Respiratory Sound and Functional Measurements Database

• RS-FMD is a multimodal database that provides synchronized pulmonary and cardio‐pulmonary measurements including auscultation, spirometry, X‐ray imaging, and questionnaires.
• It employs standardized acquisition and annotation protocols to ensure high-fidelity signals from 12-channel lung and 4-channel heart recordings, chest X‐rays, and PFT data.
• The dataset supports robust machine learning and signal processing research for detecting and monitoring respiratory diseases such as COPD and asthma.

The Respiratory Sound and Functional Measurements Database (RS-FMD) is a multimodal dataset designed to support comprehensive research in pulmonary and cardio-pulmonary medicine through the provision of synchronized multimedia and clinical data. RS-FMD combines multi-channel auscultation recordings, chest X-ray images, spirometric and pulmonary function test (PFT) variables, and standardized health-related quality-of-life questionnaire results from a clinically diverse subject cohort. Its structure, data acquisition protocols, annotation procedures, and integration strategies facilitate methodologically rigorous analysis and algorithm development for both conventional and machine learning approaches (Altan et al., 2021).

1. Cohort Characteristics and Modalities

RS-FMD encompasses five primary modalities: (i) lung auscultation sounds (12 channels), (ii) heart auscultation sounds (4 channels), (iii) chest X-ray imaging, (iv) PFT variables with full spirometric curve capture, and (v) St. George’s Respiratory Questionnaire for COPD (SGRQ-C). Data were collected from 75 individuals (30 healthy controls and 45 subjects with pulmonary diseases including asthma, COPD stages 0–4 as per GOLD criteria, chronic bronchitis, and other lower respiratory tract disorders). The demographic composition includes 64 males and 13 females, with an age range of 38–68 years. Each data type contributes to a holistic view of respiratory function and pathology.

Modality	Channel/Sites	File Format / Resolution
Lung sounds	12 channels (upper, middle, lower lungs & costophrenic angles, anterior/posterior, L/R sides)	WAV (16-bit/4 kHz)
Heart sounds	4 channels (aortic, pulmonic, tricuspid, mitral)	WAV (16-bit/4 kHz)
Chest X-ray	Standard PA	DICOM (2048×2048, 12–16 bit)
PFT/Spirometry	FEV₁, FVC, FEV₁/FVC, (PEF optional), full curve	CSV (50–200 Hz)
SGRQ-C	Symptoms, Activity, Impacts	CSV (structured questionnaire)

2. Acquisition and Standardization Protocols

Auscultation recordings utilize two Bluetooth-enabled Littmann 3200™ electronic stethoscopes, enabling simultaneous capture from left and right anatomical sites with frequency options: Bell (20–200 Hz), Diaphragm (100–500 Hz), and Extended (20–1000 Hz). Data are natively in .zsa format and converted to standardized 16-bit mono WAV at 4 kHz using Littmann API. Each subject’s 12-channel lung and 4-channel heart recordings are time-synchronized via cough-peak alignment. Spirometry is performed using a standard clinical spirometer and disposable accessories; the protocol involves a forced vital capacity maneuver (≥5 seconds), and generates time-stamped CSV outputs capturing volume-time and flow-time curves. Chest X-ray images (posterior–anterior) are acquired in DICOM format, subjected to preprocessing (window/level adjustment) and lung-field ROI annotation within CMI software. SGRQ-C questionnaires are administered by clinicians or self-reported, with domain-specific and total scores stored as CSV for import into CMI (Altan et al., 2021).

3. Data Annotation and Clinical Validation

RS-FMD applies rigorous annotation and validation protocols, with all recordings and radiographic images reviewed independently by two pulmonologists. Diagnostic labels are based on the integrated assessment of auscultation, chest X-ray, and PFT findings. Adventitious sounds (e.g., wheezes, crackles) are time-stamped at the channel level. Annotation schema conforms to the following structure:

• Auscultation events: CSV file per channel, with onset_time, offset_time, and event label.
• Patient-level metadata: PatientID, age, sex, smoking history, clinical diagnosis, COPD stage, SGRQ total and domain scores.

This systematic annotation enables reproducibility in downstream analyses and supports machine learning applications that require high-fidelity ground truth.

4. Data Organization and Licensing

The database directory follows a hierarchical structure keyed to PatientID, with discrete subdirectories for auscultation sounds and annotations, X-rays, PFT records, and questionnaire responses. File formats are standardized: WAV for auscultation, CSV for annotations, PFTs, and questionnaires, and DICOM for imaging. Access is granted under a CC-BY-NC-4.0 license for non-commercial academic research, contingent on user agreement compliance as stipulated at www.respiratorydatabase.com (Altan et al., 2021).

Example of organization for a given PatientID:

/RS-FMD/ /P0001/ /ausc/ P0001_L1_Bell.wav P0001_L1_Bell.ann ... /xray/ P0001_PA.dcm /pft/ P0001_spirometry.csv /questionnaire/ P0001_SGRQC.csv

5. Principal Clinical and Statistical Formulas

Standard physiological parameters and quality-of-life scoring are computed as follows:

• Forced expiratory volume in one second (FEV₁): measured directly from spirometer volume–time curve at $t=1$ s.
• Forced vital capacity (FVC): total exhaled volume from expiration onset to volume plateau.
• FEV₁/FVC ratio: $\frac{\text{FEV}_1}{\text{FVC}}$
• Peak expiratory flow (PEF): $\max_t \left( \frac{dV(t)}{dt} \right )$
• SGRQ-C total score: $$S_{\rm total} = \frac {\sum_i w_i} {W_{\max}\times 100}$$
• Domain scores (Symptoms, Activity, Impacts) are calculated analogously using domain-specific item weights.

This formulaic transparency facilitates method comparability and reproducibility in statistical and machine learning studies.

6. Image and Signal Processing Methodologies

Chest X-ray analysis employs lung-field segmentation via intensity thresholding, optionally refined by morphological opening/closing operations. Extracted features include geometric (area, perimeter, aspect ratio) and texture (GLCM) descriptors. These radiomic features are extensible to algorithmic pipelines via CMI export. The capability for synchronized, multi-channel auscultation sound analysis enables detailed spatial and temporal correlation studies of cardiac and respiratory acoustics with imaging and functional phenotypes (Altan et al., 2021).

7. Research Applications and Integration

RS-FMD supports comprehensive workflows for loading multi-channel WAV and annotation files, aligning PFT curves and clinical labels, pre-processing DICOM radiographs, and extracting/analysing questionnaire metrics. The resource is purpose-built for algorithmic development and validation in pulmonary and cardiac disease detection, classification, and severity stratification, as well as for exploratory studies on the interplay of anatomical, acoustic, functional, and subjective health indicators. Its design enables end-to-end research spanning traditional signal processing to modern deep learning approaches, underpinning robust investigations into respiratory and cardio-pulmonary medicine (Altan et al., 2021).