The first goal of this thesis was to prove the possibility that those with ADHD could be diagnosed using sounds derived from their brainwaves. For this, I made the following specific goals. The first was to develop ADHD diagnosing algorithms utilizing EEG brainwaves through several mathematical methods, specially using eyes-open, resting-state brainwaves. The second was to develop a sonifcation algorithm to convert brainwaves to musical sounds, which is needed for a future brainwave stethoscope for diagnosis of ADHD patients. For this, after recording the brainwaves of 20 people (10: controls, 10: ADHD) during eyes-open resting-state, Fast Fourier Transform was applied to their brainwaves to get energy per frequency band, and per channel, of the brain. Finally, 11 among 248 features, which would be used for training classification function, were selected after applying an ANOVA test and Fishier discriminant analysis, and comparing the level of overlap. By training SVM with the selected 11 features based on the brainwaves of 20 people, SVM was given the ability to classify. Then, the distances derived from the SVM result were transformed to musical sounds while making the difference between the brainwaves of ADHD and control subjects bigger than before and arranging that they be expressed as a brainwave music form which enabled people to classify two classes with a high precision level. To validate this brain music, I played the musical sounds to 29 participants and let them distinguish the experimental class of the sounds and measure their certainty based on a 7th LIKERT scale. The experimental results were as follows: for SVM classification, 97. 667% precision was gained, and for the music classification test, a sensitivity of 97.9%, a specificity of 96.55%, a positive predictive rate of 96.6%, and a negative predictive rate of 97.9% were gained. These results imply that the brainwaves of those with and without ADHD can be distinguished using musical sounds derived from their brainwaves, and that a brainwave stethoscope can be utilized for ADHD patients in the future. Furthermore, the high SVM classification result proved that brainwaves during eyes-open resting-state between healthy people and ADHD patients also have such significantly meaningful differences that they can be utilized for making a future stethoscope. Moreover, after applying power spectrum analysis to the brainwaves during eyes-open resting-state; differences in the gamma, beta, and theta activities according to the location on the brain were found. These findings were not discovered in previous relevant studies in detail and the reason for those brainwave-activity distribution-differences, according to the location of the brain. has to be identified in future research. In conclusion, future applications of this research could result in doctors being able to diagnose ADHD by just listening to sounds converted from brainwaves without consuming much time, and SVM could be utilized to make brainwave music for this application and to make a high classification result.
본 논문은 주의력 결핍 과잉 행동 장애인을 위하여 EEG를 활용한 뇌파 청진기용 알고리즘 개발과 사운드 제작을 목적으로 한다. 기존 연구는 ADHD 환자 진단을 위해서 여러가지 수학적 메소드들이 개발 또는 활용되었고 특히 ERP 뇌파를 가지고 진단을 시도하였으나 본 연구는 처음으로, ERP 가 아닌 Eye-Open Resting 상태에서의 EEG 뇌파를 가지고 Feature 들을 선택한 후 SVM 머신 러닝을 통한 적절한 hyper plane 의 parameter 값을 찾은 후 그 것을 가지고 향후에 들어오는 환자를 진단 하기 위한 메소드를 제시하였다. ERP 가 아닌 Eye-Open 상태에서의 뇌파는 향후 real time 으로 쓰일 수 있는 뇌파 청진기의 알고리즘으로 보다 쉽게 활용될 수 있다. 또한 이 연구는 수학적인 분류 알고리즘을 제시할 뿐만 아니라 더 나아가 실질적으로 뇌파 청진기용 사운드 매핑 함수 알고리즘을 제시하고 직접 ADHD patients의 뇌파를 활용하여 듣기 좋은 사운드(음악) 을 생성함으로써 실질적인 뇌파 청진기의 가능성을 실험적으로 검증 및 제시하였다. 첫번째 방법인 수학적인 분류 메소드 뿐만 아니라 sonification 두 가지 방법 모두 실험적으로 96% 이상의 정확도를 보인다는 것을 실험적으로 검증하였으며 , 두번째 메소드인 sonifciation 입력 데이터는 첫번째 수학적 메소드의 출력값에 실질적으로 의존하기 때문에 첫번째 분류 메소드의 성능이 뇌파 청진기의 정확도에 있어서 가장 중요하다. 차후에 더 많은 피 실험자를 (환자 데이터를 ) 모집하게 된다면 보다 정확한 parameter를 찾을 수 있어 보다 실질적이고 정확한 분류기 즉 ADHD 환자용 뇌파 진단기로 이용 될 수 있을 것이다.
