AI is reshaping pulmonary medicine and critical care — but unevenly. From detecting COPD and predicting sepsis to flagging hypotension before it happens, AI tools are showing real clinical promise. Still, experts warn of hype, generalizability gaps, and the persistent "black box" problem standing between innovation and widespread adoption.
AI is making meaningful strides in pulmonary medicine and critical care, but experts at the American Thoracic Society 2026 International Conference were quick to temper excitement with caution. Radiology still dominates — accounting for ~76% of the ~1,400 FDA-cleared AI medical devices — while pulmonology-specific tools are maturing but not yet fully ready for prime time. Key areas of progress include COPD detection, interstitial lung disease (ILD), pulmonary hypertension (PH), and ICU applications like sepsis prediction and hypotension forecasting.
In the ICU, AI tools like the Epic Sepsis Model V2 and the FDA-authorized Sepsis ImmunoScore are gaining traction, while the Acumen Hypotension Prediction Index (HPI) — backed by a randomized controlled trial — has shown it can meaningfully reduce time spent in hypotension during cardiac surgery. Experts also stressed that AI needs continuous monitoring post-deployment and that interpretability remains a critical unsolved challenge.
By the Numbers
Why it matters: As specialist shortages grow and patient complexity increases, AI could be a force multiplier in pulmonary and critical care — but only if issues of bias, transparency, and real-world performance are addressed head-on.
