Respiratory Sound Track Grand Challenge 2024: Respiratory Sound Compression and Event Detection for SPRSound Dataset

About

According to the World Health Organization, respiratory diseases, such as pneumonia, asthma, bronchitis, lung cancer, and chronic obstructive pulmonary disease (COPD), are among the most common mortality factors in the world, causing the death of more than 3 million people each year worldwide. These respiratory diseases have a direct impact on people’s social, economic, and health lives. Early diagnosis is the key factor for preventing the spread of respiratory diseases and limiting the adverse effects on people’s lives.

Digital stethoscopes, advancing the traditional auscultation technique, have emerged as a valuable tool in this context. These devices can record and transmit lung sounds to specialists remotely, aiding in the diagnosis of respiratory diseases. However, digital stethoscopes face limitations in their storage, computational, and transmission capacities, making the development of a respiratory sound compression algorithm essential.

Compressive sensing (CS)-based compression algorithms have attracted considerable attention in recent years. Such algorithms achieve compression through subsampling data below the Nyquist frequency, resulting in a simple compression process and high compression ratio. However, as the reconstruction speeds of these algorithms tend to be slow, there is a need to develop algorithms with faster reconstruction speeds while maintaining high compression ratios and reconstruction fidelity.

Furthermore, the application of sound event detection algorithms to respiratory sounds is particularly critical in the context of digital stethoscope-based remote auscultation. By automatically locating and categorizing segments of respiratory sounds, these algorithms can significantly aid in the early and accurate identification of abnormal lung sounds, such as wheezes, crackles, or diminished breath sounds. This innovative approach not only enhances the capabilities of digital stethoscopes but also reduces the reliance on the expertise of medical professionals, paving the way for more accurate and scalable screening processes.

The integration of digital stethoscopes with advanced sound compression and sound event detection algorithms holds the promise of transforming the diagnosis and management of respiratory diseases globally, ultimately improving patient outcomes and healthcare efficiency.

The IEEE BioCAS 2024 grand challenge on respiratory sound compression and event detection invites participants to explore different feature extraction techniques and models to improve the current state-of-the-art works. This new dataset is collected from the Shanghai Children’s Medical Center (SCMC), targeting children ranging from 1 month to 18 years old, which presents unseen challenges as compared to the prior datasets.

Timeline

Participation

1. The competition is open to individuals, colleges/universities, scientific research institutions, and enterprises. The maximum number of team members is 3.
2. To ensure compliance with local regulations during the competition, all participants should comply with the export control laws of their country. In case of any negative impact on the competition due to violation of any export control laws, the team members reserve the right to disqualify the relevant contestants and take legal action.
3. Participants shall express their interest to participate in this Grand Challenge by filling in the application form (https://forms.gle/SQvre2v58dDvDxXZ7). The participants are invited to submit their work as BioCAS papers selecting the special session "Lung Sound Design Contest". The papers will be regularly reviewed and, if accepted, must be presented at BioCAS 2024.

Awards

1. Top-4 team will be invited to present their work in IEEE BioCAS 2024.

Challenge and Dataset

Datasets

Our database is the first open access respiratory sound database in pediatric population, aging from 1 month to 18 years old. The respiratory sounds contained in the dataset were recorded at the pediatric respiratory department in Shanghai Children’s Medical Center (SCMC) using Yunting model II Stethoscope with the sampling frequency and quantization resolution of 8 kHz and 16 bit.

Our database contains 2,683 respiratory records with a total duration of 8.2 hours and each record lasts over 9 seconds. The recordings are saved in .wav format with naming rules as follows: Each name is compromised with 5 elements separated with underscores, including the patient number, age, gender, the recording location, and the recording number of the participants.

1. Patient number (e.g., 65101170)

2. Age (e.g., 0.4)

3. Gender

   a.  Male (0)

   b.  Female (1)

4. Recording location

   a.  left posterior (p1)

   b.  left lateral (p2)

   c.  right posterior (p3)

   d.  right lateral (p4)

5. Recording number (e.g., 3246)

The annotations at the event level are provided in this database. Each recording was segmented into multiple respiratory events and annotated as Normal, Rhonchi, Wheeze, Stridor, Coarse Crackle, Fine Crackle, or Wheeze+Crackle.

The event level annotation information of each recording is saved in .json format with the same filename. The annotation consists of the start (ms), the end (ms), and the corresponding type of the respiratory events (Normal, Rhonchi, Wheeze, Stridor, Coarse Crackle, Fine Crackle, Wheeze+Crackle).

An example of the annotation file is as follow:

{
    "event_annotation": [
        {
            "start": 342,
            "end": 2515,
            "type": "Normal"
        }, {
            "start": 2557,
            "end": 3776,
            "type": "Normal"
        }, {
            "start": 4547,
            "end": 5651,
            "type": "Normal"
        }, {
            "start": 6439,
            "end": 8065,
            "type": "Normal"
        }, {
            "start": 8363,
            "end": 9201,
            "type": "Normal"
        }
    ]
}

This database is freely available for research and can be downloaded from https://github.com/SJTU-YONGFU-RESEARCH-GRP/SPRSound. Publications using this database should cite the following paper:

Q. Zhang et al., "SPRSound: Open-Source SJTU Paediatric Respiratory Sound Database," in IEEE Transactions on Biomedical Circuits and Systems, vol. 16, no. 5, pp. 867-881, Oct. 2022, doi: 10.1109/TBCAS.2022.3204910.

Main Tracks

Track 1 (Respiratory Recording Compression): This track deals with respiratory recordings compression using compressive sensing-based compression methods.

Track 2 (Respiratory Event Detection): This track deals with the detection of onsets and offsets in addition to the assignment of event labels of respiratory events in respiratory recordings using sound event detection methods.

Evaluation Metrics

Submissions are assessed using specific metrics tailored to each track. For Track 1, evaluation criteria comprise Compression Ratio (CR), Percent Root Mean Square Difference (PRD), and Correlation Coefficient (CC). Track 2 submissions are evaluated based on event-based F-score (F) and event-based Error Rate (ER). The calculation methods are as follows:

Track 1:

(1) CR = (Length of initial respiratory recording) / (Length of compressed respiratory recording)

(2) PRD = √(∑(X_r(i)-X_i(i))² / ∑X_i(i)²)

(3) CC = Cov(X_i, X_r) / σ(X_i)σ(X_r)

Track 2:

(1) F = 2·TP / (2·TP+FP+FN)

(2) ER = (S+D+I) / N

(3) S = min(FN, FP)

(4) D = max(0, FN-FP)

(5) I = max(0, FP-FN)

(6) N = the number of respiratory events marked as active in the annotation

Rules

Every challenge participant agrees to use the provided data only in the scope of the DATA USE AND CONFIDENTIALITY AGREEMENT for access to data.

For training the model, no external data is allowed except the official dataset provided in this challenge.

The challenge does not encourage excessive stacking of models and hardware to brush up the score of the challenge.

Every challenge member agrees that the decisions of the challenge committee will be final and binding on all matters related to this challenge. If there is any change to data, schedule, instructions of participation, or these rules, the registered participants will be notified of the email addresses they provided when they are registered.

If an unforeseen or unexpected event (including, but not limited to: someone cheating; a virus, bug, or catastrophic event corrupting data or the submission platform; someone discovering a flaw in the data or modalities of the challenge) that cannot be reasonably anticipated or controlled, (also referred to as force majeure) affects the fairness and/or integrity of this challenge, the committee reserve the right to cancel, change or suspend this challenge. This right is reserved whether the event is due to human or technical error.

Submission

All individuals, teams and each member should submit the data use agreement in advance. (If not, your submissions will not be accepted.)

Multiple submissions are allowed. Please limit your submission to 1 per day.

The submitted files must be compressed in zip format. The main script (main.py) should be provided in the submitted files, which is the executable file for model evaluation. The readme.md and requirements.txt should be provided to describe the model architecture and list all the dependencies of your python project, respectively.

The command line and parameter requirements of the execution code are as follows:

python3 main.py --track track_number --wav /path/to/wav_path/ --out /path/to/output/

The track_number is 1 or 2 (representing track 1 and track 2, respectively). The output indicates the directory where the output files are located. The output folder structure for track 1 and track 2 are as follows:

Track 1:

<Team Name>
├─ wav_file_name1
│    ├─ compressed.wav
│    └─ reconstructed.wav
├─ wav_file_name2
│    ├─ compressed.wav
│    └─ reconstructed.wav
└─ wav_file_name3
       ├─ compressed.wav
       └─ reconstructed.wav
 ...

Track 2:

<Team Name>
├─ wav_file_name1.json
├─ wav_file_name2.json
└─ wav_file_name3.json
 ...

The format of output json (UTF-8) is as follows:

{
    {
        "start": 342,
        "end": 2515,
        "type": "Normal"
    }, {
        "start": 2557,
        "end": 3776,
        "type": "Normal"
    }, {
        "start": 4547,
        "end": 5651,
        "type": "Normal"
    }, {
        "start": 6439,
        "end": 8065,
        "type": "Normal"
    }, {
        "start": 8363,
        "end": 9201,
        "type": "Normal"
    }
}

Live Demo Submission

As part of this grand challenge, we encourage our participants to submit a live demo in which you deploy your model on a cell phone for real-time applications. We hope you can open-source your project to help accelerate this research direction.

You can refer to our past demo (BioCAS 2019) in the following Github link:

https://github.com/SJTU-YONGFU-RESEARCH-GRP/Lung-Sound-Classification-System-LungSys-I