Deep Learning on New Neonatal Chest X-Ray Dataset Boosts RDS and Aspiration Diagnosis Accuracy by 4-7%

A deep learning model trained on a new neonatal pulmonary ailment dataset achieved 4-7% higher accuracy over baselines in classifying respiratory distress syndrome and aspiration syndrome on chest X-rays.

• Novel Chinese neonatal pulmonary dataset (NPA) was created, including normal and diseased cases with varying severity.
• Proposed method uses Polluted CutMix augmentation and uncertainty-aware pseudo-labeling to handle severe class imbalance.
• Method outperformed existing baselines by 4-7% in classification accuracy (primary result, no statistical measures provided).

Automated summary

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