A new statistical diagnostic tool for respiratory diseases
Martin Wiegand, Saralees Nadarajah, John A. Smith, Kimberley J. Holt, Kevin Mcguinness · 2020
Background: Respiratory diseases can often be long-term afflictions and may severely affect a patient’s quality of life. Though treatments are available, the underlying mechanisms of the symptoms are not well understood. Aim: Develop and validate algorithms able to distinguish patients with different respiratory diseases, based on 24h cough sound recording data. Methods: With appropriate equipment, accurate recordings of patients’ coughing sequences can be created with relative ease. Based on this data our algorithm (Fig. 1) deduces the most probable diagnosis of a patient, or withholds a prediction if multiple classes are likely. Results: For the study 305 patient data sets were analysed: The rounded results of the best performing sub-algorithm for each tested combination are as follows: Conclusion: While the classification approach may provide a useful tool for the identification of respiratory diseases, it also provides valuable insight into the similarities of cough characteristics across diagnoses.Applications of the algorithm in different clinical setting could be considered.