• 한방비만학회지
• /
• 제21권2호
• /
• pp.80-94
• /
• 2021

Objectives: The purpose of this study is to re-evaluate the importance of questions of obesity syndrome differentiation (OSD) questionnaire based on real-world survey and to explore the possibility of simplifying OSD types. Methods: The OSD frequency was identified, and variance threshold feature selection was performed to filter the questions. Filtered questions were clustered by K-means clustering and hierarchical clustering. After principal component analysis (PCA), the distribution patterns of the subjects were identified and the differences in the syndrome distribution were compared. Results: The frequency of OSD in spleen deficiency, phlegm (PH), and blood stasis (BS) was lower than in food retention (FR), liver qi stagnation (LS), and yang deficiency. We excluded 13 questions with low variance, 7 of which were related to BS. Filtered questions were clustered into 3 groups by K-means clustering; Cluster 1 (17 questions) mainly related to PH, BS syndromes; Cluster 2 (11 questions) related to swelling, and indigestion; Cluster 3 (11 questions) related to overeating or emotional symptoms. After PCA, significant different patterns of subjects were observed in the FR, LS, and other obesity syndromes. The questions that mainly affect the FR distribution were digestive symptoms. And emotional symptoms mainly affect the distribution of LS subjects. And other obesity syndrome was partially affected by both digestive and emotional symptoms, and also affected by symptoms related to poor circulation. Conclusions: In-depth data mining analysis identified relatively low importance questions and the potential to simplify OSD types.

• 사상체질의학회지
• /
• 제33권4호
• /
• pp.1-9
• /
• 2021

Objectives The purpose of this study was to check whether the classification of the individual's Body Mass Index (BMI) could be predicted by analyzing the voice data constructed at the Korean medicine data center (KDC) using machine learning. Methods In this study, we proposed a convolutional neural network (CNN)-based BMI classification model. The subjects of this study were Korean adults who had completed voice recording and BMI measurement in 2006-2015 among the data established at the Korean Medicine Data Center. Among them, 2,825 data were used for training to build the model, and 566 data were used to assess the performance of the model. As an input feature of CNN, Mel-frequency cepstral coefficient (MFCC) extracted from vowel utterances was used. A model was constructed to predict a total of four groups according to gender and BMI criteria: overweight male, normal male, overweight female, and normal female. Results & Conclusions Performance evaluation was conducted using F1-score and Accuracy. As a result of the prediction for four groups, The average accuracy was 0.6016, and the average F1-score was 0.5922. Although it showed good performance in gender discrimination, it is judged that performance improvement through follow-up studies is necessary for distinguishing BMI within gender. As research on deep learning is active, performance improvement is expected through future research.

• 한국의학물리학회지:의학물리
• /
• 제32권4호
• /
• pp.92-98
• /
• 2021

Purpose: This study automatically discriminates homogeneous and structure edge regions on computed tomography (CT) images, and it evaluates the noise level and edge preservation ratio (EPR) according to the different types of iterative reconstruction (IR). Methods: The dataset consisted of CT scans of 10 patients reconstructed with filtered back projection (FBP), statistical IR (iDose⁴), and iterative model-based reconstruction (IMR). Using the 10th and 85th percentiles of the structure coherence feature, homogeneous and structure edge regions were localized. The noise level was estimated using the averages of the standard deviations for five regions of interests (ROIs), and the EPR was calculated as the ratio of standard deviations between homogeneous and structural edge regions on subtraction CT between the FBP and IR. Results: The noise levels were 20.86±1.77 Hounsfield unit (HU), 13.50±1.14 HU, and 7.70±0.46 HU for FBP, iDose⁴, and IMR, respectively, which indicates that iDose⁴ and IMR could achieve noise reductions of approximately 35.17% and 62.97%, respectively. The EPR had values of 1.14±0.48 and 1.22±0.51 for iDose⁴ and IMR, respectively. Conclusions: The iDose⁴ and IMR algorithms can effectively reduce noise levels while maintaining the anatomical structure. This study suggested automated evaluation measurements of noise levels and EPRs, which are important aspects in CT image quality with patients' cases of FBP, iDose⁴, and IMR. We expect that the inclusion of other important image quality indices with a greater number of patients' cases will enable the establishment of integrated platforms for monitoring both CT image quality and radiation dose.

• The Journal of Advanced Prosthodontics
• /
• 제13권4호
• /
• pp.191-204
• /
• 2021

PURPOSE. The purpose of this study is to present a methodology to evaluate the accuracy of intraoral scanners (IOS) used in vivo. MATERIALS AND METHODS. A specific feature-based gauge was designed, manufactured, and measured in a coordinate measuring machine (CMM), obtaining reference distances and angles. Then, 10 scans were taken by an IOS with the gauge in the patient's mouth and from the obtained stereolithography (STL) files, a total of 40 distances and 150 angles were measured and compared with the gauge's reference values. In order to provide a comparison, there were defined distance and angle groups in accordance with the increasing scanning area: from a short span area to a complete-arch scanning extension. Data was analyzed using software for statistical analysis. RESULTS. Deviations in measured distances showed that accuracy worsened as the scanning area increased: trueness varied from 0.018 ± 0.021 mm in a distance equivalent to the space spanning a four-unit bridge to 0.106 ± 0.08 mm in a space equivalent to a complete arch. Precision ranged from 0.015 ± 0.03 mm to 0.077 ± 0.073 mm in the same two areas. When analyzing angles, deviations did not show such a worsening pattern. In addition, deviations in angle measurement values were low and there were no calculated significant differences among angle groups. CONCLUSION. Currently, there is no standardized procedure to assess the accuracy of IOS in vivo, and the results show that the proposed methodology can contribute to this purpose. The deviations measured in the study show a worsening accuracy when increasing the length of the scanning area.

• 한국전문물리치료학회지
• /
• 제28권2호
• /
• pp.123-131
• /
• 2021

Background: Only 2% of falls in older adults result in serious injuries (i.e., hip fracture). Therefore, it is important to differentiate injurious versus non-injurious falls, which is critical to develop effective interventions for injury prevention. Objects: The purpose of this study was to a. extract the best features of surface electromyography (sEMG) for classification of injurious falls, and b. find a best model provided by data mining techniques using the extracted features. Methods: Twenty young adults self-initiated falls and landed sideways. Falling trials were consisted of three initial fall directions (forward, sideways, or backward) and three knee positions at the time of hip impact (the impacting-side knee contacted the other knee ("knee together") or the mat ("knee on mat"), or neither the other knee nor the mat was contacted by the impacting-side knee ("free knee"). Falls involved "backward initial fall direction" or "free knee" were defined as "injurious falls" as suggested from previous studies. Nine features were extracted from sEMG signals of four hip muscles during a fall, including integral of absolute value (IAV), Wilson amplitude (WAMP), zero crossing (ZC), number of turns (NT), mean of amplitude (MA), root mean square (RMS), average amplitude change (AAC), difference absolute standard deviation value (DASDV). The decision tree and support vector machine (SVM) were used to classify the injurious falls. Results: For the initial fall direction, accuracy of the best model (SVM with a DASDV) was 48%. For the knee position, accuracy of the best model (SVM with an AAC) was 49%. Furthermore, there was no model that has sensitivity and specificity of 80% or greater. Conclusion: Our results suggest that the classification model built upon the sEMG features of the four hip muscles are not effective to classify injurious falls. Future studies should consider other data mining techniques with different muscles.

• 한국컴퓨터정보학회논문지
• /
• 제28권1호
• /
• pp.87-92
• /
• 2023

음성 인식 시스템의 지속적인 발전으로 음성에 대한 인식율은 급속도로 발전되었지만 사용 환경에서의 잡음과 여러 음성이 혼합되어 발생하는 잡음으로 정확한 음성을 인식할 수 없는 단점을 가진다. 환경 잡음이 있는 음성을 처리할 때 음성 인식률을 높이기 위해서는 잡음을 제거해야 하며, 기존의 HMM, CHMM, GMM, 그리고 AI 모델이 적용된 DNN에서도 예상치 못한 잡음이 발생하거나 기본적으로 디지털 신호에 양자화 잡음이 추가되면 소스 신호가 변경되거나 손상되어 인식률이 저하된다. 이를 해결하기 위해 각 음성 프레임에 대한 음성 신호의 특징을 효율적으로 추출하기 위해 MFCC를 개선하여 처리하였으며, 음성 신호에 대한 잡음을 제거하기 위해 가우시안 모델을 적용한 잡음 편차 추정을 이용한 잡음 제거 방법을 개선하여 적용하였다. 제안된 모델에 대한 성능 평가는 음성에 대한 정확성 평가를 위해 교차 상관 계수를 사용하여 처리하였으며, 제안하는 방법의 인식률을 평가한 결과 이들에 대한 상관 계수에 대한 평균값 차이는 0.53 dB 개선된 것을 확인하였다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제16권1호
• /
• pp.153-172
• /
• 2022

In this study, the authors intend to analyze factors contributing to research performance using Backpropagation Neural Network and Support Vector Machine. The analyzing factors contributing to lecturer research performance start from defining the features. The next stage is to collect datasets based on defining features. Then transform the raw dataset into data ready to be processed. After the data is transformed, the next stage is the selection of features. Before the selection of features, the target feature is determined, namely research performance. The selection of features consists of Chi-Square selection (U), and Pearson correlation coefficient (CM). The selection of features produces eight factors contributing to lecturer research performance are Scientific Papers (U: 154.38, CM: 0.79), Number of Citation (U: 95.86, CM: 0.70), Conference (U: 68.67, CM: 0.57), Grade (U: 10.13, CM: 0.29), Grant (U: 35.40, CM: 0.36), IPR (U: 19.81, CM: 0.27), Qualification (U: 2.57, CM: 0.26), and Grant Awardee (U: 2.66, CM: 0.26). To analyze the factors, two data mining classifiers were involved, Backpropagation Neural Networks (BPNN) and Support Vector Machine (SVM). Evaluation of the data mining classifier with an accuracy score for BPNN of 95 percent, and SVM of 92 percent. The essence of this analysis is not to find the highest accuracy score, but rather whether the factors can pass the test phase with the expected results. The findings of this study reveal the factors that have a significant impact on research performance and vice versa.

• Structural Engineering and Mechanics
• /
• 제86권3호
• /
• pp.291-309
• /
• 2023

This paper addresses the finite element modeling of functionally graded porous (FGP) beams for free vibration and buckling behaviour cases. The formulated finite element is based on simple and efficient higher order shear deformation theory. The key feature of this formulation is that it deals with Euler-Bernoulli beam theory with only three unknowns without requiring any shear correction factor. In fact, the presented two-noded beam element has three degrees of freedom per node, and the discrete model guarantees the interelement continuity by using both C⁰ and C¹ continuities for the displacement field and its first derivative shape functions, respectively. The weak form of the governing equations is obtained from the Hamilton principle of FGP beams to generate the elementary stiffness, geometric, and mass matrices. By deploying the isoparametric coordinate system, the derived elementary matrices are computed using the Gauss quadrature rule. To overcome the shear-locking phenomenon, the reduced integration technique is used for the shear strain energy. Furthermore, the effect of porosity distribution patterns on the free vibration and buckling behaviours of porous functionally graded beams in various parameters is investigated. The obtained results extend and improve those predicted previously by alternative existing theories, in which significant parameters such as material distribution, geometrical configuration, boundary conditions, and porosity distributions are considered and discussed in detailed numerical comparisons. Determining the impacts of these parameters on natural frequencies and critical buckling loads play an essential role in the manufacturing process of such materials and their related mechanical modeling in aerospace, nuclear, civil, and other structures.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2023년도 학술발표회
• /
• pp.168-168
• /
• 2023

The frequent occurrence of cyanobacterial harmful algal blooms (CHABs) in inland waters under climate change seriously damages the ecosystem and human health and is becoming a big problem in South Korea. Satellite remote sensing is suggested for effective monitoring CHABs at a larger scale of water bodies since the traditional method based on sparse in-situ networks is limited in space. However, utilizing a standalone variable of satellite reflectances in common CHABs dual-models, which relies on both chlorophyll-a (Chl-a) and phycocyanin or cyanobacteria cells (Cyano-cell), is not fully beneficial because their seasonal variation is highly impacted by surrounding meteorological and bio-environmental factors. Along with the development of Artificial Intelligence (AI), monitoring CHABs from space with analyzing the effects of environmental factors is accessible. This study aimed to investigate the potential application of AI in the dual-model strategy (Chl-a and Cyano-cell are output parameters) for monitoring seasonal dynamics of CHABs from satellites over Korean inland waters. The Sentinel-3 satellite was selected in this study due to the variety of spectral bands and its unique band (620 nm), which is sensitive to cyanobacteria. Via the AI-based feature selection, we analyzed the relationships between two output parameters and major parameters (satellite water-leaving reflectances at different spectral bands), together with auxiliary (meteorological and bio-environmental) parameters, to select the most important ones. Several AI models were then employed for modelling Chl-a and Cyano-cell concentration from those selected important parameters. Performance evaluation of the AI models and their comparison to traditional semi-analytical models were conducted to demonstrate whether AI models (using water-leaving reflectances and environmental variables) outperform traditional models (using water-leaving reflectances only) and which AI models are superior for monitoring CHABs from Sentinel-3 satellite over a Korean inland water body.

• 한국음향학회지
• /
• 제42권2호
• /
• pp.94-101
• /
• 2023

음성 향상에서 많이 사용되는 U-Net과 같이 인코더와 디코더로 구성된 심층 신경망 모델은 skip-connection을 통해 인코더의 특징을 디코더에 연결하는 구조로 구성되어 있다. Skip-connection은 디코더에서 향상된 스펙트럼을 재구성하는데 도움을 주며 인코더를 통해 손실된 정보를 보완해줄 수 있다. 이때 skip-connection을 통해 연결되는 인코더의 특징과 디코더의 특징의 의미는 서로 다르다. 본 논문에서는 복소 스펙트럼 기반 음성 향상의 성능 향상을 위해 디코더에 연결되는 인코더의 특징을 디코더 특징의 의미에 가깝게 변환해주도록 skip-connection에 Self-Attention(SA)을 적용하는 방안을 연구하였다. SA는 시퀀스-시퀀스 문제에서 출력 시퀀스를 생성할 때, 입력 시퀀스의 가중 산술 평균을 이용하여 결정적인 부분을 집중해서 볼 수 있도록 하는 기법으로, 음성 향상 분야에서도 이를 적용함으로써 성능 향상에 효과적임을 입증하는 연구가 진행되었다. SA를 skip-connection에 적용하기 위해 인코더 특징과 디코더 특징을 이용하는 총 3가지의 방법에 대해 연구하였다. TIMIT 데이터베이스를 이용한 음성 향상 실험 결과, 제안하는 방법이 기존 skip-connection으로만 연결된 Deep Complex U-Net(DCUNET)과 비교하여 모든 성능 평가 지표에서 향상된 결과를 보였다.