Automatic pneumonia detection based on ultrasound video analysis

Area of Science:

• Medical Imaging
• Pediatric Pulmonology
• Artificial Intelligence in Medicine

Background:

• Pneumonia is a leading cause of mortality in children under five, especially in developing nations.
• Current diagnostic methods can be limited in resource-constrained settings.
• Novel, accessible diagnostic tools are crucial for early detection and treatment.

Purpose of the Study:

• To develop and evaluate a novel ultrasound video analysis application for pneumonia detection in children.
• To assess the algorithm's performance across various clinical and usage protocols.
• To explore the potential of this technique for screening applications.

Main Methods:

• Processing of short ultrasound video chunks using an image processing algorithm.
• Analysis of video frames to derive overall video statistics.
• Exploitation of geometrical properties of anatomical and pathological features in lung sonography.

Main Results:

• Encouraging performance with an Area Under the Curve (AUC) ranging from 0.7851 to 0.9177.
• Tested on 346 videos from children aged three to five years.
• Algorithm demonstrated effectiveness across diverse scanning protocols and probe maneuvers.

Conclusions:

• The proposed ultrasound video analysis technique shows significant potential for pneumonia detection in children.
• The algorithm can serve as a standalone classifier or as a preliminary stage in a larger diagnostic pipeline.
• This method offers a promising avenue for improving pneumonia screening and diagnosis, particularly in underserved regions.