ABSTRAKSI: Sistem pengenalan gelombang ECG telah banyak dikembangkan. Ada beberapa pendekatan yan digunakan yaitu waveform detection (algoritma Pan Tompskin dan metode threshold), template matching, pendekatan neural network, dan pendekatan model probabilitas.
Hidden Markov Model yang menggunakan pendekatan model probabilitas terbukti menghasilkan performansi yang bagus dalam speech recognition. Oleh karena itu, dalam tugas akhir ini akan diujikan metode HMM dalam kasus biomedis Electrocardiogram.
Prinsip kerja sistem pendeteksian penyakit jantung ini adalah dengan mengasumsikan ECG seperti speech recognition, terutama untuk kasus isolated word. Dalam sistem pendeteksian penyakit jantung terdapat 2 proses yaitu proses pemodelan dan pengenalan. Pada proses pemodelan akan dibuat suatu model labelisasi sinyal ECG dengan menghitung parameter-paramter HMM, yaitu probabilitas inisial state, probabilitas transisi antar state, dan probabilitas observasi pada suatu state. Setelah diperoleh ketiga nilai tersebut yang optimal, maka terbentuklah suatu model HMM untuk masing-masing penyakit jantung. Sedangkan Proses pengenalan penyakit jantung dilakukan pada tiap data ECG dengan menghitung likelihood dari data testing yang akan dikenali terhadap semua model data ECG yang telah di-training sebelumnya. Dengan labelisasi yang teliti dan penentuan nilai probabilitas observasi yang optimal, HMM dapat digunakan untuk mengenali penyakit jantung.
Hasil dari pengujian menunjukkan, nilai probabilitas observasinya didapat dari random uniform distribution menghasilkan akurasi yg selalu berubah/dinamis seiring dengan perubahan nilai random. Sedangkan jika nilai observasinya disamaratakan dengan jumlah state dan jumlah pengamatan, akurasinya 33,33%.
Kata Kunci : Hidden Markov Model, Electrocardiogram, probabilitas inisial state, probabilitas transisi antar state, distribusi probabilitas obsevasi.ABSTRACT: ECG signal recognition system has been developed. Several approach which used are waveform detection (Pan Tompskin algorithm and threshold method), template matching, neural network approach, and probability model approach.
Hidden Markov Model which used probability model apptoach has been proved can generate a good performance in speech recognition. So, This final project will tested HMM method in Electrocardiogram biomedic.
The priciple of heart disease detection system is by assuming ECG as a speech recognition, mainly on isolated word case. In the heart disease detection system, there are 2 process which are modeling and recognition process. In the modeling process, ECG signal labelling will be made by measuring HMM parameters, which are initial state probability, inter-state transition probability, and on-a-state observation probability. After the optimum outcome of the parameters are achieved, then a HMM model for each heart disease is formed. Meanwhile, the heart disease recognition is done in every ECG data by measuring the likelihood of tested data which will be recognized by all previously-trained ECGdata. By accurate labelling and optimum observation probability value, HMM can be used for heart disease recognition.
The analysis of the outcome shows that the observation probability which is obtained from random uniform distribution will have a dynamic value of accuracy along the alteration of random value.In the other hand, The accuracy, if the observation value is equally divided with the number of state and observation, is 33,33%
Keyword: Hidden Markov Model, electrocardiogram, intial state probability, state transition probability, observation probability distribution