• International journal of advanced smart convergence
• /
• 제8권1호
• /
• pp.24-34
• /
• 2019

In this research, a practical deep learning framework to differentiate the lesions and nodules in breast acquired with ultrasound imaging has been proposed. 7408 ultrasound breast images of 5151 patient cases were collected. All cases were biopsy proven and lesions were semi-automatically segmented. To compensate for the shift caused in the segmentation, the boundaries of each lesion were drawn using Fully Convolutional Networks(FCN) segmentation method based on the radiologist's specified point. The data set consists of 4254 benign and 3154 malignant lesions. In 7408 ultrasound breast images, the number of training images is 6579, and the number of test images is 829. The margin between the boundary of each lesion and the boundary of the image itself varied for training image augmentation. The training images were augmented by varying the margin between the boundary of each lesion and the boundary of the image itself. The images were processed through histogram equalization, image cropping, and margin augmentation. The networks trained on the data with augmentation and the data without augmentation all had AUC over 0.95. The network exhibited about 90% accuracy, 0.86 sensitivity and 0.95 specificity. Although the proposed framework still requires to point to the location of the target ROI with the help of radiologists, the result of the suggested framework showed promising results. It supports human radiologist to give successful performance and helps to create a fluent diagnostic workflow that meets the fundamental purpose of CADx.

• 대한의용생체공학회:의공학회지
• /
• 제43권5호
• /
• pp.361-368
• /
• 2022

The purpose of this study was to develop a hand rehabilitation training system for hemiplegic patients. We also tried to find out five hand postures (WF: Wrist Flexion, WE: Wrist Extension, BG: Ball Grip, HG: Hook Grip, RE: Rest) in real-time using multi-channel EMG-based deep learning. We performed a pre-processing method that converts to Spider Chart image data for the classification of hand movement from five test subjects (total 1,500 data sets) using Convolution Neural Networks (CNN) deep learning with an 8-channel armband. As a result of this study, the recognition accuracy was 92% for WF, 94% for WE, 76% for BG, 82% for HG, and 88% for RE. Also, ten physical therapists participated for the usability evaluation. The questionnaire consisted of 7 items of acceptance, interest, and satisfaction, and the mean and standard deviation were calculated by dividing each into a 5-point scale. As a result, high scores were obtained in immersion and interest in game (4.6±0.43), convenience of the device (4.9±0.30), and satisfaction after treatment (4.1±0.48). On the other hand, Conformity of intention for treatment (3.90±0.49) was relatively low. This is thought to be because the game play may be difficult depending on the degree of spasticity of the hemiplegic patient, and compensation may occur in patient with weakened target muscles. Therefore, it is necessary to develop a rehabilitation program suitable for the degree of disability of the patient.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권12호
• /
• pp.4439-4455
• /
• 2021

Compared with two-stage object detection algorithms, one-stage algorithms provide a better trade-off between real-time performance and accuracy. However, these methods treat the intermediate features equally, which lacks the flexibility to emphasize meaningful information for classification and location. Besides, they ignore the interaction of contextual information from different scales, which is important for medium and small objects detection. To tackle these problems, we propose an image pyramid network based on dual attention mechanism (DAIPNet), which builds an image pyramid to enrich the spatial information while emphasizing multi-scale informative features based on dual attention mechanisms for one-stage object detection. Our framework utilizes a pre-trained backbone as standard detection network, where the designed image pyramid network (IPN) is used as auxiliary network to provide complementary information. Here, the dual attention mechanism is composed of the adaptive feature fusion module (AFFM) and the progressive attention fusion module (PAFM). AFFM is designed to automatically pay attention to the feature maps with different importance from the backbone and auxiliary network, while PAFM is utilized to adaptively learn the channel attentive information in the context transfer process. Furthermore, in the IPN, we build an image pyramid to extract scale-wise features from downsampled images of different scales, where the features are further fused at different states to enrich scale-wise information and learn more comprehensive feature representations. Experimental results are shown on MS COCO dataset. Our proposed detector with a 300 × 300 input achieves superior performance of 32.6% mAP on the MS COCO test-dev compared with state-of-the-art methods.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제16권9호
• /
• pp.2991-3007
• /
• 2022

Two dimensional locality preserving projections (2D-LPP) is an improved algorithm of 2D image to solve the small sample size (SSS) problems which locality preserving projections (LPP) meets. It's able to find the low dimension manifold mapping that not only preserves local information but also detects manifold embedded in original data spaces. However, 2D-LPP is simple and elegant. So, inspired by the comparison experiments between two dimensional linear discriminant analysis (2D-LDA) and linear discriminant analysis (LDA) which indicated that matrix based methods don't always perform better even when training samples are limited, we surmise 2D-LPP may meet the same limitation as 2D-LDA and propose a novel matrix exponential method to enhance the performance of 2D-LPP. 2D-MELPP is equivalent to employing distance diffusion mapping to transform original images into a new space, and margins between labels are broadened, which is beneficial for solving classification problems. Nonetheless, the computational time complexity of 2D-MELPP is extremely high. In this paper, we replace some of matrix multiplications with multiple multiplications to save the memory cost and provide an efficient way for solving 2D-MELPP. We test it on public databases: random 3D data set, ORL, AR face database and Polyu Palmprint database and compare it with other 2D methods like 2D-LDA, 2D-LPP and 1D methods like LPP and exponential locality preserving projections (ELPP), finding it outperforms than others in recognition accuracy. We also compare different dimensions of projection vector and record the cost time on the ORL, AR face database and Polyu Palmprint database. The experiment results above proves that our advanced algorithm has a better performance on 3 independent public databases.

• Smart Structures and Systems
• /
• 제29권1호
• /
• pp.181-193
• /
• 2022

Data-driven structural health monitoring (SHM) of civil infrastructure can be used to continuously assess the state of a structure, allowing preemptive safety measures to be carried out. Long-term monitoring of large-scale civil infrastructure often involves data-collection using a network of numerous sensors of various types. Malfunctioning sensors in the network are common, which can disrupt the condition assessment and even lead to false-negative indications of damage. The overwhelming size of the data collected renders manual approaches to ensure data quality intractable. The task of detecting and classifying an anomaly in the raw data is non-trivial. We propose an approach to automate this task, improving upon the previously developed technique of image-based pre-processing on one-dimensional (1D) data by enriching the features of the neural network input data with multiple channels. In particular, feature engineering is employed to convert the measured time histories into a 3-channel image comprised of (i) the time history, (ii) the spectrogram, and (iii) the probability density function representation of the signal. To demonstrate this approach, a CNN model is designed and trained on a dataset consisting of acceleration records of sensors installed on a long-span bridge, with the goal of fault detection and classification. The effect of imbalance in anomaly patterns observed is studied to better account for unseen test cases. The proposed framework achieves high overall accuracy and recall even when tested on an unseen dataset that is much larger than the samples used for training, offering a viable solution for implementation on full-scale structures where limited labeled-training data is available.

• Advances in nano research
• /
• 제12권2호
• /
• pp.185-195
• /
• 2022

Deep learning is another field of artificial intelligence (AI) utilized for computer aided diagnosis (CAD) and image processing in scientific research. Considering numerous mechanical repetitive tasks, reading image slices need time and improper with geographical limits, so the counting of image information is hard due to its strong subjectivity that raise the error ratio in misdiagnosis. Regarding the highest mortality rate of Lung cancer, there is a need for biopsy for determining its class for additional treatment. Deep learning has recently given strong tools in diagnose of lung cancer and making therapeutic regimen. However, identifying the pathological lung cancer's class by CT images in beginning phase because of the absence of powerful AI models and public training data set is difficult. Convolutional Neural Network (CNN) was proposed with its essential function in recognizing the pathological CT images. 472 patients subjected to staging FDG-PET/CT were selected in 2 months prior to surgery or biopsy. CNN was developed and showed the accuracy of 87%, 69%, and 69% in training, validation, and test sets, respectively, for T1-T2 and T3-T4 lung cancer classification. Subsequently, CNN (or deep learning) could improve the CT images' data set, indicating that the application of classifiers is adequate to accomplish better exactness in distinguishing pathological CT images that performs better than few deep learning models, such as ResNet-34, Alex Net, and Dense Net with or without Soft max weights.

• Physical Therapy Rehabilitation Science
• /
• 제12권2호
• /
• pp.80-91
• /
• 2023

Objective: The purpose of this study is to identify factors that affect the incidence of hypertension using logistic regression and decision tree analysis, and to build and compare predictive models. Design: Secondary data analysis study Methods: We analyzed 9,859 subjects from the Korean health panel annual 2019 data provided by the Korea Institute for Health and Social Affairs and National Health Insurance Service. Frequency analysis, chi-square test, binary logistic regression, and decision tree analysis were performed on the data. Results: In logistic regression analysis, those who were 60 years of age or older (Odds ratio, OR=68.801, p<0.001), those who were divorced/widowhood/separated (OR=1.377, p<0.001), those who graduated from middle school or younger (OR=1, reference), those who did not walk at all (OR=1, reference), those who were obese (OR=5.109, p<0.001), and those who had poor subjective health status (OR=2.163, p<0.001) were more likely to develop hypertension. In the decision tree, those over 60 years of age, overweight or obese, and those who graduated from middle school or younger had the highest probability of developing hypertension at 83.3%. Logistic regression analysis showed a specificity of 85.3% and sensitivity of 47.9%; while decision tree analysis showed a specificity of 81.9% and sensitivity of 52.9%. In classification accuracy, logistic regression and decision tree analysis showed 73.6% and 72.6% prediction, respectively. Conclusions: Both logistic regression and decision tree analysis were adequate to explain the predictive model. It is thought that both analysis methods can be used as useful data for constructing a predictive model for hypertension.

• 한국컴퓨터정보학회논문지
• /
• 제28권10호
• /
• pp.9-16
• /
• 2023

태권도 겨루기의 전자호구, 축구의 VAR 등 스포츠에서 기술 발전이 고도화되고 있다. 하지만 태권도 품새는 사람이 직접 자세를 눈으로 보고 판단하며 지도하기 때문에 때로는 대회의 현장에서 판정시비가 일어난다. 본 연구는 인공지능을 이용하여 태권도 동작을 더 정확하게 판단하고 평가할 수 있는 인공지능 모델을 제안한다. 본 연구에서는 촬영 및 수집한 데이터를 전처리한 후 학습, 테스트, 검증 세트로 분리한다. 분리한 데이터를 각 모델과 조건을 적용하여 학습한 후 비교하여 가장 좋은 성능의 모델을 제시한다. 각 조건의 모델은 정확도, Precision, Recall, F1-Score, 학습 소요 시간, Top-n error의 값을 비교하였고 그 결과 ResNet50과 Adam을 사용한 조건에서 학습한 모델의 성능이 가장 우수한 것으로 나타났다. 본 연구에서 제시한 모델을 활용하여 교육 현장이나 대회 등 다양한 방면에서 활용할 수 있을 것으로 기대한다.

• 벤처창업연구
• /
• 제9권1호
• /
• pp.119-132
• /
• 2014

본 연구는 코스닥 기업의 횡령 배임 및 최대주주변경을 고려한 부실기업 예측 모형을 연구하였다. 모형개발을 위해 2009년부터 2012년까지 코스닥시장에서 상장폐지된 제조기업 83개사를 부실기업표본으로 선정하였고. 정상기업 표본은 같은 기간 코스닥시장에 상장되어 정상적인 영업활동을 하고 있으며 부실기업과 동일아이템 혹은 동종업종에 속한 83개사를 선정하여 총 166개사를 쌍대표본 추출법으로 구성하였다. 본 연구는 상기 표본기업의 상장폐지 직전 5년간 재무비율 80개를 선정하여 T-test를 실시하여 유의미한 변수 중에서 5년 연속 출현한 19개를 도출하였고 전진선택법을 이용하여 로지스틱 회귀분석 모형식을 추정하였다. 기존 연구에서는 상장폐지 직전 3년간 자료만을 분석하였으나 본 연구는 직전 5년간 자료를 분석하여 기업이 부실화되는 초기과정부터 어떤 유의미한 재무적 특성이 시차를 두고 부실화에 영향을 미치는 지를 연구했다는 점과 선행 연구에서 시도되지 않은 횡령 배임과 최대 주주변경이라는 비재무적인 특성을 더미변수로써 고려된 부실기업예측모형을 구축하여 그 정보의 유용함을 실증적으로 분석한 점이 기존 선행연구들과 차별화 된다. 연구결과, 더미변수를 추가한 모형의 판별력은 T-1년에 95.2%, T-2년에 88.0%, T-3년에 81.3%, T-4년에 79.5%, T-5년에 74.7%로 나타났으며, 상장폐지 년도에 가까워지면서 판별력도 점차 올라갔으며 기존 선행연구의 결과보다도 대체로 높은 판별력을 보였다. 본 연구가 사전에 부실화될 가능성이 높은 기업을 찾아냄으로써 해당기업은 물론 투자자, 금융기관 및 기타 이해관계자들의 피해를 조금이나마 줄여 줄 수 있을 것이라고 기대된다.

• 한국전자통신학회논문지
• /
• 제13권3호
• /
• pp.661-668
• /
• 2018

피부 저항을 이용한 생체 신호는 스트레스성 질환에 따라 각각 다른 특성을 보이고 있으며 이 특성을 이용하여 스트레스성 질환을 진단하는 생체진단 장비들이 개발 되었으며, 장비들은 피부 저항 측정기에서 측정한 신호를 해석하기 쉽게 출력해주며, 그 분야의 전문가는 출력 신호를 직접 보고 어떤 스트레스성 질환의 가능성이 높은지를 판단하게 된다. 하지만 각 측정 대상자에게서 측정된 생체 신호를 분석하여 측정 대상자가 어떤 스트레스성 질환을 가지고 있는지를 사람이 정확히 판단하기는 매우 어려울 뿐만 아니라 판단의 결과가 잘못될 가능성도 매우 높다. 이런 문제점을 해결하기 위하여 본 연구에서는 머신러닝 기법을 이용하여 측정된 신호가 어떤 스트레스성 질환의 신호에 해당하는지를 판단하는 기능을 구현하였다. 측정 장비의 낮은 컴퓨팅 능력을 고려하여 분류 기법은 SVM을 사용하였으며, 훈련 데이터와 테스트 데이터는 13개의 질환을 중심으로 오차범위 5를 사용하여 각 질환 당 1,000개를 랜덤하게 생성하여 사용하였다. 모의실험 결과에서 90% 이상의 판단 정확도를 보였으며 앞으로 측정 장비가 실제로 환자들에게 적용되면 다시 생성된 데이터로 분류기를 재훈련 할 수 있게 구성하였다.