{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"medRxiv"}],"indexed":{"date-parts":[[2026,5,18]],"date-time":"2026-05-18T16:11:18Z","timestamp":1779120678150,"version":"3.51.4"},"posted":{"date-parts":[[2026,5,15]]},"group-title":"Health Informatics","reference-count":48,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,5,15]],"date-time":"2026-05-15T00:00:00Z","timestamp":1778803200000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nc-nd\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,5,15]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                <jats:sec>\n                  <jats:title>Objectives<\/jats:title>\n                  <jats:p>To evaluate whether class-conditional conformal prediction (CP) can provide reliable uncertainty quantification (UQ) under severe class imbalance and distribution shift, using multiple sclerosis (MS) diagnosis from magnetic resonance imaging (MRI) as a clinical exemplar.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Methods<\/jats:title>\n                  <jats:p>We evaluated marginal and class-conditional CP using 720 T2-weighted MRI scans (142 MS, 578 controls). A convolutional neural network trained on 3 T data was evaluated under distribution shift (1.5 T acquisitions and synthetic image degradations). Through 100 Monte Carlo experiments, we assessed coverage guarantees, class-specific performance, and relationships between calibration set size, coverage variance, and uncertainty.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Results<\/jats:title>\n                  <jats:p>Marginal CP severely under-covered the minority MS class (16.9% mean coverage at 1.5 T vs. 95.2% for controls) despite valid population-level guarantees. Class-conditional CP dramatically improved MS coverage to 77.5% at 1.5 T and 85.8% at 3 T, significantly reducing severe undercoverage (&lt;80%) frequency while maintaining &gt;89% control coverage. Minority class coverage variance increased due to limited calibration samples, matching theoretical Beta-binomial predictions. CP maintained validity under distribution shift; prediction set sizes scaled monotonically with shift severity, yielding clinically interpretable UQ.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Conclusions<\/jats:title>\n                  <jats:p>Class-conditional CP successfully mitigates systematic undercoverage of minority disease classes while maintaining validity under distribution shift. The approach offers a practical, model-agnostic solution for uncertainty quantification applicable across clinical AI systems, though increased coverage variance for less represented conditions reflects fundamental statistical constraints. By characterizing these variance trade-offs, this framework enables more reliable deployment of diagnostic AI in heterogeneous clinical environments across diverse medical domains where minority disease class detection is critical.<\/jats:p>\n                <\/jats:sec>","DOI":"10.64898\/2026.05.12.26353057","type":"posted-content","created":{"date-parts":[[2026,5,15]],"date-time":"2026-05-15T22:35:24Z","timestamp":1778884524000},"source":"Crossref","is-referenced-by-count":0,"title":["Reliable Uncertainty Under Class Imbalance and Distribution Shift: Class-Conditional Conformal Prediction of Multiple Sclerosis"],"prefix":"10.64898","author":[{"ORCID":"https:\/\/orcid.org\/0009-0002-2094-3025","authenticated-orcid":false,"given":"Alexander S.","family":"Millar","sequence":"first","affiliation":[{"name":"Department of Biomedical Informatics, University of Utah, Suite 208, 421 Wakara Way, Salt Lake City, UT 84108, United States"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-7145-1640","authenticated-orcid":false,"given":"Cortnee","family":"Rom\u00e1n","sequence":"additional","affiliation":[{"name":"Department of Biomedical Informatics, University of Utah, Suite 208, 421 Wakara Way, Salt Lake City, UT 84108, United States"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-4345-9669","authenticated-orcid":false,"given":"Ramkiran","family":"Gouripeddi","sequence":"additional","affiliation":[{"name":"Department of Biomedical Informatics, University of Utah, Suite 208, 421 Wakara Way, Salt Lake City, UT 84108, United States"},{"name":"Utah Clinical and Translational Science Institute, University of Utah, Suite 208, 421 Wakara Way, Salt Lake City, UT 84108, United States"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-1449-477X","authenticated-orcid":false,"given":"Julio C.","family":"Facelli","sequence":"additional","affiliation":[{"name":"Department of Biomedical Informatics, University of Utah, Suite 208, 421 Wakara Way, Salt Lake City, UT 84108, United States"},{"name":"Utah Clinical and Translational Science Institute, University of Utah, Suite 208, 421 Wakara Way, Salt Lake City, UT 84108, United States"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"54368","reference":[{"key":"2026051808350938000_2026.05.12.26353057v1.1","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-16452-1_52"},{"key":"2026051808350938000_2026.05.12.26353057v1.2","unstructured":"A. Angelopoulos , S. Bates , J. Malik , and M. I. Jordan , \u201cUncertainty Sets for Image Classifiers using Conformal Prediction,\u201d Sep. 2022."},{"key":"2026051808350938000_2026.05.12.26353057v1.3","doi-asserted-by":"publisher","DOI":"10.1001\/jamainternmed.2021.2626"},{"key":"2026051808350938000_2026.05.12.26353057v1.4","doi-asserted-by":"publisher","DOI":"10.1038\/s41746-022-00611-y"},{"key":"2026051808350938000_2026.05.12.26353057v1.5","doi-asserted-by":"publisher","DOI":"10.3390\/healthcare13070701"},{"key":"2026051808350938000_2026.05.12.26353057v1.6","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pmed.1002683"},{"key":"2026051808350938000_2026.05.12.26353057v1.7","doi-asserted-by":"publisher","DOI":"10.1038\/s41746-024-01085-w"},{"key":"2026051808350938000_2026.05.12.26353057v1.8","doi-asserted-by":"publisher","DOI":"10.3233\/SHTI231113"},{"key":"2026051808350938000_2026.05.12.26353057v1.9","doi-asserted-by":"publisher","DOI":"10.1016\/j.cpet.2021.09.009"},{"key":"2026051808350938000_2026.05.12.26353057v1.10","doi-asserted-by":"publisher","DOI":"10.1177\/1352458520970841"},{"key":"2026051808350938000_2026.05.12.26353057v1.11","doi-asserted-by":"publisher","DOI":"10.3389\/fneur.2024.1448377"},{"key":"2026051808350938000_2026.05.12.26353057v1.12","doi-asserted-by":"publisher","DOI":"10.1007\/s44197-025-00353-6"},{"key":"2026051808350938000_2026.05.12.26353057v1.13","doi-asserted-by":"publisher","DOI":"10.3389\/fneur.2021.693017"},{"key":"2026051808350938000_2026.05.12.26353057v1.14","doi-asserted-by":"publisher","DOI":"10.1016\/S1474-4422(20)30067-3"},{"key":"2026051808350938000_2026.05.12.26353057v1.15","doi-asserted-by":"publisher","DOI":"10.1001\/jama.2018.20588"},{"key":"2026051808350938000_2026.05.12.26353057v1.16","doi-asserted-by":"publisher","DOI":"10.1016\/S1474-4422(17)30470-2"},{"key":"2026051808350938000_2026.05.12.26353057v1.17","doi-asserted-by":"publisher","DOI":"10.1177\/13524585231157276"},{"key":"2026051808350938000_2026.05.12.26353057v1.18","doi-asserted-by":"publisher","DOI":"10.1038\/s41591-024-03113-4"},{"key":"2026051808350938000_2026.05.12.26353057v1.19","doi-asserted-by":"publisher","DOI":"10.1016\/j.ijmedinf.2025.106014"},{"key":"2026051808350938000_2026.05.12.26353057v1.20","doi-asserted-by":"publisher","DOI":"10.1016\/j.ijmedinf.2025.105970"},{"key":"2026051808350938000_2026.05.12.26353057v1.21","doi-asserted-by":"publisher","DOI":"10.1007\/s41666-021-00113-8"},{"key":"2026051808350938000_2026.05.12.26353057v1.22","doi-asserted-by":"publisher","DOI":"10.1049\/iet-ipr.2019.0366"},{"key":"2026051808350938000_2026.05.12.26353057v1.23","doi-asserted-by":"publisher","DOI":"10.1056\/NEJMc2104626"},{"key":"2026051808350938000_2026.05.12.26353057v1.24","unstructured":"A. Kendall and Y. Gal , \u201cWhat uncertainties do we need in Bayesian deep learning for computer vision?,\u201d in Proceedings of the 31st International Conference on Neural Information Processing Systems, in NIPS\u201917. Red Hook, NY, USA: Curran Associates Inc., 2017, pp. 5580\u20135590."},{"key":"2026051808350938000_2026.05.12.26353057v1.25","doi-asserted-by":"publisher","DOI":"10.1038\/s41746-020-00367-3"},{"key":"2026051808350938000_2026.05.12.26353057v1.26","unstructured":"B. Lakshminarayanan , A. Pritzel , and C. Blundell , \u201cSimple and Scalable Predictive Uncertainty Estimation using Deep Ensembles,\u201d in Proceedings of the 31st International Conference on Neural Information Processing Systems, in NIPS\u201917. Red Hook, NY, USA: Curran Associates Inc., 2017, pp. 6405\u20136416."},{"key":"2026051808350938000_2026.05.12.26353057v1.27","unstructured":"T. Wang , et al., \u201cFrom Aleatoric to Epistemic: Exploring Uncertainty Quantification Techniques in Artificial Intelligence,\u201d Jan. 2025."},{"key":"2026051808350938000_2026.05.12.26353057v1.28","doi-asserted-by":"publisher","DOI":"10.1016\/j.metrad.2023.100003"},{"key":"2026051808350938000_2026.05.12.26353057v1.29","doi-asserted-by":"publisher","DOI":"10.1016\/j.jbi.2022.103996"},{"key":"2026051808350938000_2026.05.12.26353057v1.30","unstructured":"L. Peracchio et al., \u201cEvaluation of Predictive Reliability to Foster Trust in Artificial Intelligence. A case study in Multiple Sclerosis,\u201d Feb. 2024."},{"key":"2026051808350938000_2026.05.12.26353057v1.31","unstructured":"P. Schulam and S. Saria , \u201cCan You Trust This Prediction? Auditing Pointwise Reliability After Learning,\u201d Feb. 2019."},{"key":"2026051808350938000_2026.05.12.26353057v1.32","doi-asserted-by":"publisher","DOI":"10.1007\/b106715"},{"key":"2026051808350938000_2026.05.12.26353057v1.33","first-page":"371","article-title":"A Tutorial on Conformal Prediction","volume":"9","year":"2008","journal-title":"J. Mach. Learn. Res"},{"key":"2026051808350938000_2026.05.12.26353057v1.34","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v36i11.21459"},{"key":"2026051808350938000_2026.05.12.26353057v1.35","doi-asserted-by":"publisher","DOI":"10.1561\/2200000101"},{"key":"2026051808350938000_2026.05.12.26353057v1.36","unstructured":"\u201cClass-Conditional Conformal Prediction for Imbalanced Data via Top-k Classes,\u201d OpenReview. Accessed: Sep. 21, 2025. [Online]. Available: https:\/\/openreview.net\/pdf?id=Dtxc7mlKRg"},{"key":"2026051808350938000_2026.05.12.26353057v1.37","doi-asserted-by":"publisher","DOI":"10.1039\/C6TX00252H"},{"key":"2026051808350938000_2026.05.12.26353057v1.38","doi-asserted-by":"crossref","unstructured":"T. Ding , A. N. Angelopoulos , S. Bates , M. I. Jordan , and R. J. Tibshirani , \u201cClass-Conditional Conformai Prediction with Many Classes,\u201d in Proceedings of the 37th International Conference on Neural Information Processing Systems, in NIPS \u201923. Red Hook, NY, USA: Curran Associates Inc., 2023.","DOI":"10.52202\/075280-2817"},{"key":"2026051808350938000_2026.05.12.26353057v1.39","unstructured":"[39] Y. Shi , S. Ghosh , T. Belkhouja , J. R. Doppa , and Y. Yan , \u201cConformal Prediction for Class-wise Coverage via Augmented Label Rank Calibration,\u201d in Proceedings of the 38th International Conference on Neural Information Processing Systems, in NIPS \u201924. Red Hook, NY, USA: Curran Associates Inc., 2025."},{"key":"2026051808350938000_2026.05.12.26353057v1.40","doi-asserted-by":"publisher","DOI":"10.1038\/s41746-025-01616-z"},{"key":"2026051808350938000_2026.05.12.26353057v1.41","doi-asserted-by":"publisher","DOI":"10.1016\/j.neuroimage.2016.12.064"},{"key":"2026051808350938000_2026.05.12.26353057v1.42","unstructured":"\u201cIXI dataset (RRID:SCR_005839),\u201d Feb. 23, 2015, Imperial College London, Biomedical Image Analysis Group. Accessed: Sep. 21, 2025. [Online]. Available: http:\/\/brain-development.org\/ixi-dataset\/"},{"key":"2026051808350938000_2026.05.12.26353057v1.43","doi-asserted-by":"publisher","DOI":"10.17632\/8bctsm8jz7.1"},{"key":"2026051808350938000_2026.05.12.26353057v1.44","unstructured":"B. B. Avants , N. J. Tustison , and H. J. Johnson , \u201cAdvanced Normalization Tools (ANTs), v2.5.4,\u201d Nov. 26, 2024, GitHub \/ ANTs community: 2.5.4. Accessed: Sep. 22, 2025. [Online]. Available: https:\/\/stnava.github.io\/ANTs\/"},{"key":"2026051808350938000_2026.05.12.26353057v1.45","doi-asserted-by":"publisher","DOI":"10.1109\/TMI.2010.2046908"},{"key":"2026051808350938000_2026.05.12.26353057v1.46","doi-asserted-by":"publisher","DOI":"10.1101\/2021.02.10.430678"},{"key":"2026051808350938000_2026.05.12.26353057v1.47","doi-asserted-by":"publisher","DOI":"10.5281\/zenodo.13762311"},{"key":"2026051808350938000_2026.05.12.26353057v1.48","doi-asserted-by":"publisher","DOI":"10.1109\/CBMS.2009.5255454"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.05.12.26353057","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,5,18]],"date-time":"2026-05-18T15:35:19Z","timestamp":1779118519000},"score":1,"resource":{"primary":{"URL":"http:\/\/medrxiv.org\/lookup\/doi\/10.64898\/2026.05.12.26353057"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,5,15]]},"references-count":48,"URL":"https:\/\/doi.org\/10.64898\/2026.05.12.26353057","relation":{},"subject":[],"published":{"date-parts":[[2026,5,15]]},"subtype":"preprint"}}