Recursive Confidence Propagation in Medical Diagnosis: A Hierarchical Uncertainty Framework Using Confident Learning and GANs

Vasu Jindal, Huijin Ju, Zili Lv

Confirmation: I have read and agree with the IEEE BHI 2025 conference submission's policy on behalf of myself and my co-authors.

Keywords: Medical diagnosis, Uncertainty quantification, Confident learning, Generative adversarial networks

Abstract:

Medical diagnosis involves hierarchical uncertainty
where confidence at each level depends on deeper diagnostic
levels, but existing methods treat this as a single-layer problem.
We introduce MRC, the first framework for modeling recursive
diagnostic uncertainty in healthcare. MRC combines hierarchi-
cal confident learning with uncertainty-aware GANs to handle
complex probability distributions from recursive confidence re-
lationships. The approach identifies unreliable examples across
diagnostic levels while accounting for cross-level dependencies
and learns realistic clinical uncertainty patterns. Evaluated on
cardiology, radiology, and neurology datasets, MRC shows 2.5-
4.1% improvement in diagnostic accuracy and 23-41% improve-
ment in confidence calibration versus existing methods. MRC
provides the first principled approach to recursive diagnostic un-
certainty while maintaining clinical interpretability and workflow
integration.

Track: 4. Clinical Informatics

Registration Id: QPNRMHBLRJY

Submission Number: 362