Computerized Diagnostic Assistant for the Automatic Detection of Pneumothorax on Ultrasound: A Pilot Study

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonology

Background:

• Bedside thoracic ultrasound (US) is a rapid diagnostic tool for pneumothorax (PTX), surpassing physical examination and chest radiography (CXR) in some aspects.
• However, the accuracy of US for PTX diagnosis is operator-dependent.
• A computerized diagnostic assistant was developed to automatically detect sonographic signs of PTX.

Purpose of the Study:

• To evaluate the diagnostic accuracy of a novel computer algorithm designed to detect pneumothorax (PTX) on bedside thoracic ultrasound (US) images.
• To compare the algorithm's performance against an expert panel of US-trained physicians.

Main Methods:

• A retrospective study analyzed 146 archived bedside thoracic US examinations from adult emergency department patients.
• A computer algorithm interpreted B-mode video clips for pleural sliding and M-mode images for the seashore sign.
• The algorithm's interpretations were compared to a blinded, independent expert panel of three experienced physicians.

Main Results:

• The algorithm achieved a sensitivity of 79% (95% CI [63-89]) and a specificity of 87% (95% CI [77-93]) in detecting pneumothorax (PTX).
• For the highest quality images, the algorithm demonstrated 100% sensitivity and 92% specificity.
• Out of 133 analyzed examinations, 79 were true negatives, 33 true positives, 9 false negatives, and 12 false positives.

Conclusions:

• The developed computer algorithm shows potential as an aid for clinicians in identifying sonographic signs of pneumothorax (PTX).
• Further refinement and prospective validation are necessary before the algorithm can be integrated into clinical practice.
• The tool could help overcome the operator dependency of bedside thoracic ultrasound.