Apsari, Francisca Cindy Meilia (2025) Sistem Prediksi Kejang Epileptik Menggunakan Radar Continuous Wave Berbasis Multivariate Statistical Process Control. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5023211021-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5023211021-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy |
Abstract
Penelitian ini mengembangkan sistem prediksi kejang epileptik berbasis radar non-kontak Continuous Wave (CW) menggunakan metode Multivariate Statistical Process Control (MSPC). Sistem ini dirancang untuk memantau sinyal detak jantung dari pergerakan mikro dinding dada menggunakan sensor radar RFbeam K-LC7. Untuk mengekstraksi sinyal detak jantung secara akurat, digunakan metode Combined Instantaneous Power yang lebih robust terhadap gangguan I/Q imbalance dibandingkan teknik demodulasi konvensional. Sinyal I dan Q dari radar diproses melalui filter bandpass, dikombinasikan menjadi instantaneous power, dan distabilkan dengan Automatic Gain Control. Deteksi detak dilakukan menggunakan zero-crossing yang adaptif terhadap estimasi frekuensi jantung dari analisis autocorrelation FFT dan filter bank. Hasil validasi sinyal fisiologis menunjukkan sistem mampu mengekstraksi Beat-to-Beat Interval (BBI) dengan rata-rata Root Mean Square Error (RMSE) sebesar 67,22 ms terhadap data ECG. Fitur HRV seperti SDNN, RMSSD, dan rasio LF/HF menunjukkan korelasi moderat-kuat dengan ECG (r = 0,663; 0,701; 0,859). Selanjutnya, fitur HRV yang diekstraksi dari sinyal ECG pada dataset publik SeizeIT2 digunakan sebagai input untuk memvalidasi software MSPC dalam mendeteksi fase preictal. Model dibangun dengan PCA dan diuji menggunakan pendekatan hybrid global projection dan personalized control limit. Hasil prediksi menunjukkan nilai presisi sebesar 93,9%, sensitivitas 71,1%, spesifisitas 89,4%, dan F1-Score sebesar 80,9%. Rata-rata false alarm rate tercatat sebesar 1,27 FP/h, kompetitif dengan literatur sebelumnya. Sistem ini juga mampu memberikan peringatan rata-rata 11,4 menit sebelum kejang, dengan prediksi paling awal tercatat 14,6 menit sebelumnya. Kesimpulan dari penelitian ini menunjukkan bahwa metode instantaneous power efektif digunakan untuk memperoleh HRV yang baik dan stabil, sedangkan model MSPC mampu digunakan untuk memprediksi kejang berdasarkan analisis fitur HRV dengan akurasi yang memadai.
==================================================================================================================================================================
This study develops a non-contact epileptic seizure prediction system based on Continuous Wave (CW) radar using the Multivariate Statistical Process Control (MSPC) method. The system is designed to monitor heart rate signals from the micro-movements of the chest wall using the RFbeam K-LC7 radar sensor. To accurately extract heart rate signals, the Combined Instantaneous Power method is used, which is more robust against I/Q imbalance interference compared to conventional demodulation techniques. The radar’s I and Q signals are processed through a bandpass filter, combined into instantaneous power, and stabilized with Automatic Gain Control. Heartbeat detection is performed using zero-crossing, which is adaptive to the heart rate frequency estimation from autocorrelation FFT analysis and a filter bank. Physiological signal validation shows that the system is able to extract the Beat-to-Beat Interval (BBI) with an average Root Mean Square Error (RMSE) of 67.22 ms compared to ECG data. HRV features such as SDNN, RMSSD, and the LF/HF ratio show moderate-to-strong correlations with ECG (r = 0.663; 0.701; 0.859). Furthermore, the HRV features extracted from the ECG signals in the public SeizeIT2 dataset are used as input to validate the MSPC software in detecting the preictal phase. The model is built using PCA and tested using a hybrid global projection and personalized control limit approach. The prediction results show a precision of 93.9%, sensitivity of 71.1%, specificity of 89.4%, and an F1-Score of 80.9%. The average false alarm rate is recorded at 1.27 FP/h, competitive with previous literature. The system was also able to provide warnings on average 11.4 minutes before a seizure, with the earliest prediction occurring 14.6 minutes prior. The conclusion of this study indicates that the instantaneous power method is effective for obtaining stable and accurate HRV, while the MSPC model can be used to predict seizures based on HRV feature analysis with adequate accuracy.
Actions (login required)
 |
View Item |