Validating the accuracy of deep learning for the diagnosis of pneumonia on chest x-ray against a robust multimodal reference diagnosis: a post hoc analysis of two prospective studies
Problem
Deep learning pneumonia diagnosis on chest X-rays had mainly been validated against radiologists, whose own diagnoses can be inaccurate, so accuracy was tested against a robust multimodal expert reference.
Why it matters
Reliable pneumonia diagnosis from chest X-rays directly affects treatment decisions and patient outcomes.
What the AI did
A DL convolutional neural network was trained to diagnose pneumonia and evaluated in two prospective cohorts (430 patients) against a multidisciplinary expert panel reference, compared with radiologists and emergency physicians.
Results
DL and radiologists showed similar CXR accuracy in both cohorts (p ≥ 0.269); cohort 1 radiologist 1: 75.5% (95% CI 69.1-80.9); DL accuracy comparable.
Impact
Prospective validation found DL accuracy statistically similar to senior radiologists, supporting AI as a diagnostic aid.
Record
| Venue | European Radiology Experimental |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Jérémy Hofmeister, Nicolás Garin, Xavier Montet, Max Scheffler, Alexandra Platon, Pierre‐Alexandre Poletti, Jérôme Stirnemann, Marie‐Pierre Debray et al. |
| Citations | 25 (OpenAlex, at import time) |
| References | 37 |
| Model or method | deep learning convolutional neural network |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1186/s41747-023-00416-y |
| Open access | gold |
| Keywords | deep learning, pneumonia, chest x-ray, validation, prospective study, diagnosis |
Importance rating
significant — thousands to millions affected
Reliable pneumonia diagnosis from chest X-rays guides treatment decisions; DL matched senior radiologists in 430-patient prospective validation, but no scale stated.