Brachycephalic Obstructive Airway Syndrome (BOAS) classification in dogs based on respiratory noise analysis using machine learning
Publicerad
Författare
Typ
Examensarbete för masterexamen
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
Brachycephalic Obstructive Airway Syndrome (BOAS) is a problem in several dog breeds
due to a compressed shape of the skull. It is classified as BOAS grade 0-3, where 0 is normal
breathing and 3 is the most severe grade of the syndrome. Grade 2-3 can cause great
suffering for the affected dogs and needs treatment. This study aimed to find a method using
machine learning to classify the BOAS grade based on audio recordings of respiratory
noise. The recordings were converted into Mel-Frequency Cepstral Coefficients (MFCCs)
to be processed as images by the network. The results proved that Recurrent Neural Network
- Long Short-Term Memory (RNN-LSTM) was a successful method to classify the
four different BOAS grades with an accuracy of about 86-87% for dictaphone recordings
and about 62-66% for stethoscope recordings. Convolutional Neural Networks (CNN) also
managed to classify the BOAS grades but this method was less accurate, with an accuracy
of approximately 74-76% for dictaphone recordings and 50-54% for stethoscope recordings.
The study was a collaboration between Chalmers University of Technology and
Swedish University of Agricultural Sciences.
Beskrivning
Ämne/nyckelord
Brachycephalic Obstructive Airway Syndrome, BOAS, Machine learning, Convolutional Neural Network, CNN, Mel-Frequency Cepstral Coefficients, MFCC, Recurrent Neural Network, Long Short-Term Memory, RNN-LSTM, Respiratory noise analysis