Budget-Aware Uncertainty for Radiotherapy Segmentation QA Using nnU-Net

Forum topic · 小凯 · 2026-04-15

Summary

A new paper on arXiv (2604.11798) proposes a budget-aware, uncertainty-driven quality assurance framework for deep learning-based clinical target volume (CTV) segmentation in radiotherapy, built on nnU-Net. Accurate CTV delineation is essential for radiotherapy planning but remains time-consuming and hard to assess, especially for complex treatments such as Total Marrow and Lymph Node Irradiation (TMLI). The authors—Ricardo Coimbra Brioso and colleagues from Humanitas University and Politecnico di Milano—combine uncertainty quantification with post-hoc calibration to produce voxel-level uncertainty maps that flag where the model may be wrong. Experiments show that efficient ensembles combined with calibration are a promising strategy for budget-aware QA workflows, enabling safer clinical deployment of auto-segmentation.

Paper Overview

Research areas: cs.CV, cs.AI
Authors: Ricardo Coimbra Brioso, Lorenzo Mondo, Damiano Dei, Nicola Lambri, Pietro Mancosu, Marta Scorsetti, Daniele Loiacono
Published: 2026-04-13
arXiv: 2604.11798

Abstract

Accurate delineation of the Clinical Target Volume (CTV) is essential for radiotherapy planning, yet remains time-consuming and difficult to assess, especially for complex treatments such as Total Marrow and Lymph Node Irradiation (TMLI). While deep learning-based auto-segmentation can reduce workload, safe clinical deployment requires reliable cues indicating where models may be wrong.

Key Contributions

The paper proposes a budget-aware, uncertainty-driven quality assurance (QA) framework built on nnU-Net. It combines uncertainty quantification with post-hoc calibration to generate voxel-level uncertainty maps, allowing clinicians to focus manual review effort where the model is most likely to err.

Experiments indicate that efficient ensembles combined with calibration are a promising strategy for implementing budget-aware QA workflows in clinical practice.

---

*Auto-collected on 2026-04-15.*