EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Characterization of local linearity in deep learning CT denoising networks

Kwok Lung Fan, Junyuan Li, Craig Abbey, Cheng Ting Lin, J Webster Stayman, Grace J Gang

Medical Physics · 2026-11 · 53 (11)

DOI: 10.1002/mp.70756 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Medical Physics, online erschienen am 05.11.2026. Verfasst von Kwok Lung Fan und 5 weiteren Autor:innen. Beteiligte Einrichtung: Department of Radiology, University of Pennsylvania, Philadelphia, Pennsylvania, USA. Hauptthemen laut MEDLINE: Deep Learning, Tomography, X-Ray Computed, Signal-To-Noise Ratio, Image Processing, Computer-Assisted. Die Arbeit nennt 20 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NCI NIH HHS, National Institutes of Health.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Deep learning algorithms for image processing have been proposed for a wide range of applications. In particular, deep learning CT denoising has demonstrated superior ability to reduce noise compared to conventional methods. However, the nonlinear nature of these algorithms makes it difficult to inspect and visualize how the model accomplishes the denoising task.

Purpose

We propose a local linearization approach to characterize and analyze the behavior of deep learning algorithms. Example analysis was performed for a classic denoising architecture with convolutional neural network and ReLU activation functions, and was extended to GELU activation and a transformer‐based architecture.

Methods

We defined strictly local linear regions as the input space over which the activation patterns of all ReLU activation functions are the same. The extent of these regions was investigated by perturbing clinically relevant features of a lung lesion. We further defined approximately local linear regions as the input space where Jacobians are similar and linear approximation results in small error. We developed an algorithm to partition input space over a range of lesion and background features into approximately local linear regions, and separately quantified the contribution of the noise realization and anatomical background to this approximation error. The behavior of networks trained with different levels of denoising, as well as alternative architectures, was compared.

Results

The extent of strictly local linear regions was found to be very small. Perturbations on the scale of were enough to push the network to a neighboring strictly local linear region. However, the Jacobians over a large number of strictly local linear regions were found to be highly similar, which suggests that the network behaves approximately as a piecewise linear system. The partitioning algorithm identified a small number of approximately local linear regions within input spaces characterized by clinically relevant features; this number was similar for the GELU variant but increased for networks trained with more aggressive noise reduction and for a transformer‐based architecture, while Jacobians transferred across anatomical backgrounds also increased approximation error.

Conclusions

The local linearity analysis framework can be a useful tool for elucidating the behavior of neural networks that are traditionally considered as “black‐boxes.” Preliminary results suggest that the framework can be applied to other differentiable architectures, subject to empirical validation, to characterize these highly nonlinear, data‐dependent algorithms over input spaces of interest.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Deep LearningTomography, X-Ray ComputedSignal-To-Noise RatioImage Processing, Computer-AssistedConvolutional Neural NetworksAlgorithmsLinear ModelsHumans

Stichwörter der Autor:innen

CT denoisingDeep learning evaluationImage qualityJacobian analysisLocal linearityNeural Network interpretability

Autor:innen und Institutionen

  • Department of Radiology, University of Pennsylvania, Philadelphia, Pennsylvania, USA.
  • Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, New York, USA.
  • Department of Biomedical Engineering, Johns Hopkins University, Baltimore, Maryland, USA.
  • Department of Psychological and Brain Sciences, University of California Santa Barbara, Santa Barbara, California, USA.
  • Department of Radiology, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA.

Förderung

  • NCI NIH HHS
  • National Institutes of Health

Zitieren

APA

Fan, K. L., Li, J., Abbey, C., Lin, C. T., Stayman, J. W., & Gang, G. J. (2026). Characterization of local linearity in deep learning CT denoising networks. Medical Physics, 53(11). https://doi.org/10.1002/mp.70756

Vancouver

Fan KL, Li J, Abbey C, Lin CT, Stayman JW, Gang GJ. Characterization of local linearity in deep learning CT denoising networks. Medical Physics. 2026;53(11). doi:10.1002/mp.70756

BibTeX
@article{Fan2026,
  author = {Fan, Kwok Lung and Li, Junyuan and Abbey, Craig and Lin, Cheng Ting and Stayman, J Webster and Gang, Grace J},
  title = {Characterization of local linearity in deep learning CT denoising networks},
  journal = {Medical Physics},
  year = {2026},
  volume = {53},
  number = {11},
  doi = {10.1002/mp.70756}
}
Kostenlos starten →