Deep Learning-Driven Early Diagnosis of Respiratory Diseases using CNN-RNN Fusion on Lung Sound Data
Bibliographic record
Abstract
This research depicts a deep learning-based algorithm designed for lung sound analysis, which combines Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) architectures to improve early disease detection. With comprehensive datasets from Coswara and ICBHI, the algorithm is proficient in distinguishing a spectrum of respiratory diseases, including pneumonia, asthma, and Chronic Obstructive Pulmonary Disease (COPD). Model pre-processing data with high pass filtering and segmented analysis of lung sound recordings, with Mel-spectrograms used as pivotal input features. The complete fusion model architecture integrates three CNN layers, three max-pooling layers, and two fully connected layers, the result is a feature map that highlights the presence of detected features, complemented by including two Long Short-Term Memory (LSTM) layers in the RNN component. The training process is devoted to the Adam optimizer alongside the cross-entropy loss function. Data augmentation techniques were applied to handle class imbalances and enhance model generalizability. The experimental results demonstrate high accuracy, sensitivity, specificity, and F1-score across various respiratory diseases. The performance metrics on the ICBHI dataset underscore the model's exceptional accuracy: 93.3% for healthy individuals, 93.8% for pneumonia patients, 91.7% for asthma patients, and 94.0% for COPD patients. The model outperforms alternative algorithms such as decision trees, support vector machines, and random forest regarding precision, recall, F1 score, and accuracy across the ICBHI and Coswara datasets. This noteworthy outcome positions the algorithm as an advanced and effective solution for progressing the domain of respiratory disease diagnosis through lung sound analysis. The model also provides interpretable visual explanations using Grad-CAM, along with confidence estimates, to enhance clinical trust.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".