Case study · Lung-sound classification
Hearing Disease in Breath Sounds
COPD and pneumonia from lung sounds: EMD/DWT decomposition, MFCC/GTCC features, QDA.
Classifying chronic obstructive pulmonary disease (COPD) and pneumonia from recorded lung sounds, part of a master's thesis where I provided technical support and carried out the biomedical signal-processing and machine-learning work in MATLAB. Empirical mode decomposition (EMD), discrete wavelet transform (DWT), MFCC and GTCC features and a QDA classifier reached 99% accuracy.
- Org
- MS thesis
- Status
- Research
- Channels
- CH2CH3
The problem
COPD and pneumonia sound different through a stethoscope. Can a model tell them apart from recorded lung sounds?
What I built
- Part of a master's thesis: I provided technical support and carried out the signal-processing and ML work
- EMD and DWT decomposition of lung-sound recordings
- MFCC and GTCC features with a QDA classifier
Stack
MATLAB · EMD · DWT · MFCC · GTCC · QDA
Results
- 99% COPD/pneumonia classification accuracy
Lung-sound recordings are decomposed with EMD and DWT, described with Mel-frequency (MFCC) and gammatone (GTCC) cepstral coefficients, and classified with quadratic discriminant analysis (QDA).
Automated analysis of breath sounds points towards digital stethoscopes and low-cost respiratory screening.