[논문]마이크로스코프 이미지의 딥러닝 기반 이상검출

정희용; 고정원; 신춘성

doi:10.9723/jksiis.2021.26.2.001

마이크로스코프 이미지의 딥러닝 기반 이상검출
Abnormal Detection with Microscope through Deep Learning 원문보기

한국산업정보학회논문지 = Journal of the Korea Industrial Information Systems Research, v.26 no.2, 2021년, pp.1 - 10

정희용 (전남대학교 AI융합대학 IoT인공지능융합전공) , 고정원 (전남대학교 AI융합대학 IoT인공지능융합전공) , 신춘성 (전남대학교 문화전문대학원 미디어예술공학전공)

초록
AI-Helper

흡연자 중에서 담배가 인체에 유해하다는 사실을 모르는 사람은 없을 것임에도 불구하고, 정작 금연 성공률은 높지 않다. 금연을 위한 의지를 지속적으로 굳건하게 다지기 위하여 병원에서 실시하는 건강검진과 PET(positron emission tomography) 이미지를 통한 암 검사의 결과가 도움이 되지만, 일상생활 중에 간단히 실시할 수 있는 방법이 아니다. 본 연구에서는 일상생활 중에 관찰 가능한 흡연자의 신체 부위를 딥러닝 기반 마이크로스코프 이미지 측정 및 분석을 통하여 흡연자와 비흡연자의 차이를 검출할 수 있는 비침습적 방법을 제안하였다. 우선, 관찰 부위를 흡연시 직접적인 접촉을 하는 혓바닥 표면으로 설정하였다. 다음으로, 마이크로스코프로 혓바닥 표면(410배 확대)을 흡연자 10명과 비흡연자 10명의 실험 참가자를 통하여 데이터 셋(총 1,000장)을 구축하여 그 중 80%를 딥러닝 모델의 학습에 사용하였고, 나머지 20%는 예측에 사용하였다. 딥러닝 모델을 스케일링하는 방법(width scaling, depth scaling, resolution scaling) 중 한 가지 방법만 적용하는 VGG, ResNet, DenseNet과 세 가지를 모두 적용하여 스케일링하는 EfficientNet의 성능을 비교하여 모세혈관 이미지 처리에 EfficientNet의 우수성을 확인해 볼 수 있었다.

Abstract ▼ AI-Helper

The success rate of the no-smoking campaign has been low, although everybody knows that cigarettes are harmful to the human health. The results of both regular health and cancer checks in the hospital are useful for strengthening the human intention for quitting the smoking, however, those methods are difficult to use in daily life because of the use of large-scaled particular devices such as PET(positron emission tomography). Thus, this study proposed a non-invasive method that detects the difference between smokers and non-smokers through deep-learning-based analysis. At first, observing parts were decided to the tongue surface. Then, a data set(total 1,000) was made through the experiment to measure the tongue surface(410 times magnification) with the participants of 10 smokers and 10 non-smokers. The 80% ratio of data set was used for the train, and the left 20% was for the prediction. As a result, it was found that the classification through EfficientNet with the compound scaling including three scaling methods of width scaling, depth scaling and resolution scaling was much better than other models including VGG, ResNet, and DenseNet with the only one scaling.

주제어

표/그림 (8)

그림 Figure 1. A description of the microscope with the low and high magnification.
그림 Figure 2. An explanation to measure the tongue surface.
그림 Figure 3. Model scaling. (a) is a baseline network example; (b)^~(d) are conventional scaling that only increases one dimension of network width, depth, or resolution. (e) is the compound scaling method that uniformly scales all three dimension with a fixed ratio (Tan et al, 2019).
그림 Figure 4. An example of dataset used for this study.
그림 Figure 5. Results of performance from EfficientNet-B0 to EfficientNet-B3 and the regulated EfficientNet-B1 through target size, batch size, and dropout.
그림 Figure 6. Relationship between the number of parameters and the accuracy to decide the hyperparameter for good tuning.
그림 Figure 7. Results of comparison with VGG16, ResNet50, DenseNet121, and EfficientNet-B1.
그림 Figure 8. Results of tangue buds between smokers and non-smokers.

참고문헌 (29)

Ayanian, JZ. (1999). Perceived risks of heart disease and cancer among cigarette smokers, JAMA, 281(11), (pp. 1019-1-21).

상세보기
Carosella, AM, Ossip-Klein, DJ, Watt, CA, and Podgorski, C. (2002). Smoking history, knowledge, and attitudes among older residents of a long-term care facility, Nicotine & Tobacco Research, 4, (pp. 161-169).

상세보기
Chen, WW, and Lindsey, R. (2001). Evaluation of a tobacco prevention program on knowledge, attitudes, intention and behavior of tobacco use among fourth grade students - a preliminary study, J Drug Educ, 31(4), (pp. 399-410).

상세보기
Frank, E, Denniston, M, and Pederson, L. (2002). Declines in smokers' understanding of tobacco's hazards between 1986 and 1998: a report from north Georgia, South Med J , 95(7), (pp. 675-680).

상세보기
Geckova, A, Van Dijk, JP, van lttersum-Gritter, T, and Groothoff, JW. (2002). Determinants of adolescents' smoking behaviour: A literature review, Cent Eur J publ Health, 10(3), (pp. 79-87).
Hakeem, R, Thomas, J, and Badruddin, SH. (2001). Urbanisation and health related knowledge and attitudes of South Asian children, J Pak Med Assoc, 51(12), (pp. 437-443).
He, K, Zhang, X, Ren, S, and Sun, J. (2016). Deep residual learning for image recognition, CVPR, (pp. 770-778).
He, Y, Lin, J, Liu, Z, Wang, H, Li, LJ, and Han, S. (2018). Amc: Automl for model compression and acceleration on mobile devices, ECCV.
Howard, AG, Zhu, M, Chen, B, Kalenichenko, D, Wang, W, Weyand, T, Andreetto, M, and Adam, H. (2017). Mobilenets: Efficient convolutional neural networks for mobile vision applications, arXiv preprint arXiv: 1704.04861.
Hu, J, Shen, L, and Sun, G. (2018). Squeeze-and-excitation networks, CVPR.
Huang, Y, Cheng, Y, Chen, D, Lee, H, Ngiam, J, Le, QV, and Chen, Z. (2018). Gpipe: Efficient training of giant neural networks using pipeline parallelism, arXiv preprint arXiv: 1808.07233.
Jeong, YS, Lee, HR, Baek, JH, Kim, KH, Chung YS, and Lee, CW. (2020). Deep Learning-based Rice Seed Segmentation for Phynotyping, Journal of the Korea Industrial Information Systems Research, 25(5), (pp. 23-29).

원문보기 상세보기
Kale, YS, Vibhute, N, Belgaumi, U, Kadashetti, V, Bommanavar, S, and Kamate, W. (2019). Effect of using tobacco on taste perception, J Family Med P rim Care, 8(8), (pp. 2699-2702).
Kim, SW, Cha, KA, and Park, SH. (2020). Lip-reading System based on Bayesian Classifier, Journal of the Korea Industrial Information Systems Research, 25(4), (pp. 9-16).

원문보기 상세보기
Kim, DH, Hwang, BW, Lee, SW, and Kwak, SY. (2020). 3D Human Shape Deformation using Deep Learning, Journal of the Korea Industrial Information Systems Research, 25(2), (pp. 19-27).

원문보기 상세보기
Kornblith, S, Shlens, J, and Le, QV. (2019). Do better imagenet models transfer better?, CVPR.
Krizhevsky, A, Sutskever, I, and Hinton, GE. (2012). Imagenet classification with deep convolutional neural networks, NIPS, (pp. 1097-1105).
Lu, Z, Pu, H, Wang, F, Hu, Z, and Wang, L. (2018). The expressive power of neural networks: A view from the width, NeurIPS.
Pavlos, P, Vasilios, N, Antonia, A, Dimitrios, K, Georgios, K, and Georgios, A. (2009). Evaluation of young smokers and non-smokers with Electrogustometry and Contact Endoscopy, BMC Ear Nose Throat Disord, 9(9), doi: 10.1186/1472-6815-9-9.

상세보기
Sargent, JD, Beach, ML, Dalton, MA, Mott, LA, and Tickle, JJ, (2001). Ahrens, MB, Heatherton, TF. Effect of seeing tobacco use in films on trying smoking among adolescents: cross sectional study, BMJ , 323(7326), (1394-1397).

상세보기
Sharir, O, and Shashua, A. (2018). On the expressive power of overlapping architectures of deep learning, ICLR.
Suga, S, Otomo, A, Jeong, H, and Ohno, Y. (2019). Image Similarity Check of Nailfold Capillary by Template Matching, IEEE 8th Global Conference on Consumer Elecctronics (GCCE), Osaka, Japan.
Szegedy, C, Liu, W, Jia, Y, Sermanet, P, Reed, S, Anguelov, D, Erhan, D, Vanhoucke, V, and Rabinovich, A. (2015). Going deeper with convolutions, CVPR, (pp. 1-9).
Tan, M, and V. Le, Q. (2019). EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks, Proceedings of the 36th International Conference on Machine Learning (ICML), Long Beach, California, PMLR 97.
Tomassini, S, Cuoghi, V, Catalani, E, Casini, G, and Bigiani, A. (2007). Long-term effects of nicotine on rat fungiform taste buds, Neuroscience, 147(3), (pp. 803-810).

상세보기
Torabi, MR, Yang, J, and Li, J. (2002). Comparison of tobacco use knowledge, attitude and practice among college students in China and the United states, Health promot int, 17(3), (pp. 247-254).

상세보기
World Health Organization (WHO). (1983). Guidelines for the conduct of tobacco-smoking surveys among health professionals, Report of a WHO meeting held in Winnipeg, Canada.
Woo, E. (2003). Study on the Knowledge Level of Smoking and Smoking Behavior, Master Thesis, The Graduate School of Yonsei Univeristy, Seoul, Korea.
Zagoruyko, S, and Komodakis, N. (2016). Wide residual networks, BMVC.

표제어: PCR

동의어: Packet Collision Rate

용어 설명 출처 목록 (6)

용어 설명: PCR은 세균 특이성이 있는 primer를 이용하여 적은 수의 세균이 있을지라도 쉽게 검출할 수 있는 유용한 방법이며, 이를 이용하여 구강 내 치면세균막이나 타액에서 직접 세균을 검출할 수 있게 되었다[8].

내보내기 구분	파일저장 인쇄 메일전송
구성항목	기본정보 상세정보 관리번호, 논문명, 저널/프로시딩명, 저자 , 발행년, 권, 호, 시작페이지, 끝페이지, 발행기관 관리번호, 논문명, 대등논문명, 저자 , 저널/프로시딩명, 발행기관, 발행년, 발행언어, 권, 호, 시작페이지, 끝페이지, ISBN, ISSN, 주제분야, 키워드, 초록(한글), 초록(영문), 저자(소속기관)
저장형식	Text(ASCII format) Excel format RefWorks Direct Export RIS format (for Reference Manager, ProCite, EndNote), Scholar's Aids, Mendeley
메일정보	받는사람 (필수) @ 보내는사람 (선택) @ 제목 내용 KISTI 검색결과 이메일 서비스
안내	총 건의 자료가 검색되었습니다. 다운받으실 자료의 인덱스를 입력하세요. (1-10,000) 검색결과의 순서대로 최대 10,000건 까지 다운로드가 가능합니다. 데이타가 많을 경우 속도가 느려질 수 있습니다.(최대 2~3분 소요) 다운로드 파일은 UTF-8 형태로 저장됩니다. 파일의 내용이 제대로 보이지 않을실 때는 웹브라우저 상단의 보기 -> 인코딩 -> 자동선택 여부를 확인하십시오. ~ Text(ASCII format) Excel format

연합인증