Uncertainty-Aware Multiscale Treatment Response Forecasting in Advanced Non-Small Cell Lung Cancer Using Residual-Variance Conformal Prediction
- 1Department of Radiation Oncology, University of Washington, Seattle, WA
Abstract
Voxel-level prediction of treatment response from longitudinal Fluorodexoyglucose (FDG) positron emission tomography / computed tomography (PET/CT) can enable spatially adaptive dose planning in advanced non-small cell lung cancer (NSCLC), but point predictions alone lack the uncertainty quantification needed for safe clinical decision-making. Standard conformal prediction (CP) provides distribution-free coverage but yields spatially uniform intervals and assumes exchangeability--an assumption violated by intra-tumoral spatial correlation. We propose a multiscale CP framework built on a re-optimized VoxelForecast generalized least squares model that explicitly captures voxel-level covariance. Variogram sensitivity analysis identified the Stable model as significantly superior to previouly used Matern family. The framework combines hierarchical nested leave-one-patient-out calibration, which restores exchangeability for valid voxel-level coverage, with residual-variance CP (VarCP), which learns a heteroscedastic, location-dependent scale to produce adaptive intervals while preserving finite-sample marginal coverage guarantees. On two prospective NSCLC trial cohorts--FLARE-RT (N=25 locally advanced, chemoradio-therapy; 11,100 voxels) and BRIGHT (N=19 metastatic, chemoimmunotherapy; 26,980 voxels)--multiscale feature integration reduced voxel-level prediction error by 35% and 31%, respectively. VarCP maintained nominal coverage while producing 7-18% (FLARE-RT) and 4-8% (BRIGHT) narrower voxel-level intervals than a hierarchically calibrated standard CP baseline (paired Wilcoxon signed-rank, significant at most miscoverage levels), with comparable lesion-level gains (7-18% and 6-10%) via median-based aggregation of voxel residuals. VarCP delivers tighter yet mathematically guaranteed spatial uncertainty maps, supporting both voxel-level dose-adjustment and whole-lesion treatment planning in advanced NSCLC.
Related
- CP-BayesNE: Uncertainty-Aware Personalized Nutrient-Density Estimation from Food Images Using Conformal Prediction and Prior-Corrected Dietary Adaptation
- High performance white blood cell classification using ensemble CNNs with Monte Carlo dropout and conformal prediction for clinical uncertainty quantification
- Maternal–Fetal Oncology Intelligence: Self-Evolving Foundation Models and Digital Twin Ecosystems for Precision Cancer Prediction, Treatment Optimization, and Maternal–Fetal Outcome Forecasting
- A Cross-Dataset Audit of Discrimination and Conformal Prediction Coverage in Keystroke-Based Parkinson's Disease Classi cation
- Uncertainty Quantification for Aquatic Ecotoxicity Prediction
BibTeX
@misc{wang2026uncertaintyaware,
title = {{Uncertainty-Aware Multiscale Treatment Response Forecasting in Advanced Non-Small Cell Lung Cancer Using Residual-Variance Conformal Prediction}},
author = {S. Wang and L. Deng and T. Ye and P. E. Kinahan and J. Zeng and J. H. Gennari and S. R. Bowen},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.30.26364431},
url = {https://www.medrxiv.org/content/10.64898/2026.09.30.26364431v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Uncertainty-Aware Multiscale Treatment Response Forecasting in Advanced Non-Small Cell Lung Cancer Using Residual-Variance Conformal Prediction
AU - S. Wang
AU - L. Deng
AU - T. Ye
AU - P. E. Kinahan
AU - J. Zeng
AU - J. H. Gennari
AU - S. R. Bowen
PY - 2026
DA - 2026/09/30
PB - medRxiv
DO - 10.64898/2026.09.30.26364431
UR - https://www.medrxiv.org/content/10.64898/2026.09.30.26364431v1
AB - Voxel-level prediction of treatment response from longitudinal Fluorodexoyglucose (FDG) positron emission tomography / computed tomography (PET/CT) can enable spatially adaptive dose planning in advanced non-small cell lung cancer (NSCLC), but point predictions alone lack the uncertainty quantification needed for safe clinical decision-making. Standard conformal prediction (CP) provides distribution-free coverage but yields spatially uniform intervals and assumes exchangeability--an assumption violated by intra-tumoral spatial correlation. We propose a multiscale CP framework built on a re-optimized VoxelForecast generalized least squares model that explicitly captures voxel-level covariance. Variogram sensitivity analysis identified the Stable model as significantly superior to previouly used Matern family. The framework combines hierarchical nested leave-one-patient-out calibration, which restores exchangeability for valid voxel-level coverage, with residual-variance CP (VarCP), which learns a heteroscedastic, location-dependent scale to produce adaptive intervals while preserving finite-sample marginal coverage guarantees. On two prospective NSCLC trial cohorts--FLARE-RT (N=25 locally advanced, chemoradio-therapy; 11,100 voxels) and BRIGHT (N=19 metastatic, chemoimmunotherapy; 26,980 voxels)--multiscale feature integration reduced voxel-level prediction error by 35% and 31%, respectively. VarCP maintained nominal coverage while producing 7-18% (FLARE-RT) and 4-8% (BRIGHT) narrower voxel-level intervals than a hierarchically calibrated standard CP baseline (paired Wilcoxon signed-rank, significant at most miscoverage levels), with comparable lesion-level gains (7-18% and 6-10%) via median-based aggregation of voxel residuals. VarCP delivers tighter yet mathematically guaranteed spatial uncertainty maps, supporting both voxel-level dose-adjustment and whole-lesion treatment planning in advanced NSCLC.
ER -