.5%), vasculitis (5.3%), cardiac disease (3.9%), hematologic disease (2.0%), and undetermined (8.5%). Common clinical features were headache (53.8%), vomiting (43.6%) and loss of consciousness (28.2%) in the hemorrhagic stroke, and hemiparesis (94.6%), headache (35.1%) and speech disorder (31.1%) in the ischemic stroke. 86.0% of the hemorrhagic and 64.8% of the ischemic stroke patients were categorized in the ‘good' outcome group. Conclusion: The incidence of ischemic stroke increased to the similar level of hemorrhagic stroke. The most common causes were AVM in the hemorrhagic and Moyamoya disease in the ischemic stroke. Most of these patients showed good functional outcome, regardless of the causes of stroke.

머신러닝 기술을 활용한 뇌졸중 분류방안연구

채명건,sabina,황보택근 차세대컨버전스정보서비스학회 2021 차세대컨버전스정보서비스기술논문지 Vol.10 No.1

Stroke is the most common, dangerous single institutional disease and exacerbates the social burden in an aging society. Stroke can be examined through various imaging methods, and diagnosing stroke using CT images has the advantage of fewer space constraints and faster shooting time. However, the diagnosis through images is very difficult, so it is a big disadvantage of this method. In this paper, the study was conducted based on facial image data, not based on CT images, through the characteristics of imbalance in the face of stroke patients. Based on this concept, we proposed a preprocessing algorithm optimized for stroke using facial image data of patients. We developed machine learning-based algorithms by finding elements that can be quantified from facial images, and used data from stroke patients in actual hospitals to evaluate their performance. In order to evaluate the performance of the proposed algorithm and system, two hospital neurosurgeons diagnosed stroke separately for each stroke data and evaluated their performance by comparing their diagnosis with system results. Additionally, cross-checking and feedback allowed us to identify the underlying problem of stroke. 뇌졸중은 가장 흔하고, 위험한 단일 기관 질환이며 고령화 사회에서 사회적 부담을 악화시킨다. 뇌졸중은 다양한 영상 검사법을 통해 검사할 수 있으며, CT 영상을 이용해 뇌졸중을 진단하면 공간 제약이 적고 촬영 시간이 빠르다는 장점이 있다. 그러나 영상을 통한 진단은 매우 어렵기 때문에 이 방법의 큰 단점이다. 본 논문에서는 뇌졸중 환자의 얼굴에서는 불균형이 나타난다는 특징을 통해 CT 이미지 기반이 아닌, 얼굴 이미지 데이터를 기반으로 학습을 진행했다. 이 개념을 바탕으로 환자의 얼굴 이미지 데이터를 이용해 뇌졸중에 최적화된 전처리 알고리즘을 제안했다. 얼굴 이미지에서 수치화 시킬 수 있는 요소를 찾아내어 머신러닝 기반 알고리즘을 개발하였고, 성능을 평가하기 위하여 실제 병원에서 뇌졸중 환자의 데이터를 활용하였다. 이번 연구는 제안된 알고리즘과 시스템의 성능을 평가하기 위해 병원 신경외과 전문의 2명이 뇌졸중 데이터마다 별도로 뇌졸중을 진단해 자신의 진단을 시스템 결과와 비교해 성과를 평가했다. 추가로 교차 검진과 피드백을 통해 뇌졸중의 근본적 문제를 확인할 수 있었다.

Offline Handwritten Numeral Recognition Using Multiple Features and SVM classifier

Kim, Gab-Soon,Park, Joong-Jo Institute of Korean Electrical and Electronics Eng 2015 전기전자학회논문지 Vol.19 No.4

In this paper, we studied the use of the foreground and background features and SVM classifier to improve the accuracy of offline handwritten numeral recognition. The foreground features are two directional features: directional gradient feature by Kirsch operators and directional stroke feature by local shrinking and expanding operations, and the background feature is concavity feature which is extracted from the convex hull of the numeral, where the concavity feature functions as complement to the directional features. During classification of the numeral, these three features are combined to obtain good discrimination power. The efficiency of our scheme is tested by recognition experiments on the handwritten numeral database CENPARMI, where SVM classifier with RBF kernel is used. The experimental results show the usefulness of our scheme and recognition rate of 99.10% is achieved.

Offline Handwritten Numeral Recognition Using Multiple Features and SVM classifier

김갑순,박중조 한국전기전자학회 2015 전기전자학회논문지 Vol.19 No.4

In this paper, we studied the use of the foreground and background features and SVM classifier to improve the accuracy of offline handwritten numeral recognition. The foreground features are two directional features: directional gradient feature by Kirsch operators and directional stroke feature by local shrinking and expanding operations, and the background feature is concavity feature which is extracted from the convex hull of the numeral, where the concavity feature functions as complement to the directional features. During classification of the numeral, these three features are combined to obtain good discrimination power. The efficiency of our scheme is tested by recognition experiments on the handwritten numeral database CENPARMI, where SVM classifier with RBF kernel is used. The experimental results show the usefulness of our scheme and recognition rate of 99.10% is achieved.

Stroke Width-Based Contrast Feature for Document Image Binarization

( Le Thi Khue Van ),( Guee Sang Lee ) 한국정보처리학회 2014 Journal of information processing systems Vol.10 No.1

Automatic segmentation of foreground text from the background in degraded document images is very much essential for the smooth reading of the document content and recognition tasks by machine. In this paper, we present a novel approach to the binarization of degraded document images. The proposed method uses a new local contrast feature extracted based on the stroke width of text. First, a pre-processing method is carried out for noise removal. Text boundary detection is then performed on the image constructed from the contrast feature. Then local estimation follows to extract text from the background. Finally, a refinement procedure is applied to the binarized image as a post-processing step to improve the quality of the final results. Experiments and comparisons of extracting text from degraded handwriting and machine-printed document image against some well-known binarization algorithms demonstrate the effectiveness of the proposed method.

뇌졸중 검출을 위한 복합 생체 정보를 활용한 멀티 모델 프레임워크

최형선,김재승,황보택근 차세대컨버전스정보서비스학회 2022 차세대컨버전스정보서비스기술논문지 Vol.11 No.3

According to the World Health Organization, the world's population is rapidly aging. This is expected to increase medical costs and be the source of various chronic diseases. According to the World Health Organization, stroke ranks second in the world's death toll and the number continues to rise. Researchers from each country have reported various risk factors through various studies and clinical trials, and are aware of the seriousness of stroke. Previous studies have detected symptoms of stroke and investigated causality. Also, with the development of artificial intelligence, it was successful to measure the distortion of the face, which is one of the symptoms from stroke patients. However, stroke can show other symptoms, such as voice tremors, in addition to facial paralysis. Therefore, this study proposes a deep learning model framework using artificial intelligence, focusing on the inarticulateness of speech and the distortion of the face among the symptoms of stroke. The proposed model applied transfer learning to improve accuracy and overcome limitations of insufficient datasets. As a result, the performance was improved by 0.7% in training accuracy, for validation, 13.9% in stroke accuracy, and 4.6% in general patients. 세계보건기구에 따르면 세계 인구는 빠르게 고령화를 향하고 있다. 이는 곧 의료비용의 증가와 각종 만성 질환의 근원지로 예상된다. 세계보건기구에 따르면 뇌졸중은 전 세계 사망원인의 2위를 차지하고 있으며 그 수는 계속해서 증가하고 있다. 이에 각국의 연구원들은 다양한 연구와 임상 실험을 통해 여러 위험 요소가 보고하였고, 뇌졸중의 심각성을 인지하고 있다. 기존의 연구들은 뇌졸중의 증상을 탐지하고, 인과관계를 조사하였다. 또한 인공지능의 발전으로 증상 중 하나인 얼굴의 일그러진 정도를 측정하며 증상 여부를 판별하는데 성공하였다. 그러나 뇌졸중은 얼굴의 마비 증상 외에도 목소리의 떨림 등 다른 증상들을 띌 수 있다. 따라서 본 연구에서는 뇌졸중의 증상 중 말의 어눌함과 얼굴의 일그러진 정도에 초점을 맞추어 인공지능을 활용한 딥러닝 모델 프레임워크를 제안한다. 제안된 모델은 정확도 향상과 부족한 데이터셋의 한계점을 극복하기 위해 전이학습을 적용하였다. 그 결과 훈련 정확도에서 0.7%, 검증 정확도에서는 뇌졸중 환자는 13.9% 일반 환자는 4.6% 개선된 성능을 보였다.

딥러닝과 XGBoost를 이용한 뇌졸중전조증상진단 애플리케이션

노정현,전왕수,이상용 한국지능시스템학회 2023 한국지능시스템학회논문지 Vol.33 No.4

In this paper, after segmenting the lip area in the facial image, using the featurepoints of the lip, it is determined whether there is a precursor symptom of stroke. UNet and FCN, which are semantic image segmentation models, were used forsegmentation of the lip region, and at this moment, VGGNet16, ResNet101, andDenseNet121 were used as backbone networks. As a result of the experiment, themIoU of UNet using DenseNet121 was the highest at 92.5%. And, as a result oflearning with XGBoost using the feature map of the lip area and diagnosing astroke, it showed 98.8% accuracy. As a result of comparison with the existingstroke diagnosis method, the accuracy improved by 7.74~10.8%.

Stroke Width-Based Contrast Feature for Document Image Binarization

Van, Le Thi Khue,Lee, Gueesang Korea Information Processing Society 2014 Journal of information processing systems Vol.10 No.1

Automatic segmentation of foreground text from the background in degraded document images is very much essential for the smooth reading of the document content and recognition tasks by machine. In this paper, we present a novel approach to the binarization of degraded document images. The proposed method uses a new local contrast feature extracted based on the stroke width of text. First, a pre-processing method is carried out for noise removal. Text boundary detection is then performed on the image constructed from the contrast feature. Then local estimation follows to extract text from the background. Finally, a refinement procedure is applied to the binarized image as a post-processing step to improve the quality of the final results. Experiments and comparisons of extracting text from degraded handwriting and machine-printed document image against some well-known binarization algorithms demonstrate the effectiveness of the proposed method.

딥러닝과 XGBoost를 이용한 뇌졸중 질환 예측

노정현(Jeong-Hyun Noh),전왕수(Wang-Su Jeon),이상용(Sang-Yong Rhee) 한국정보기술학회 2022 Proceedings of KIIT Conference Vol.2022 No.6

본 논문에서는 FAST 검사방법 중 하나인 얼굴영역을 이용하여 뇌졸중 조기 증상 진단 분류 방법을 제안하였다. 이 방법은 U-Net과 FCN을 이용하여 얼굴 영역의 입술을 분할하고, 입술영역의 특징을 사용하여 정상, 비정상을 분류한다. 분할 모델은 U-Net과 FCN의 백본망은 ResNet101과 VGGNet16을 사용했고, ResUnet이 87.8%의 정확도를 보였다. 그리고 분류성능을 비교한 결과 96.7%로 가장 좋은 성능을 보였다. In this paper, the use of face region, one of the FAST approaches, it is proposed this research for early stroke classification. This approach uses U-Net and FCN to segment lips in the face region and discriminate between positive and negative based on lip features. ResNet101 and VGGNet16 are used in the segmentation model for the U-Net and FCN backbone networks, respectively, and ResUnet has accuracy of 87.8%. In addition, due to the normal and abnormal comparison using xgboost by extracting features from the lips, the accuracy was 96.6%.

MSER를 위한 획 너비변환과 특징추출

김민우(Min-Woo Kim),오일석(Il-Seok Oh) 한국정보과학회 2014 정보과학회 컴퓨팅의 실제 논문지 Vol.20 No.1

장면 텍스트 추출은 다양한 이미지 기반의 응용에 중요한 단서를 제공한다는 점에서 중요하다. MSER(maximally stable extremal regions)는 가장 우수한 영역검출 알고리즘 중 하나로 전처리로서 텍스트 후보 영역을 찾는데 자주 사용된다. MSER 추출 후 각 영역이 텍스트인지 판단하기 위해 획 정보를 추출해야 한다. 기존에 제안된 SWT(stroke width transform)는 에지를 기반으로 밝기 이미지를 획 너비 이미지로 변환하여 획 정보를 추출한다. 이 방법은 획 정보 추출이 쉬우나 에지를 기반으로 동작하기 때문에 MSER에 바로 적용할 수 없다. 본 논문에서는 MSER의 획 너비 이미지를 생성하는 새로운 방법을 제안한다. 특히 제안하는 방법은 획의 너비 정보뿐만 아니라 영역이 평행한 경계선을 갖는 정도를 나타내는 특징도 추출할 수 있으며, 이들 특징은 텍스트와 비텍스트를 구분하는 데 유용하다. Scene text extraction is crucial for diverse image-based applications due to the fact that the text provides contextual clues. Maximally stable extremal regions (MSER) are one of the most outstanding region detectors, and often used to detect text candidates in the preprocessing step of scene text extraction. Once MSER is extracted, stroke information is needed to classify each candidate regions as text or background. The conventional stroke width transform (SWT) converts an intensity image into stroke width image based on edge points, and then extracts stroke information. This method is helpful to extract stroke information, but inappropriate for MSER to apply directly due to the fact that works based on edge points. In this paper, we propose a new method to make stroke width image for MSER. Especially, the proposed method can extract useful feature to classify a region into text or nontext about a degree that region is composed of parallel boundary besides stroke width information.