Classification of Pulmonary Crackle and Normal Lung Sound Using Spectrogram and Support Vector Machine

Achmad Rizal; Wahmisari Priharti; Dien Rahmawati; Husneni Mukhtar

doi:10.4028/p-tf63b7

Paper Titles

Hand Exoskeleton Development Based on Voice Recognition Using Embedded Machine Learning on Raspberry Pi
p.81

Development of Electromyography Signal Contraction Detection Using the Double Threshold Method for Controlling Wheelchair
p.93

Comparison of Time-Frequency Analyzes for a Sleep Staging Application with CNN
p.109

Using Pectoral Muscle Removers in Mammographic Image Process to Improve Accuracy in Breast Cancer
p.131

Classification of Pulmonary Crackle and Normal Lung Sound Using Spectrogram and Support Vector Machine
p.143

Machine Learning Based Real-Time Diagnosis of Mental Stress Using Photoplethysmography
p.154

The Effect of Feature Selection on Gray Level Co-Occurrence Matrix (GLCM) for the Four Breast Cancer Classifications
p.168

Near Infrared LEDs-Based Non-Invasive Blood Sugar Testing for Detecting Blood Sugar Levels on Diabetic Care
p.183

Multimodal Instrumentations as Fatigue Detection Using Fuzzy Logic Method
p.192

HomeJournal of Biomimetics, Biomaterials and...Journal of Biomimetics, Biomaterials and...Classification of Pulmonary Crackle and Normal...

Classification of Pulmonary Crackle and Normal Lung Sound Using Spectrogram and Support Vector Machine

Abstract:

Crackles is one of the types of adventitious lung sound heard in patients with interstitial pulmonary fibrosis or cystic fibrosis. Pulmonary crackles of discontinuous short duration appear on inspiration, expiration, or both. To differentiate these pulmonary crackles, the medical staff usually uses a manual method, called auscultation. Various methods were developed to recognize pulmonary crackles and distinguish them from normal pulmonary sounds to be applied in digital signal processing technology. This paper demonstrates a feature extraction method to classify pulmonary crackle and normal lung sounds using Support Vector Machine (SVM) method using several kernels by performing spectrograms of the pulmonary sound to generate the frequency profile. Spectrograms with various resolutions and 3-fold cross-validation were used to divide the training data and the test data in the testing process. The resulting accuracy ranges from 81.4% - 100%. More accuracy values of 100% are generated by a feature extraction in several SVM kernels using 256 points FFT with three variations of windowing parameters compared to 512 points, where the best accuracy of 100% was produced by STFT-SVM method. This method has a potential to be used in the classification of other biomedical signals. The advantages of that are that the number of features produced is the same as the N-point FFT used for any signal length, the flexibility in the STFT parameters changes, such as the type of window and the window's length. In this study, only the Keiser window was tested with specific parameters. Exploration with different window types with various parameters is fascinating to do in further research.