Dataset Description: Tracheal sound analysis for detection of respiratory depressions 

Overview

Increasingly, both surgical and non-surgical procedures—such as dental work, endoscopy, cosmetic surgery, and cataract surgery—are performed under sedation analgesia using a combination of sedative and narcotic drugs. These agents must be accurately titrated to meet individual patient needs, with close monitoring of their effects on respiratory function. 

Common monitoring techniques, such as pulse oximetry, often have a considerable delay in detecting respiratory complications. Additionally, side-stream capnography has limited effectiveness for detecting respiratory depression due to challenges such as sampling errors, lumen obstruction by airway secretions, and frequent detachment from the patient’s airway. Therefore, direct monitoring of airway patency using auscultatory techniques is crucial during sedation analgesia.

Continuous monitoring of tracheal sounds with traditional or electronic stethoscopes can reliably and rapidly detect airway complications before they lead to serious issues. However, the overall efficacy of tracheal stethoscopes as airway monitors depends on the continuous listening to respiratory sounds by the anesthesia team. Continuous operator listening may not be practical, and interrupted auscultation can be associated with significant airway problems. Therefore, the development of real-time, automatic, and continuous techniques for tracheal sound monitoring and analysis is essential for effective airway monitoring during sedation analgesia.

Data Acquisition

Following approval from the institutional ethical committee and informed consent from the patients, we conducted a study involving 16 adults classified as ASA I and II, all scheduled for cataract surgery under sedation anesthesia. Patients with a history of respiratory diseases were excluded from the study. 

Upon positioning on the operating table, all patients received supplemental oxygen via a mask. Monitoring included ECG, non-invasive blood pressure, and pulse oximetry. Tracheal sound recording commenced one minute prior to the administration of sedative drugs using a C417 omni-directional condenser Lavalier microphone (AKG Acoustics, Vienna, Austria), secured over the suprasternal notch with double-sided adhesive tape. 

Tracheal sound recording continued throughout the procedure at a sampling rate of 44.1 KHz. After the study's completion, the recorded sounds were analyzed by the anesthesiologist to identify periods of respiratory depression, defined as episodes of apnea, breath-holding, or airway obstruction with a resolution of one second.

Data Structure

This dataset was utilized in one of the challenges at the IAI_2024 event, with the corresponding paper available at [paper link]. The dataset is divided into three categories: training, testing, and hidden datasets. Each phase of the competition provided the dataset sequentially: first the training dataset, followed by the testing dataset, and finally, the hidden dataset was revealed during the live competition to validate the results.

The dataset consists of 21,564 seconds of tracheal sound recordings, with 10.97% of the data comprising intervals of respiratory depression (RD), collected from 16 subjects. It is important to note that the training, testing, and hidden sets originate from distinct individuals. The table below provides detailed information about the tracheal sound dataset, reporting the duration of the recordings in seconds, with RD events indicated in parentheses. 

To clarify the table, consider the following example: the training folder contains 11 .wav files. The associated labels are provided in separate .docx files (e.g., label_S1.docx corresponds to S1.wav). Each .docx file includes the onset and offset timings of the respiratory depression intervals.

To download the Tracheal sound dataset for detection of Respiratory Depression (RD) intervals please click here 

If you wish to use any part of this dataset, kindly reference the following paper:-