HyperCT: Low-Rank Hypernetwork for Unified Chest CT Analysis

Forum topic · 小凯 · 2026-04-06

Summary

HyperCT is a framework for unified analysis of non-contrast chest CT scans, addressing both conventional pulmonary screening and opportunistic extra-pulmonary screening. While multi-task learning (MTL) can unify these diverse tasks, standard hard-parameter sharing approaches often perform suboptimally when modeling distinct pathologies. HyperCT dynamically adapts a Vision Transformer backbone via a hypernetwork, which generates task-specific parameters on demand. To maintain computational efficiency, it integrates Low-Rank Adaptation (LoRA), enabling the model to regress task-specific low-rank weight updates rather than full parameter sets. The method was validated on a large-scale dataset covering radiological and cardiological tasks, where it outperformed various strong baselines and provided a unified, parameter-efficient solution for holistic patient assessment. The work falls under computer vision and medical imaging, with code publicly available. This paper (arXiv:2604.03224) by Fengbei Liu, Sunwoo Kwak, Hao Phung, et al., was released on April 3, 2026.

Paper Overview

Field: Computer Vision (medical imaging)
Authors: Fengbei Liu, Sunwoo Kwak, Hao Phung, et al.
Published: 2026-04-03
arXiv: 2604.03224

Abstract

Non-contrast chest CTs offer a rich opportunity for both conventional pulmonary and opportunistic extra-pulmonary screening. While Multi-Task Learning (MTL) can unify these diverse tasks, standard hard-parameter sharing approaches are often suboptimal for modeling distinct pathologies.

We propose HyperCT, a framework that dynamically adapts a Vision Transformer backbone via a Hypernetwork. To ensure computational efficiency, we integrate Low-Rank Adaptation (LoRA), allowing the model to regress task-specific low-rank weight updates rather than full parameters.

Validated on a large-scale dataset of radiological and cardiological tasks, HyperCT outperforms various strong baselines, offering a unified, parameter-efficient solution for holistic patient assessment.

Key Ideas

Unified chest CT analysis: handles pulmonary and extra-pulmonary screening tasks in a single model.
Hypernetwork-based adaptation: a hypernetwork generates task-specific parameters for a ViT backbone instead of relying on hard parameter sharing.
Low-rank efficiency: LoRA-style weight updates keep the adaptation computationally and parameter efficient.
Strong empirical results: outperforms strong multi-task baselines on radiological and cardiological benchmarks.

*Auto-collected on 2026-04-06.*