What is the COPD Detection System?

The COPD Detection System is an advanced deep learning application designed to screen and classify the severity of Chronic Obstructive Pulmonary Disease (COPD). By analyzing digital recordings of respiratory sounds (lung auscultations), the system leverages Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to provide rapid, non-invasive diagnostic insights.

Problem Statement

COPD is a leading cause of global mortality. Early detection is critical, but traditional spirometry tests require specialized equipment and trained personnel. There is a pressing need for accessible, low-cost, and reliable screening methods that can operate in remote or resource-constrained environments using simple audio inputs.

Tech Stack

• Machine Learning: Python, TensorFlow, PyTorch, Scikit-learn
• Audio Processing: Librosa (for extracting MFCCs and Mel-spectrograms)
• Model Architectures: CNN, CNN-LSTM hybrid models
• Data Visualization: Matplotlib, Seaborn

Architecture & Implementation

The project utilizes the ICBHI Respiratory Sound Database. The implementation pipeline includes:

1. Signal Processing: Raw audio files are preprocessed to remove noise, and Mel-Frequency Cepstral Coefficients (MFCCs) are extracted to convert audio data into visual spectrogram representations.
2. Feature Extraction: A deep CNN architecture is utilized to extract spatial features from the spectrograms.
3. Sequential Modeling: The extracted features are fed into an LSTM layer to capture the temporal dynamics and sequential nature of breathing cycles.
4. Classification: The final dense layers output a probability distribution predicting whether the patient is healthy or has COPD, along with severity classification.

Results & Metrics

The deployed CNN-LSTM hybrid model achieved exceptional performance on the validation dataset:

• Overall Accuracy: Achieved over 95% accuracy in binary classification (Healthy vs. COPD).
• Robustness: The integration of LSTM significantly reduced false positives compared to a standalone CNN baseline.