복시환자의 임상양상

김민석,최진,김정훈,김재석,이주화,Min Seok Kim,Jin Choi,Jung Hoon Kim,Jae Suk Kim,Joo Hwa Lee 대한안과학회 2013 대한안과학회지 Vol.54 No.11

Purpose: To evaluate the clinical features, causes and outcomes of patients with diplopia. Methods: All patients presenting with diplopia from October 2010 to March 2012 and followed up for more than 3 months were retrospectively investigated. Results: During the study period, 59 patients with diplopia were identified. There were 42 males and 17 females with an average age of 50 years. Binocular diplopia accounted for 54 cases (92%) and 5 cases (8%) had monocular diplopia. Cranial nerve palsies were the most common cause of binocular diplopia (28 cases, 52%). Within the cranial nerve palsies group, 14 cases (50%) were accompanied by hypertension or diabetes mellitus. Binocular diplopia spontaneously resolved in 35 cases (65%) by 3 months rising to 41 cases (76%) by 7 months. Thirteen (93%) out of 14 cases of cranial nerve palsies with hypertension or diabetes resolved spontaneously by 3 months. Conclusions: Binocular diplopia was caused most commonly by cranial nerve palsy and resolved after 3 months in 65% of patients. A spontaneous recovery from diplopia was observed after 3 months in 93% of patients with cranial nerve palsies and microvascular disease such as hypertension or diabetes. Therefore, the initial observation without additional treatment would be sufficient in these patients. J Korean Ophthalmol Soc 2013;54(11):1772-1777

열처리 및 단조조건에 따른 알루미늄 6056 소재의 특성변화 및 자동차의 피스톤 블록 설계

김민석,정병덕,박효영,최정묵,김정민,박준식,Kim, Min Seok,Jung, Hyung Duck,Park, Hyo young,Choi, Jeong Mook,Kim, Jeong Min,Park, Joon Sik 한국재료학회 2015 한국재료학회지 Vol.25 No.10

The mechanical properties and microstructures of Aluminum 6056 alloys were investigated for their use in the fabrication of a piton block. The EN-AW6056 alloys exhibited a tensile strength of 375 MPa for a solution treatment temperature of $550^{\circ}C$ for 2 h followed by an aging treatment at $190^{\circ}C$ for 4 h. The microstructures of the heat treated specimen showed that the $Mg_2Si$ phase with a size of 3~5 um was dispersed throughout the aluminum matrix when the heat treatment was done. Moreover, in order to identify the forgeability of the specimen, upsetting tests were done. For up to 80 % of the upsetting ratio, the specimen maintained its original shape, and at above 80 % of the upsetting ratio, the specimen underwent crack development. The specimen was successfully forged without any defects with the examined material conditions. The material conditions together with the forging conditions are discussed in terms of the microstructures and mechanical properties.

갑상선독성주기마비 증례; 마비증상시와 회복 후에 연속적으로 시행한 신경전도 검사 결과

김민석,박정아,석정임,Kim, Min Suck,Park, Jung A,Seok, Jung Im 대한임상신경생리학회 2015 Annals of Clinical Neurophysiology Vol.17 No.2

Simplexity in Hydrology: 복잡함과 간단함 사이에서의 균형

김민석,Kim, Min-Seok 한국수자원학회 2021 물과 미래(한국수자원학회지) Vol.54 No.9

심층신경망 구조에 따른 구개인두부전증 환자 음성 인식 향상 연구

김민석,정재희,정보경,윤기무,배아라,김우일,Kim, Min-seok,Jung, Jae-hee,Jung, Bo-kyung,Yoon, Ki-mu,Bae, Ara,Kim, Wooil 한국음향학회 2019 韓國音響學會誌 Vol.38 No.6

본 논문에서는 구개인두부전증(VeloPharyngeal Insufficiency, VPI) 환자의 음성을 효과적으로 인식하기 위해 컨볼루션 신경망 (Convolutional Neural Network, CNN), 장단기 모델(Long Short Term Memory, LSTM) 구조 신경망을 은닉 마르코프 모델(Hidden Markov Model, HMM)과 결합한 하이브리드 구조의 음성 인식 시스템을 구축하고 모델 적응 기법을 적용하여, 기존 Gaussian Mixture Model(GMM-HMM), 완전 연결형 Deep Neural Network(DNN-HMM) 기반의 음성 인식 시스템과 성능을 비교한다. 정상인 화자가 PBW452단어를 발화한 데이터를 이용하여 초기 모델을 학습하고 정상인 화자의 VPI 모의 음성을 이용하여 화자 적응의 사전 모델을 생성한 후에 VPI 환자들의 음성으로 추가 적응 학습을 진행한다. VPI환자의 화자 적응 시에 CNN-HMM 기반 모델에서는 일부층만 적응 학습하고, LSTM-HMM 기반 모델의 경우에는 드롭 아웃 규제기법을 적용하여 성능을 관찰한 결과 기존 완전 연결형 DNN-HMM 인식기보다 3.68 % 향상된 음성 인식 성능을 나타낸다. 이러한 결과는 본 논문에서 제안하는 LSTM-HMM 기반의 하이브리드 음성 인식 기법이 많은 데이터를 확보하기 어려운 VPI 환자 음성에 대해 보다 향상된 인식률의 음성 인식 시스템을 구축하는데 효과적임을 입증한다. This paper proposes speech recognition systems employing Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) structures combined with Hidden Markov Moldel (HMM) to effectively recognize the speech of VeloPharyngeal Insufficiency (VPI) patients, and compares the recognition performance of the systems to the Gaussian Mixture Model (GMM-HMM) and fully-connected Deep Neural Network (DNNHMM) based speech recognition systems. In this paper, the initial model is trained using normal speakers' speech and simulated VPI speech is used for generating a prior model for speaker adaptation. For VPI speaker adaptation, selected layers are trained in the CNN-HMM based model, and dropout regulatory technique is applied in the LSTM-HMM based model, showing 3.68 % improvement in recognition accuracy. The experimental results demonstrate that the proposed LSTM-HMM-based speech recognition system is effective for VPI speech with small-sized speech data, compared to conventional GMM-HMM and fully-connected DNN-HMM system.

승차감 관련 주관평가와 객관평가의 상관성 연구

김민석,김연태,문원길,안세진,유완석,Kim, Min-Seok,Kim, Yon-Tae,Moon, Won-Kil,Ahn, Se-Jin,Yoo, Wan-Suk 한국자동차공학회 2007 한국 자동차공학회논문집 Vol.15 No.5

In order to compare subjective and objective assessments, a passenger car was driven at several speeds over several road profiles. To measure the acceleration signals experienced by the seated subject who provided an subjective assessment, four triaxial translational accelerometers and one triaxial gyro sensor were mounted on the steering wheel and on the passenger seat and floor, respectively. Correlations were determined between the measured accelerations and the subjective assessments of 3 expert subjects and 9 general subjects using psychophysical power law.

당뇨망막병증에서 나비라스 레이저 시스템과 기존 범망막광응고술의 소요시간과 통증 비교

김민석,이성우,김재석,Min Seok Kim,Seong Woo Lee,Jae Suk Kim 대한안과학회 2014 대한안과학회지 Vol.55 No.8

Purpose: To compare the pain scale and time necessary for panretinal photocoagulation (PRP) between Navilas<sup><font size=`2`>Ⓡ</font> (OD-OS, Teltow, Germany) and conventional laser in diabetic retinopathy. Methods: Fifteen patients who required PRP for diabetic retinopathy were enrolled in the present study. PRP was performed using Navilas<sup><font size=`2`>Ⓡ</font> (5 x 5 array patterned system) in the superior, nasal and inferior areas, and using conventional laser at the temporal area 1 week later. Total time of laser application and number of laser shots were counted for calculating required time per 100 spots of each laser system. Immediately after the laser photocoagulation, patients were asked to quantify their pain on a visual analog pain scale (0 = no pain; 10 = worst pain). Results: PRP using Navilas<sup><font size=`2`>Ⓡ</font> required shorter time per 100 laser spots (27.7 sec vs. 102.0 sec, <em>p</em> < 0.001) and subjects had lower treatment-related pain than with the conventional laser system (3.3 vs. 6.9,<em> p</em> < 0.001). Conclusions: PRP using Navilas<sup><font size=`2`>Ⓡ</font> can be considered as an efficient method for improving patient and operator's comfort with faster laser application and lower treatment-related pain. J Korean Ophthalmol Soc 2014;55(8):1150-1154

「내외수 연계 내수침수예측시스템」 기술 소개

김민석,육지문,문영일,Kim, Min-Seok,Yuk, Ji-Mun,Mun, Yeong-Il 한국수자원학회 2017 물과 미래(한국수자원학회지) Vol.50 No.4

영상처리 기법을 이용한 샥-하트만 파면 센서의 측정범위 확장에 대한 연구

김민석,김지연,엄태경,윤성기,이준호,Kim, Min-Seok,Kim, Ji-Yeon,Uhm, Tae-Kyung,Youn, Sung-Kie,Lee, Jun-Ho 한국광학회 2007 한국광학회지 Vol.18 No.6

샥-하트만 센서는 렌즐릿 배열을 이용하여 파면의 국부적인 기지를 계산하고 이로부터 파면의 왜곡된 정도를 측정하는 파면측정 센서이다. 일반적으로 하나의 점영상에 대한 중심점을 추출하여 기준 점영상의 중심점에 대한 상대적 이동량을 통해 파면의 국부 기울기를 계산, 전체 파면을 복원하는 원리이다. 따라서 하나의 점영상에 대한 정확한 중심점의 추출을 위하여 해당하는 하부개구 내에 점영상이 존재하여야 한다. 하지만 큰 기울기의 파면이 샥-하트만 센서에 입사하게 되면 측정범위를 벗어나게 되어 일반적인 방법으로는 파면의 국부적인 기울기를 계산할 수 없게 되는 문제점이 있다. 이를 해결하기 위해 특정 측정시스템에만 국한된 알고리즘이 아닌 모든 파면에 대해 측정 가능하면서도 추가적인 장비 없이 일반적인 샥-하트만 파면 센서가 지니고 있는 측정범위의 한계를 극복할 수 있는 알고리즘이 필요하다. 본 연구에서는 영상처리 기법을 이용하여 각 점영상에 대한 유동적인 탐색영역을 설정하고 이러한 탐색영역 내에서 중심점을 추출하는 방법을 제안하였다. 또 점영상 대응법을 적용하여 측정 점영상과 기준 점영상을 일대일 대응시킴으로써 전체 측정파면을 복원할 수 있는 알고리즘을 제안함으로써 측정범위에 제한 없이 파면왜곡을 측정할 수 있는 측정방법을 제시하였다. 이렇게 제안된 알고리즘을 샥-하트만 파면 센서 시스템을 이용한 초점벗어남 측정실험을 통하여 검증하였다. The Shack-Hartmann wavefront sensor is composed of a lenslet array generating the spot images from which local slope is calculated and overall wavefront is measured. Generally the principle of wavefront reconstruction is that the spot centroid of each lenslet array is calculated from pixel intensity values in its subaperture, and then overall wavefront is reconstructed by the local slope of the wavefront obtained by deviations from reference positions. Hence the spot image of each lenslet array has to remain in its subaperture for exact measurement of the wavefront. However the spot of each lenslet array deviates from its subaperture area when a wavefront with large local slopes enters the Shack-Hartmann sensor. In this research, we propose a spot image searching method that finds the area of each measured spot image flexibly and determines the centroid of each spot in its area Also the algorithms that match these centroids to their reference points unequivocally, even if some of them are situated off the allocated subaperture, are proposed. Finally we verify the proposed algorithm with the test of a defocus measurement through experimental setup for the Shack-Hartmann wavefront sensor. It has been shown that the proposed algorithm can expand the dynamic range without additional devices.

2.3 GHz 대역에서 단일 Half-LO 주파수를 이용한 Double-Conversion Down Mixer 설계

김민석,문주영,윤상원,Kim Min-Seok,Moon Ju-Young,Yun Sang-Won 한국전자파학회 2006 한국전자파학회논문지 Vol.17 No.8

본 논문에서는 Half-LO 주파수를 이용하는 새로운 방식의 double-conversion down mixer를 2.3 GHz 대역에서 설계하였다. 제안된 mixer에서는 Half-LO 주파수를 2개의 HEMT에 인가하여 두 HEMT에서의 변환 방식을 gate type과 resistive type으로 다르게 함으로서 IF 주파수를 얻게 된다. 기존의 sub-harmonic mixers와 같이 Half-LO 주파수를 사용하지만 제안된 mixer는 Half-LO의 second harmonic 성분을 이용하지 않고 Half-LO의 기본 주파수를 이용하여 주파수 변환을 하고 두 번째 HEMT에서 Resistive mixer type으로 IF 신호를 얻기 때문에 기존의 능동 mixer보다 향상된 선형성을 얻을 수 있다. 2.3 GHz 대역에서 설계 제작된 mixer는 Half-LO 전력 10 dBm을 이용하여 5dB의 변환 손실과 16.25dBm의 IIP3 특성을 얻을 수 있었다. In this paper, we designed the double conversion down mixer by using Half-LO frequency in 2.3 GHz band. The IF frequency is obtained by supplying two LO frequencies to HEMT in both gate type and resistive type. The proposed mixer uses Half-LO frequency the same way as conventional sub-harmonic mixers. However the proposed one uses fundamental component of Half-LO frequency in first stage instead of using second harmonic components of Half-LO frequency, and the IF frequency is obtained by resistive type mixer in second stage, thereby the proposed mixer can improve linearity in comparison with conventional active mixer. We can verify that the proposed mixer has an conversion loss of 5dBm and IIP3 of 16.25dBm by using 10 dBm Power.