Deep learning feasibility for intussusception detection on POCUS

Fine-tuning deep learning models outperformed classical and transfer learning for detecting intussusception on point-of-care ultrasound, with a threshold-based approach yielding the best predictive performance in a study of 49 patients (8 positive). Small dataset limits generali…

• Fine-tuning models outperformed classical machine learning, ensemble, and transfer learning across all analyses.
• A threshold-based case analysis approach provided the greatest predictive performance.
• The study demonstrates feasibility of deep learning for intussusception detection using a small POCUS dataset.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.