Gated Recurrent Units dalam Mendeteksi Obstructive Sleep Apnea

Abstract

AbstrakDalam melakukan penelitian obstructive sleep apnea (OSA), polysomnography (PSG) digunakan untuk diagnosis. Namun subjek diharuskan menginap dilaboratorium selama beberapa malam untuk melakukan tes dengan PSG dan karena banyaknya alat yang harus dikenakan pada tubuh dapat membuat tidak nyaman saat pengambilan data. Belakangan ini, beberapa peneliti mengunakan single-lead ECG untuk melakukan deteksi OSA. Untuk menghasilkan model terbaik, akan dilakukan eksperimen training, dengan batch normalization dan dropout yang berbeda. Pada penelitian ini apnea-ecg dataset digunakan, RR-Interval dan amplitudo QRS complex dari released set berjumlah 35 data akan disegmentasi permenit untuk digunakan sebagai input dari arsitektur yang diajukan adalah gated recurrent unit (GRU). Lalu withheld set berjumlah 35 data akan digunakan untuk pengujian per-segment dan per-recording. Kinerja sistem diukur berdasarkan accuracy, sensitifity, dan specificity dengan pengujian per-segment mendapat hasil accuracy 83.92%, sensitifity 81.28%, dan specificity 85.55%, dan pengujian per-recording mendapat hasil accuracy 97.14%, sensitifity 95.65% dan specificity 100%.Kata kunci: Obstructive sleep apnea, GRU, ECG, RR-Interval, QRS complex.AbstractIn conducting obstructive sleep apnea (OSA) studies, polysomnography (PSG) was used for the diagnosis. However, the subject was required to stay in the laboratory for several nights to carry out tests with the PSG and because of the many devices that had to be worn on the body, it could be uncomfortable to collect data. Recently, several researchers have used single-lead ECG to detect OSA. To produce the best model, training experiments will be conducted, with different batch normalization and dropout. In this study, the apnea-ecg dataset is used, the RR-Interval and the QRS complex amplitude from the released set totaling 35 data will be segmented per minute to be used as input for the proposed architecture is the gated recurrent unit (GRU). Then the withheld set of 35 data will be used for per-segment and per-recording testing. System performance was measured based on accuracy, sensitivity, and specificity with per-segment testing getting 83.92% accuracy, 81.28% sensitivity, and 85.55% specificity, and per-recording testing got 97.14% accuracy, 95.65% sensitivity and 100% specificity.Keywords: Obstructive sleep apnea, GRU, ECG, RR-Interval, QRS complex.