Deteksi Kejang Epilepsi Berdasarkan Sinyal Electroencephalogram Menggunakan Graph Transformer Dengan Transformasi Wavelet

Makarim, Muhammad Faishal (2026) Deteksi Kejang Epilepsi Berdasarkan Sinyal Electroencephalogram Menggunakan Graph Transformer Dengan Transformasi Wavelet. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002221005-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (22MB) | Request a copy

Abstract

Kejang epilepsi dapat diperiksa melalui sinyal electroencephalogram (EEG) dan mem pengaruhi sekitar 50 juta orang di seluruh dunia. Namun, analisis konvensional memer lukan waktu yang lama serta rentan terhadap kesalahan manusia. Sinyal EEG juga memi liki keterbatasan berupa resolusi spasial yang rendah. Penelitian ini menggunakan model Graph Transformer dengan transformasi wavelet untuk mendeteksi kejang epilepsi berda sarkan sinyal EEG. Discrete Wavelet Transform (DWT) dengan lter Daubechies 4 (Db4) digunakan untuk mendekomposisi sinyal EEG ke dalam beberapa sub-band yang merep resentasikan karakteristik sinyal pada resolusi yang berbeda. Selanjutnya, setiap koe sien wavelet dalam domain waktu-frekuensi dikarakterisasi menggunakan parameter statistik untuk mendapatkan informasi penting terkait kejadian kejang. Parameter statistik terse but digunakan sebagai node feature pada setiap kanal EEG, sedangkan hubungan spasial antar kanal dihitung menggunakan Normalized Cross-Correlation (NCC) dalam struktur graf. Representasi graf ini kemudian diproses oleh Graph Transformer untuk menangkap informasi spasiotemporal antar kanal EEG. Model diuji menggunakan dataset Children’s Hospital BostonMassachusetts Institute of Technology (CHB-MIT) yang berisi rekaman EEG pasien epilepsi. Evaluasi dilakukan dengan skema Leave-One-Patient-Out (LOPO) untuk mengukur kemampuan generalisasi model dalam mendeteksi kejadian kejang pada pasien yang tidak terlibat selama proses pelatihan. Hasil penelitian menunjukkan bah wa model Graph Transformer dengan transformasi wavelet mampu mendeteksi kejang epilepsi dan waktu kejadiannya pada setiap pasien. Analisis Explainable AI (XAI) meng gunakan Integrated Gradients (IG) mengidenti kasi tur dominan dan daerah kanal yang paling berpengaruh selama kejadian kejang sehingga dapat memberikan informasi yang mendukung diagnosis klinis.
======================================================================================================================================
Epileptic seizures can be examined using electroencephalogram (EEG) signals and a ect approximately 50 million people worldwide. However, conventional EEG analysis is time-consuming and prone to human error. In addition, EEG signals have the limitation of low spatial resolution. This study proposes a Graph Transformer model combined with wavelet transformation for epileptic seizure detection based on EEG signals. The Discre te Wavelet Transform (DWT) with the Daubechies 4 (Db4) wavelet lter was employed to decompose EEG signals into multiple sub-bands representing signal characteristics at di erent resolutions. Subsequently, each wavelet coe cient in the time-frequency domain was characterized using statistical parameters to extract important information related to seizure events. These statistical parameters were used as node features for each EEG channel, while the spatial relationships between channels were computed using Normali zed Cross-Correlation (NCC) to construct a graph representation. The resulting graph was then processed by the Graph Transformer to capture spatiotemporal dependencies among EEG channels. The model was evaluated using the Children’s Hospital Boston Massachusetts Institute of Technology (CHB-MIT) dataset, which contains EEG reco rdings from epilepsy patients. Evaluation was conducted using the Leave-One-Patient Out (LOPO)scheme to assess the model’s generalization ability in detecting seizure events from previously unseen patients. The results demonstrate that the proposed Graph Tran sformer with wavelet transformation is capable of detecting epileptic seizures and their occurrence time for each patient. Furthermore, Explainable Arti cial Intelligence (XAI) analysis using Integrated Gradients (IG) identi ed the dominant features and EEG chan nel regions that contributed most signi cantly during seizure events, thereby providing information that can support clinical diagnosis.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Electroencephalogram, Deteksi Kejang Epilepsi, Graph Transformer, Transformasi Wavelet, Electroencephalogram, Epileptic Seizure Detection, Graph Transformer, Wavelet Transform.
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA403.3 Wavelets (Mathematics) R Medicine > R Medicine (General) > R858 Deep Learning R Medicine > RC Internal medicine > RC386.5 Electroencephalography. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Faishal Makarim
Date Deposited:	03 Aug 2026 07:25
Last Modified:	03 Aug 2026 07:25
URI:	http://repository.its.ac.id/id/eprint/142784

Actions (login required)

View Item