×
ABSTRAK
Penelitian untuk deteksi aritmia pada elektrokardiogram dengan metode Jaringan Syaraf Tiruan (JST) kelas jamak menggunakan fitur interval RR, interval QRS, serta gradien gelombang R telah berhasil dilaksanakan. Sistem deteksi tersebut diimplementasikan pada perangkat lunak MATLAB. Tipe aritmia yang dideteksi dalam penelitian adalah Premature Ventricular Contraction (PVC), Premature Atrial Contraction (PAC), dan Left Bundle Branch Block (LBBB). Tahapan pada penelitian ini antara lain pengumpulan data, persiapan perangkat lunak, pra-proses, ekstraksi fitur, pelatihan JST, pengujian JST, penentuan kinerja serta perancangan antar muka Graphycal User Interface (GUI). Tahap pelatihan dilakukan dengan menggunakan data sebanyak 6% sedangkan tahap pengujian sebanyak 94% dari total keseluruhan data. JST yang digunakan pada penelitian ini antara lain JST RBF, MLP-BP dan LVQ. Rancangan sistem kemudian dibuat dalam tampilan GUI untuk mempermudah tampilan antarmukanya. Pada penelitian ini dilakukan variasi jumlah fitur sebagai masukan JST, yaitu tiga macam fitur dan dua macam fitur. Hasil terbaik ditunjukkan pada variasi gabungan tiga macam fitur (interval RR, lebar QRS, gradien gelombang R) menggunakan JST RBF dengan kinerja berupa sensitivitas, spesifisitas, serta akurasi cukup baik yaitu 96,08%, 92,37% dan 93,34%.
Kata kunci : gradien gelombang R, interval RR,lebar QRS, PVC, PAC, LBBB, JST kelas jamak
ABSTRACT
Research for arrhythmias detection using multiclass Artificial Neural Network (ANN) method has been successfully implemented. It utilized RR interval, QRS width, and R wavegradient features. Arrhythmia types used in this study were Premature Ventricular Contraction (PVC), Premature Atrial Contraction (PAC), and Left Bundle Branch Block (LBBB). The stages in this study include data collection, software preparation, pre-processing, feature extraction, ANN training, ANN testing, determination of the performance as well as the GUI layout. The training phase was done by using 6% of the data while the testing stage using 94% of the data.The neural network used in this study were RBF NN, MLP-BP NN and LVQ NN. The detection system was made in GUI layout for the user friendly interface. This study was conducted by varying features number as the input of ANN. The variation includes two and three kinds of features. The best results were found when three features (R wave gradient, RR interval, QRS width) using RBF NN were included. The best performance were 96,08%, 92,37% and 93,34% in terms of sensitivity, sprcificity and accuracy, respectively.
Keywords: R wave gradient, RR interval, QRS width, PVC, PAC, LBBB, multiclass neural network