Datasets
Our sadaco framework currently supports the datasets below. Here, we provide information on how to download and prepare each dataset.
ICBHI 2017 Challenge dataset [Link]
ICBHI dataset is an audio dataset containing audio samples, collected independently by two research teams in two different countries, over several years. Most of the database consists of audio samples recorded by the School of Health Sciences, University of Aveiro (ESSUA) research team at the Respiratory Research and Rehabilitation Laboratory (Lab3R), ESSUA and at Hospital Infante D. Pedro, Aveiro, Portugal. The second research team, from the Aristotle University of Thessaloniki (AUTH) and the University of Coimbra (UC), acquired respiratory sounds at the Papanikolaou General Hospital, Thessaloniki and at the General Hospital of Imathia (Health Unit of Naousa), Greece.
Dataset Specification
Classes
Normal : Healthy breathing sound without any other symptoms.
Crackle : Crackle sound is a series of short, explosive sounds. They can also sound like bubbling, rattling, or clicking. You’re more likely to have them when you breathe in, but they can happen when you breathe out, too.
You can have fine crackles, which are shorter and higher in pitch, or coarse crackles, which are lower. Either can be a sign that there’s fluid in your air sacs.
They can be caused by:Pneumonia
Heart disease
Pulmonary fibrosis
Cystic fibrosis
COPD
Lung infections, like bronchitis
Asbestosis, a lung disease caused by breathing in asbestos
Pericarditis, an infection of the sac that covers your heart
Wheeze : This high-pitched whistling noise can happen when you’re breathing in or out. It’s usually a sign that something is making your airways narrow or keeping air from flowing through them.
Two of the most common causes of wheezing are lung diseases called chronic obstructive pulmonary disease (COPD) and asthma. But many other issues can make you wheeze, too, including:Allergies
Bronchitis or bronchiolitis
Emphysema
Epiglottitis (swelling of the top flap of your windpipe)
Gastroesophageal reflux disease (GERD)
Heart failure
Lung cancer
Sleep apnea
Pneumonia
Respiratory syncytial virus (RSV)
Vocal cord problems
An object stuck in your voice box or windpipe
Data Count and Duration
The database consists of a total of 5.5 hours of recordings containing 6898 respiratory cycles, of which 1864 contain crackles, 886 contain wheezes, and 506 contain both crackles and wheezes, in 920 annotated audio samples from 126 subjects.
Fraiwan dataset [Link]
Fraiwan dataset is collected by Jordan University of Science and Technology Faculty of Computer and Information Technology, King Abdullah University Hospital. These data contain actual lung sound recordings from 112 Middle Eastern people suffering from a variety of pulmonary health issues. The data complement, augment, and balance the limited publicly available comparable datasets. Furthermore, the data can be used to train health professionals’ auscultatory skills.
Dataset Specification
There are three types of filters encoded in Fraiwan:
Bell mode filtration (B): Amplifies sounds in the frequency range 20-100Hz, but emphasizes the low frequency sounds in the range 20-200Hz.
Diaphragm mode filtration (D): Amplifies sounds in the frequency range 20-200Hz, but emphasizes the high frequency sounds in the range 100-500Hz.
Extended mode filtration (E): Amplifies sounds in the frequency range 20-1000Hz, but emphasizes the frequency sounds in the range 50-500Hz.
Classes
There are seven different classes in Fraiwan dataset:
Sound Type |
No. of Subjects |
|---|---|
Wheezes |
41 |
Normal |
35 |
Crepitations |
23 |
Crackles |
8 |
Wheezes and Crackles |
2 |
Bronchial and Crackles |
2 |
Bronchial |
1 |
Data Count and Duration
This dataset includes respiratory sounds from 112 subjects, 35 of whom are healthy and the rest are not, with just one recording per person and three types of filters applied to each recording. In other words, for each subject, we have three filtered recordings. Hence, we have 112 * 3, 336 recordings in total. For example, the file named BP60_heart failure,Crep,P L L,83,F.wav is a Bell filtered crepitation sound captured from the posterior left lower zone of the chest of an 83-year-old female with heart failure problem.
These volunteers ranged in age from 21 to 90 years old, with 43 females and 69 males. Each recording lasts 5 to 30 seconds, which is enough time to cover at least one breathing cycle. The maximum duration is limited by the recording capabilities of the electronics stethoscope. There were no minimum duration requirements imposed on the physician examining the subjects at the annotation stage.