Analisis Performa CNN-PatchTST untuk Klasifikasi Aritmia melalui Sinyal EKG

Q: What is the clinical evidence from this study?

Population: Arrhythmia (n=29858). Intervention: Hybrid CNN-PatchTST model vs. Standalone CNN model. Primary outcome: Classification accuracy.

Q: What does this research mean for the field?

A hybrid CNN-PatchTST model achieves 88.8% accuracy and an 82.5% macro F1-score in classifying five types of cardiac arrhythmia from ECG signals, improving upon standalone CNN models. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Humam Maulana Tsubasanofa Ramadhan; Achmad Junaidi; Muhammad Muharrom Al Haromainy

doi:10.33005/santika.v6i1.1250

June 30, 2026Prosiding Seminar Nasional Informatika Bela NegaraOpen Access

Analisis Performa CNN-PatchTST untuk Klasifikasi Aritmia melalui Sinyal EKG

View Full Paper

Ask AI

Bookmark

Share

Key result

Hybrid CNN-PatchTST model improves ECG arrhythmia classification accuracy by ~3 percentage points over standalone CNN.

n=29,858

Why the study?

Manual detection of arrhythmia from ECG signals is time-consuming and prone to inter-observer errors, motivating development of automated classification methods.

Key Points

To develop and evaluate a hybrid deep learning model combining a one-dimensional convolutional neural network with PatchTST to classify five cardiac rhythm categories from electrocardiogram signals.
Applied preprocessing pipelines comprising signal standardization, noise filtering, Z-score normalization, and hybrid sampling to correct class imbalances.
Combined a 1D CNN for local morphological feature extraction with a Transformer-based PatchTST for long-range temporal dependencies across three data split ratios.
Attained the highest class-specific F1-scores in Atrial Fibrillation (91%) and Normal Sinus Rhythm (93%).
Demonstrated lower classification performance for First-degree AV Block and Sinus Bradycardia due to morphological waveform similarities with Normal Sinus Rhythm.

Structured PICO

P

Population

29,858 ECG signal records from six public Kaggle datasets used to evaluate a hybrid CNN-PatchTST model for classifying five types of arrhythmias.

I

Intervention

Hybrid 1D Convolutional Neural Network (CNN) and Patch Time Series Transformer (PatchTST) architecture for ECG signal classification

C

Comparator

Standalone 1D Convolutional Neural Network (CNN) model

O

Outcome

Model classification performance measured by accuracy, macro sensitivity, macro specificity, and macro F1-score

Main Result

Absolute Event Rate: 88.8% vs 85.8%

Limitations

Lower performance in classifying First-degree AV Block (IAVB) and Sinus Bradycardia (SB) due to signal morphological similarity with Normal Sinus Rhythm.
Limitations of synthetic data generated by hybrid sampling in fully representing variations between classes.
The model exhibited a tendency to misclassify other arrhythmias as Normal Sinus Rhythm.
Morphological overlap between classes (e.g., First-degree AV Block and Sinus Bradycardia heavily resemble Normal Sinus Rhythm), leading to misclassification
Limitations of synthetic data generated via hybrid sampling in fully representing true class variations

Implication

A hybrid deep learning model combining 1D CNNs and PatchTST effectively captures both local morphology and long-term temporal dependencies in ECG signals, achieving high accuracy in automated arrhythmia classification.

Abstract

Penyakit kardiovaskular, khususnya aritmia, merupakan salah satu penyebab kematian tertinggi di dunia. Deteksi aritmia secara manual melalui interpretasi sinyal elektrokardiogram (EKG) membutuhkan waktu lama dan rentan terhadap kesalahan antar-pengamat. Penelitian ini mengusulkan model hybrid CNN-PatchTST untuk mengklasifikasikan lima jenis aritmia, yaitu Atrial Fibrillation (AF), First-degree AV Block (IAVB), Normal Sinus Rhythm (NSR), Sinus Bradycardia (SB), dan Sinus Tachycardia (STach) dari sinyal EKG. Model ini menggabungkan CNN satu dimensi sebagai ekstraktor fitur morfologi lokal dengan PatchTST berbasis Transformer untuk menangkap dependensi temporal jangka panjang. Dataset yang digunakan merupakan gabungan enam sumber publik dari Kaggle dengan total 29.858 rekaman sinyal EKG pada frekuensi 500 Hz. Preprocessing mencakup standarisasi sinyal, noise filtering, normalisasi Z-score, dan hybrid sampling untuk mengatasi ketidakseimbangan kelas. Pengujian dilakukan pada tiga skenario rasio pembagian data yang berbeda. Hasil terbaik diperoleh pada rasio 70:20:10 dengan akurasi 88,8%, sensitivitas makro 84,3%, spesifisitas makro 96,3%, dan F1-score makro 82,5%. Kelas AF dan NSR mencatat performa terbaik dengan F1-score masing-masing 91% dan 93%, sementara kelas IAVB dan SB menunjukkan performa lebih rendah akibat kemiripan morfologi sinyal dengan NSR.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Ramadhan et al. (2026) studied Arrhythmia (n=29,858). Hybrid CNN-PatchTST model vs. Standalone CNN model was evaluated on Classification accuracy. The hybrid CNN-PatchTST model achieved 88.8% accuracy and 82.5% macro F1-score in classifying five arrhythmia types from ECG signals, outperforming a standalone CNN model's 85.8% accuracy.

synapsesocial.com/papers/6ad0fc263afa093ea9acfc7c https://doi.org/10.33005/santika.v6i1.1250

View Full Paper

Ask AI

Bookmark

Share

Analisis Performa CNN-PatchTST untuk Klasifikasi Aritmia melalui Sinyal EKG

Key Points

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Analisis Performa CNN-PatchTST untuk Klasifikasi Aritmia melalui Sinyal EKG

Key Points

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

More related papers

More from Prosiding Seminar Nasional Informatika Bela Negara

Keep reading

More related papers

More from Prosiding Seminar Nasional Informatika Bela Negara