The lung scan and the abnormal chest X-ray: difficult diagnoses

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Ventilation-perfusion (V-P) lung scans are increasingly used for patients unable to undergo contrast CT.
• This shift may increase the proportion of V-P scans performed on patients with abnormal chest radiographs.
• Diagnosing pulmonary embolism in patients with abnormal chest X-rays presents a clinical challenge.

Purpose of the Study:

• To evaluate computer-based analysis of V-P scans for assisting clinical interpretation.
• To determine if AI can improve diagnostic accuracy in patients with abnormal chest radiographs.

Main Methods:

• Radionuclide V-P images were acquired from 262 patients (118 normal, 144 abnormal chest radiographs).
• Artificial neural networks (ANNs) were developed using objective image data to detect pulmonary embolism.
• ANN predictions were compared against clinical interpretations and pulmonary angiography results.

Main Results:

• The ANN-based method showed performance comparable to clinical interpretation for both normal and abnormal chest radiographs.
• Averaging clinical and ANN estimates yielded higher accuracy than either method alone.
• Computer analysis offers a valuable, distinct perspective for V-P scan interpretation.

Conclusions:

• Computer-based V-P scan analysis is as effective as clinical interpretation for patients with abnormal chest radiographs.
• Integrating digital analysis with standard clinical interpretation enhances diagnostic performance.
• This approach can improve diagnostic certainty for patients relying on V-P scans as a primary diagnostic tool.