Brief Review — LHSNN: Design and Application of a Laconic Heart Sound Neural Network


Sik-Ho Tsang
2 min readDec 30, 2023

Design and Application of a Laconic Heart Sound Neural Network
, by Nanjing University of Posts and Telecommunications
2019 IEEE Access (Sik-Ho Tsang @ Medium)

Heart Sound Classification
20132023 [2LSTM+3FC, 3CONV+2FC] [NRC-Net] [Log-MelSpectrum+Modified VGGNet] [CNN+BiGRU]
==== My Other Paper Readings Are Also Over Here ====

  • Laconic Heart Sound Neural Network (LHSNN) is proposed where spectrogram is extracted and used as input of the proposed Laconic Neural Network (LNN).


  1. Laconic Heart Sound Neural Network (LHSNN)
  2. Results

1. Laconic Heart Sound Neural Network (LHSNN)

1.1. Spectrogram

  • A normal heart sound signal has a typical quasi-periodicity, sparsity, and energy concentration.
  • The spectrogram represents the relationship of the audio signal over time-spectral-energy.

The heart sound is transformed to spectrogram based on [15].

1.2. Laconic Neural Network (LNN)

Laconic Neural Network (LNN)

In brief, 2 convolutional layers with max pooling are used.

Then, 2 fully connected layers are used.

2. Results

Different kernel sizes and different number of convolutional layers

With the two convolutional layers, the training time is acceptable and the recognition rate is the highest.

SOTA Comparisons (I)

Compared with the recognition results of the cardiologists in [19] and the traditional LeNet-5 model, the LeNet-5 improved [20], and the network model in [21], this paper’s model is superior.

SOTA Comparisons (II)

The proposed method has the highest modify accuracy (MAcc), which is reaching 0.8950. Therefore, the method of this paper is competitive on the problem of heart sound classication and recognition.



Sik-Ho Tsang

PhD, Researcher. I share what I learn. :) Linktree: for Twitter, LinkedIn, etc.