Deep learning using multilayer perception improves the diagnostic acumen of spirometry: a single-centre Canadian...

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence
• Medical Diagnostics

Background:

• Spirometry and plethysmography are standard pulmonary function tests (PFTs) for lung disease.
• Plethysmography's inaccessibility often leads to spirometry being used alone, risking missed or misdiagnoses, particularly for restrictive lung diseases.
• There is a need for improved interpretation of spirometry alone.

Purpose of the Study:

• To develop a deep learning model for enhanced interpretation of spirometry.
• To improve the diagnostic accuracy of spirometry without the need for plethysmography.

Main Methods:

• A multilayer perceptron deep learning model was developed using full PFT data from multiple patient cohorts.
• Inputs included spirometry, plethysmography, and biometric data.
• The model was trained, validated, and tested on independent datasets and compared against decision tree and pulmonologist interpretations.

Main Results:

• The deep learning model using biometrics and spirometry achieved 95% accuracy after refinement.
• The final model significantly outperformed decision tree (75.61%) and pulmonologist (66.67%) interpretations.
• The model's accuracy was comparable to using full PFTs including plethysmography.

Conclusions:

• Deep learning significantly enhances the diagnostic capability of spirometry.
• The model accurately classifies lung physiology, achieving performance comparable to comprehensive PFTs.
• This AI-driven approach offers a promising solution for improving lung disease diagnosis where plethysmography is unavailable.