Domain Specific Transporter Framework to Detect Fractures in Ultrasound

Abstract

Ultrasound examination for detecting fractures is ideally suited for Emergency Departments (ED) as it is fast, safe (from ionizing radiation), has dynamic imaging capability, and is easily portable. High variability in manual assessment of ultra-sound has piqued research interest in automatic assessment using Deep Learning (DL). Most DL techniques are trained on large labeled datasets which is expensive and requires many hours of careful annotation. We propose an unsupervised, domain-specific transporter framework to identify relevant key points from ultrasound scans providing a concise geometric representation highlighting regions with high structural variation. We incorporate domain-specific information using instantaneous local phase (LP) which detects bone features. We validate the technique on wrist 3DUS videos obtained from 30 subjects each independently assessed by 3 readers to identify fractures. The saliency of key points detected is compared against manual assessment based on distance from relevant features. Our approach was able to accurately detect 180/250 bone regions. We expect this technique to increase the applicability of ultrasound in fracture detection.