Comparison of window shapes and lengths in short-time feature extraction for classification of heart sound signals

2026-04-15 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Medical Devices & Health Technology · Depth: Intermediate, quick

Summary

A study evaluated the impact of window shapes and lengths on feature extraction for classifying phonocardiography (PCG) signals using a bidirectional long short-term memory (biLSTM) network. The research focused on three window shapes (Gaussian, triangular, rectangular) and three lengths for each, extracting statistical features from short signal segments. The biLSTM network was trained and tested on these features to diagnose cardiovascular pathology. Results indicated that the Gaussian window consistently yielded the best classification performance, with the triangular window performing comparably at a 75 ms length. The commonly used rectangular window was identified as the least effective option. A 75 ms Gaussian window specifically outperformed a baseline method.

Key takeaway

For Machine Learning Engineers developing heart sound classification systems, selecting the appropriate windowing technique is critical. You should prioritize using a Gaussian window for feature extraction from phonocardiography (PCG) signals, particularly exploring a 75 ms length, as it demonstrated superior performance over other shapes and a baseline. Avoid the rectangular window, despite its common availability, to prevent feature distortion and suboptimal classification.

Key insights

Gaussian windows and optimal lengths significantly improve heart sound classification with biLSTM networks.

Principles

• Window shape impacts spectral feature quality.
• Optimal window length enhances classification.
• Rectangular windows are suboptimal for PCG.

Method

Statistical features are extracted from PCG signals using sliding windows of varying shapes and lengths, then fed into a biLSTM network for classification performance evaluation.

In practice

• Prioritize Gaussian windows for PCG feature extraction.
• Test 75 ms window lengths for PCG signals.
• Avoid rectangular windows in PCG analysis.

Topics

• Heart Sound Classification
• Phonocardiography Signals
• biLSTM Networks
• Feature Extraction
• Windowing Functions

Best for: AI Engineer, Machine Learning Engineer, AI Scientist, Research Scientist

Related on AIssential

• Comparison of window shapes and lengths in short-time feature extraction for classification of heart sound signals — Optimal window shape and length significantly enhance heart sound signal classification performance.
• Phonetic Error Analysis of Raw Waveform Acoustic Models — Raw waveform models with BLSTMs and transfer learning improve phone recognition, with errors reflecting inherent phonetic similarities.
• Shape and amplitude decoupling in pulsatile physiological signal synthesis and its evaluation — Pulsatile physiological signals, crucial for understanding cardiovascular dynamics, are often generated by methods that undesirably couple waveform shape and amplitude, hindering precise control.
• Lung-SRAD: Spectral-Aware Regularized Audio DASS with Dual-Axis Patch-Mix Contrastive Learning for Respiratory Sound Classification — Lung-SRAD uses State Space Models and spectral-aware regularization to improve respiratory sound classification by preserving high-frequency details.
• Live Human Detector on Outbound Phone Calls [R] — A machine learning audio classifier can detect live human connections on outbound calls by analyzing acoustic features.
• MURMUR: An Efficient Inference System for Long-Form ASR — Murmur efficiently balances long-form ASR accuracy and latency using a two-level inference system.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.