Can AI generate diagnostic reports for radiologist approval on CXR images? A multi-reader and multi-case observer...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Detecting lung abnormalities in heterogeneous chest X-ray (CXR) images and generating reports is challenging and time-consuming.
• Current methods often require significant radiologist effort and time.

Purpose of the Study:

• To evaluate the effectiveness of a novel AI system, MOM-ClaSeg, in improving radiologist accuracy and efficiency for detecting lung abnormalities.
• To compare AI-assisted single radiologist reading with traditional double reading using a multi-reader, multi-case (MRMC) study design.

Main Methods:

• Retrospective collection of over 36,000 CXR images from 12 hospitals.
• Comparison of two reading methods: traditional double reading by two radiologists (control) versus single radiologist reading augmented by AI-generated reports (experiment).

Main Results:

• AI-assisted single reading significantly increased diagnostic accuracy by 1.49% and sensitivity by 10.95% compared to double reading (P < 0.001).
• Diagnostic specificity showed a minimal, non-significant difference (0.22%, P = 0.255).
• Average image reading and diagnostic time decreased by 54.70% with AI assistance (P < 0.001).

Conclusions:

• The MOM-ClaSeg AI system can function as a "first reader," enhancing radiologist performance.
• AI-assisted single reading achieves superior diagnostic accuracy and efficiency compared to conventional double reading for CXR interpretation.