Publications

Detailed Information

Developing a universally applicable prediction model for soft tissue changes after orthognathic surgery based on the sparse partial least squares method : 광범위 적용 가능한 sparse partial least squares 기반 턱교정 수술 후 연조직 예측 알고리즘의 개발

DC Field Value Language
dc.contributor.advisor이신재-
dc.contributor.author서희연-
dc.date.accessioned2017-07-14T05:42:28Z-
dc.date.available2017-07-14T05:42:28Z-
dc.date.issued2015-02-
dc.identifier.other000000025428-
dc.identifier.urihttps://hdl.handle.net/10371/125065-
dc.description학위논문 (박사)-- 서울대학교 대학원 : 치의과학과, 2015. 2. 이신재.-
dc.description.abstract연구 목적: 본 연구의 목적은1) 턱교정수술의 종류에 관계없이 정확하게 수술 후 연조직 변화를 예측할 수 있는 예측 모형을 개발하고, 2) sparse partial least squares (SPLS) 방법을 통해 경제적이고 해석이 보다 용이한 예측 모형을 수립하는 것이다.
재료 및 방법: 본 연구는 서울대학교치과병원에서 교정 및 턱교정수술을 받은 318명의 골격성 II급 및 골격성 III 급 부정교합자를 대상으로 하였다. 총232개의 독립변수와 64개의 종속변수(32개의 술 후 연조직 계측점 좌표)로 예측모형을 수립하였다. PLS 모형과 SPLS 모형의 예측 정확도를 비교하였다. 예측 정확도 비교를 위하여 leave-one-out cross-validation을 통해 test error를 절대 값으로 계산하였다. Sparsity를 부과하는 SPLS 방법을 사용하여 경제적이고 해석이 보다 용이한 예측 모형을 만들고자 하였다. Leave-one-out cross-validation으로 최적의 SPLS 모형을 선택하였다.
결과: SPLS 방법에 기반한 예측 모형의 예측 정확도는 PLS 기반 모형의 예측 정확도와 유사하였다. 수립된 예측 모형들은 수술의 방향이나 종류, 성별에 관계없이 높은 예측 정확도를 보였다. 이는 본 연구에서 수립된 예측 모형을 수술의 종류에 관계없이 광범위하게 적용할 수 있음을 의미한다. SPLS 기반 예측모형에서 sparsity 를 부과하여 평균적으로 선택되는 독립변수가 34개로 감소하여도 모든 변수를 투입한 PLS 모형에 비하여 예측 정확도에 있어 임상적으로 유의한 차이를 보이지 않았다.
결론: PLS 와 SPLS 방법 모두 수술의 종류에 관계없이 높은 예측 정확도를 보였으며 SPLS 방법을 통해 정확하면서도 보다 해석이 용이하고 경제적인 턱교정수술 환자의 연조직 예측 모형을 수립할 수 있었다.
-
dc.description.abstractIntroduction: The aims of the present study are 1) to develop a model for an accurate soft tissue prediction that can be applied to various modes of surgical correction: Class II correction, Class III correction, the mandibular surgery and/or the maxillary surgery, and additional genioplasty-
dc.description.abstract2) to minimize the number of variables incorporated via SPLS method in order to build a parsimonious and interpretable prediction model.
Materials and methods: The subjects of this study consisted of 318 patients who had undergone the combined surgical-orthodontic correction of skeletal Class II or skeletal Class III malocclusions. The prediction model was composed of 232 independent and 64 dependent variables. Two prediction methods, the PLS method and the SPLS method were compared. In this study, to evaluate the predictive performance, test errors were calculated in absolute values through the leave-one-out cross-validation method. We promoted sparsity by SPLS method to build a parsimonious and more interpretable prediction model. Leave-one-out cross-validation method was used to determine the optimal SPLS model, i.e., a model selection.
Results: Prediction models by the SPLS method showed prediction performance comparable to prediction model by the PLS method. Since there were no significant differences in prediction performance depending on surgical movement or gender of the subjects, PLS and SPLS models built in present study are universally applicable. The SPLS showed no significant difference in prediction performance until the mean number of selected independent variables was reduced to 34.
Conclusions: It was our observation that both PLS and SPLS methods resulted in accurate prediction of soft tissue change associated with various types of surgical intervention. This study showed that it was possible to build a parsimonious and interpretable prediction model by the SPLS method. The SPLS method was found to be a powerful and objective tool for simplification. Based on our findings, we propose that the SPLS method can provide an improved algorithm in predicting surgical outcomes after orthognathic surgeries.
-
dc.description.tableofcontentsAbstract
Contents

Ⅰ. INTRODUCTION 1
Ⅱ. REVIEW OF LITERATURE 6
1. Notations, acronyms, and terms 6
2. Limitation of the conventional regression model 6
3. Feasible approaches to solve multi-collinearity problem and large p small n situation - Multivariate regression methods using latent variables 8
4. PLS to handle multi-collinearity problem and large p small n situation 8
4.1 Characteristics of PLS in comparison with other methods 8
4.2 Application of PLS method in soft tissue prediction after surgery - Previous studies 10
5. PLS and Sparse PLS 12
6. Cross-validation for model assessment and model selection 14
6.1 Purposes 14
6.2 Definition 16
6.3 Various cross-validation methods 16
6.4 Measure of prediction performance and model selection in the PLS method 17
Ⅲ. MATERIALS AND METHODS 19
1. Subjects 19
2. Cephalometric analysis 19
3. Variables in predictor and response matrices 20
4. Two multivariate methods to make prediction equations 21
5. Comparing the prediction performance between PLS method and SPLS method 22
6. Model selection for SPLS method 22
Ⅳ. RESULTS 24
1. Detailed description of study subjects 24
2. Building a universally applicable prediction model 24
3. Development of an optimal SPLS model for an accurate soft tissue prediction 26
Ⅴ. DISCUSSION 27
1. Building a universally applicable prediction model 27
2. Development of an optimal SPLS model for an accurate soft tissue prediction 30
VI. CONCLUSIONS 33

REFERE NCES 34
TABLES 41
FIGURES 46
APPENDIX TABLE 73
-
dc.formatapplication/pdf-
dc.format.extent3665897 bytes-
dc.format.mediumapplication/pdf-
dc.language.isoen-
dc.publisher서울대학교 대학원-
dc.subjectClass II malocclusion-
dc.subjectClass III malocclusion-
dc.subjectorthognathic surgery-
dc.subjectprofile prediction-
dc.subjectpartial least squares method-
dc.subjectsparse partial least squares method-
dc.subject.ddc617-
dc.titleDeveloping a universally applicable prediction model for soft tissue changes after orthognathic surgery based on the sparse partial least squares method-
dc.title.alternative광범위 적용 가능한 sparse partial least squares 기반 턱교정 수술 후 연조직 예측 알고리즘의 개발-
dc.typeThesis-
dc.contributor.AlternativeAuthorHee-Yeon Suh-
dc.description.degreeDoctor-
dc.citation.pages73-
dc.contributor.affiliation치의학대학원 치의과학과-
dc.date.awarded2015-02-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share