Comparing expert systems for identifying chest x-ray reports that support pneumonia

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Radiology Informatics

Background:

• Accurate identification of acute bacterial pneumonia from chest x-ray reports is crucial for timely patient care.
• Computerized methods offer potential for efficient and consistent analysis of radiological reports.

Purpose of the Study:

• To evaluate the performance of four computerized methods in identifying chest x-ray reports indicative of acute bacterial pneumonia.
• To compare machine learning approaches against expert-constructed systems, keyword search, laypersons, and physicians.

Main Methods:

• Development and comparison of two expert-knowledge-based computerized methods.
• Development and comparison of two machine learning-based computerized methods.
• Performance evaluation using chest x-ray reports.

Main Results:

• Machine learning systems achieved performance comparable to expert-constructed systems.
• All computerized methods demonstrated superior performance to baseline keyword search and layperson evaluation.
• Computerized methods performed on par with physician evaluations.

Conclusions:

• Machine learning is a viable and effective tool for identifying chest x-ray reports supporting acute bacterial pneumonia.
• Computerized analysis, particularly using machine learning, shows promise in augmenting radiological report interpretation.