An uncertainty-aware vision transformer-BiLSTM Bayesian framework for reliable clinical decision support using chest X-rays.
An AI system analyzes chest X-rays with 95.5% accuracy while openly reporting its confidence level, addressing a key barrier to clinical AI deployment.
This study introduces a Bayesian ViT-BiLSTM architecture for chest X-ray diagnosis that combines spatial and temporal feature extraction with uncertainty quantification, achieving 95.5% accuracy on large public datasets while providing calibrated confidence scores. The framework addresses key limitations of prior AI diagnostic systems by explicitly quantifying both aleatoric and epistemic uncertainty, relevant to clinical decision support implementation.
What the study was
- Study design
- Deep learning model development and validation study (MIMIC-CXR-JPG + PadChest-GR datasets)
- Population
- Chest X-ray datasets (MIMIC-CXR-JPG and PadChest-GR) for pneumonia, pulmonary fibrosis, and pleural effusion detection
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- Scientific Reports
Why it surfaced
Uncertainty quantification in AI radiology is an important research direction; model validation on public datasets only — needs prospective clinical testing.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.