The Role and Diagnostic Accuracy of Artificial Intelligence in Pulmonary Function Tests: A Systematic Review

The rapid advancement of artificial intelligence (AI) has highlighted its potential as a supportive tool in pulmonary function test (PFT) interpretation given the inherent biological variation, inter-rater variability, lack of confidence in result interpretation and restricted access within resource-constrained settings. This study aims to systematically review the published literature on the diagnostic accuracy of AI-based interpretation of PFTs and evaluate its implications for current and future clinical practice in healthcare. After screening, forty-seven publications met the inclusion criteria and were analysed to create a narrative summary. Four main over-arching themes were identified from the available literature. AI appears to consistently outperform non-specialists in diagnostic accuracy of PFTs and showed a synergistic effect when used as an adjunct in both specialist and non-specialist settings. Using AI software also had greater diagnostic accuracy than clinicians when presented with suboptimal or limited clinical information and investigations. It was also observed that the implementation of AI can address the issue of inter-rater variability by giving more consultant interpretations. Chronic obstructive pulmonary disease diagnosis saw the greatest accuracy with other conditions such as obstructive sleep apnoea showing limited evidence for the introduction of AI as a diagnostic tool. The evidence suggests that AI has the potential to play a significant role in the future of healthcare but should be used as an adjunctive tool as opposed to an independent diagnostic decision maker. One such way it could be utilised is as a triage/screening tool to help bridge the gap between primary and secondary care.