Benchmarking Uncertainty and its Disentanglement in multi-label Chest X-Ray Classification
Researchers have developed a benchmark to evaluate uncertainty quantification in AI models used for multi-label chest X-ray classification. The study assessed 13 different methods across convolutional and transformer architectures using the MIMIC-CXR-JPG dataset. Findings highlight varying effectiveness and limitations in disentangling epistemic and aleatoric uncertainties depending on the method and model architecture. AI
IMPACT Establishes a benchmark for evaluating AI model trustworthiness in medical diagnostics, potentially improving diagnostic accuracy and safety.