Novel ML for pulmonary function testing on oscillometry devices

• Pulmonary function tests to classify respiratory physiology are critical for the diagnosis and management of patients with lung diseases.
• Traditional PFT methods:
  
  • Take up to 45 minutes for the gold standard PFT. 
  • Often require manual interpretation by specialists, which can be time-consuming and subject to variability
  • May be uncomfortable for patients
  • Take place in specialized centres

TECHNOLOGY

• The invention is a novel machine learning architecture that addresses the need for an automated, fast, and accurate assessment of lung function patterns during monofrequency oscillometry tests, which are less expensive, more mobile machines that can be deployed at the point-of-care.
• The architecture utilizes raw output measurements of volume, pressure, and flow during the mono-frequency modality of oscillometry devices (e.g. Thorasys). It introduces several innovative elements, including the use of MiniROCKET random filter bank algorithm for feature generation, a ridge regression classifier model, a soft voting scheme, and the concept of a "grey zone" to enhance predictive confidence. 
• Preliminary test results show that the proposed pipeline is capable of assessing restrictive, obstructive, and normal pulmonary function patterns with high accuracy.

BENEFIT

By offering a fast, accurate, and effective means of assessing lung function patterns at point of testing, the technology has the potential to disrupt traditional PFT:

• Only takes 5-10 minutes to complete vs. 45 minutes for the gold standard PFT.
• Model not only diagnoses obstruction cases with 100% accuracy but also has higher accuracy in classifying normal subjects (on raw spectral data)
• It can also help interpret oscillometry results in clinical use, leading to faster and more accurate diagnoses of respiratory problems.
• This technology can streamline the diagnostic process, reduce subjectivity, and enhance overall clinical efficiency.

STATUS

• PCT filed Sept 2024