Deteksi Obstructive Sleep Apnea Menggunakan Recurrence Plot dan Convolutional Neural Network dari Citra Termografi Inframerah

Pratapa, Muhammad Daffa Kurnia (2025) Deteksi Obstructive Sleep Apnea Menggunakan Recurrence Plot dan Convolutional Neural Network dari Citra Termografi Inframerah. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023211038-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Obstructive Sleep Apnea (OSA) adalah gangguan serius saat seseorang mengalami berhentinya pernapasan selama tidur setidaknya selama 10 detik yang mana terjadi secara berulang. OSA disebabkan oleh otot-otot pada saluran pernapasan yang menjadi rileks saat tidur sehingga menyebabkan penghambatan aliran udara pada paru-paru. Jika tidak segera diatasi, OSA dapat menimbulkan berbagai komplikasi fatal serta dapat menurunkan kualitas hidup secara signifikan dan meningkatkan risiko kematian pada pasien. Untuk mengatasi tantangan ini, perlu dikembangkan metode deteksi OSA yang non-invasif dan efisien. Pada penelitian ini, sebuah pendekatan pengunaaan kamera infrared thermography (IRT) untuk mendeteksi sinyal pernapasan dilakukan. Rata-rata laju pernapasan yang terdeteksi dari sinyal thermal menunjukkan mean absolute error (MAE) sebesar 0,96 brpm dengan nilai CAND 93,97 % dibanding ground truth respiratory belt Go-Direct. Di sisi lain, deteksi detak jantung melalui perubahan suhu wajah menghasilkan MAE 5,007 bpm dan CAND 92,62 % dari pengukuran sepuluh subjek, selanjutnya data sinyal jantung ini diubah menjadi sinyal HRV setiap menit. Sinyal pernapasan dan HRV ini kemudian ditransformasi menjadi recurrence plot (RP) sebagai input model classifier CNN ResNet. Pengujian pada tiga skenario input menunukkan bahwa penggunaan RP sinyal pernapasan dengan ResNet-50, optimizer SGD, dan learning rate 0,01 memberikan kinerja terbaik, mencapai accuracy 98,02 % dan F1-score 96,87 %. Sebaliknya, RP sinyal HRV hanya meraih accuracy 59,02 % dan F1-score 41,13 %, karena hasilnya banyak dipengaruhi oleh amplifikasi noise. Model dual-input memang memberikan performa yang hampir serupa dengan input RP pernapasan dengan accuracy 99,01 % dan F1-score 98,40 %, akan tetapi kurang efisien karena memerlukan dua proses terpisah untuk mengekstraksi sinyal pernapasan dan HRV. Dengan demikian, skenario input RP sinyal pernapasan merupakan konfigurasi optimal dalam penelitian ini.

=======================================================================================================================================

Obstructive Sleep Apnea (OSA) is a serious disorder characterized by repeated cessation of breathing during sleep for at least 10 seconds. It occurs when the muscles of the upper airway relax during sleep, causing obstruction of airflow to the lungs. If left unaddressed, OSA may lead to fatal complications, a significant decline in quality of life, and increased mortality risk. To meet this challenge, non-invasive and efficient detection methods are essential. In this study, an infrared thermography (IRT) approach was employed to capture respiratory signals. The average respiratory rate derived from the thermal signal exhibited a mean absolute error (MAE) of 0.96 bpm and a coverage probability (CAND) of 93.97 % when compared to the ground-truth respiratory belt Go-Direct. Heart rate detection based on facial temperature variations resulted in an MAE of 5.007 bpm and a CAND of 92.62 % across ten subjects; these measurements were subsequently converted into minute-by-minute heart rate variability (HRV) signals. Both respiratory and HRV time series were transformed into recurrence plots (RP) for input into a ResNet-based convolutional neural network classifier. Evaluation under three input scenarios revealed that RP of respiratory signals, processed with ResNet-50, stochastic gradient descent (SGD) optimizer, and a learning rate of 0.01, achieved the highest performance, with an accuracy of 98.02 % and an F1-score of 96.87 %. In contrast, RP of HRV signals attained only 59.02 % accuracy and a 41.13 % F1-score due to amplified noise. Although the dual-input model delivered comparable results with 99.01 % accuracy and a 98.40 % F1-score, it required separate extraction processes for respiratory and HRV signals, which made it less efficient. Consequently, the RP respiratory-signal input scenario represents the optimal configuration in this research.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Pencitraan termal, Obstructive Sleep Apnea (OSA), Laju Pernapasan, Recurerrence Plot (RP), Convolutional Neural Network (CNN), Thermal Imaging, Respiration Rate
Subjects:	R Medicine > R Medicine (General) > R858 Deep Learning R Medicine > RC Internal medicine > RC78 Diagnosis, Radioscopic--Examinations, questions, etc. T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Muhammad Daffa Kurnia Pratapa
Date Deposited:	04 Aug 2025 02:35
Last Modified:	04 Aug 2025 02:35
URI:	http://repository.its.ac.id/id/eprint/125418

Actions (login required)

View Item