aceImproving the fairness and robustness of deep learning classification models through bias mitigation and post hoc calibration
The goal of this study was to enhance the fairness and clinical applicability of deep learning models for classifying Alzheimer's disease from structural MRI scans. To achieve this, we addressed two key challenges: age-related bias and unreliable probability estimates. Specifically, we investigated in-processing debiasing strategies to achieve age-invariant predictions while preserving diagnostic accuracy. We also systematically evaluated post hoc calibration methods to improve the reliability, interpretability,…
A GLM-based residualization method (PMDN) removed age-related bias in a 3D CNN for Alzheimer's MRI classification while keeping balanced accuracy at 0.71 and improving calibration.
The adversarial debiasing approach (CF-Net) achieved only partial and unstable age-bias removal, and models with similar classification performance differed significantly in calibration, stability, and fairness.
Models achieved similar classification performance and moderate balanced accuracy (0.71), and CF-Net debiasing was unstable, so the benefits come mainly from fairness and calibration rather than from better diagnostic accuracy.
Evidence
- Peer-reviewedInternational Journal of Computer Assisted Radiology and Surgery2026-10-17
How should this claim be treated?
Truvace Impact Record TRV-2026-1958, v1: “Improving the fairness and robustness of deep learning classification models through bias mitigation and post hoc calibration.” Truvace, 2026-10-20. /record/TRV-2026-1958 (accessed at citation time). sha256 cc5731dd23e3956a…
Calibration history
Every change to this record since certification, in the open. None yet — the reading has held since it entered the record.
Certified into the record
How to verify without trusting this page
Fetch the canonical text of any version from /api/record/TRV-2026-1958 and hash it yourself — for example shasum -a 256 on the saved canonical field. The result must equal content_hash, and each version’s text ends with prev:followed by the prior version’s hash (version 1 chains to 64 zeros). If a single character of any version had been altered since certification, the chain would not reproduce.