신경망을 이용한 구순열로 인한 코변형의 정량적 평가

Abstract

Our study aimed at quantitative assessment of a cleft palate nose deformity condition by analyzing the following parameters gathered from a photographic image of a cleft palate patient: (1) angle difference between two nostril axes, (2) center of the nostril and distance between two centers, (3) overlapped area of two nostrils, and (4) the overlapped area ratio of the two nostrils. A regression equation of doctor"s grades was obtained using the eight parameters. Three plastic surgeons gave us the grades for the each photographic image by 10 increments with maximum grade of 100. The average reproducibility of the grades given by the three plastic surgeons and the three laymen using the developed program was 10.8±4.6% and 7.4±1.8%, respectively. Kappa values representing the degree of consensus of the plastic surgeons and the three laymen were 0.43 and 0.83, respectively. Correlation coefficient of the grades evaluated by the surgeons and obtained by the regression equation was 0.642 and that of the grades by the surgeons and by the neural network was 0.798. In conclusion, the developed neural network model provided us better reproducibility, much better consensus, and better correlation than doctor"s subjective evaluation in addition to objectiveness and easy application.