Comparing the Diagnostic Accuracy of Artificial Intelligence and Physicians for Pneumonia

Background: Pneumonia is one of the most important causes of death of infectious diseases worldwide, particularly in young children. Chest X-rays are a key part of the diagnostic process in getting the diagnosis right quickly; however, this may be subject to human interpretation. Artificial intelligence (AI) holds promise for image analysis. Objective: The aim of this research is to examine the diagnostic accuracy of AI systems with respect to pneumonia when compared to physicians. We assumed that a model guided by image data, and trained on that image data, can also competently recognize pneumonia from case descriptions, or even perform as well as a physician. Experimental design: Comparative (experimental) design was used. Five physicians and five AI tools such as ChatGPT and Gemini were invited to assess the ten chest X-ray case descriptions (five normal, five pneumonia). They each made a diagnosis and were scored as correct/incorrect based on known answers. Each participant's percentage accuracy was calculated. Results: With this small sample, AI accuracy was 100%, but physician accuracy was lower, ranging from 80% to 100% (mean 88%). Discussion: These results are consistent with previous research reporting high AI sensitivity, but they do not corroborate the real-world reports of AI limitations. Generalizability is limited by the small sample size and the use of case text (not images). Conclusion: AI outperforms doctors under the study conditions and there is a need for wider validation. Suggested improvements: Larger studies with real Xray images and clinical context are needed to ensure that conclusions are reliable.