David Kügler
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
David Kügler has a total of 89 co-authors. Here are the top co-authors by number of shared publications:
- Martin Reuter
German Center for Neurodegenerative Diseases (DZNE)
5 shared publications Anirban Mukhopadhyay
4 shared publicationsIgor Stenin
ENT Clinic, University Düsseldorf, Düsseldorf, Germany
3 shared publicationsJulia Kristin
ENT Clinic, University Düsseldorf, Düsseldorf, Germany
3 shared publicationsJörg Schipper
ENT Clinic, University Düsseldorf, Düsseldorf, Germany
3 shared publicationsThomas Klenzner
ENT Clinic, University Düsseldorf, Düsseldorf, Germany
3 shared publications
Latest Publications
- Open Access
Age and sex differences in lateral ventricle choroid plexus volume in a large, manually segmented, multi-site MRI dataset
Background: The choroid plexus, which produces cerebrospinal fluid and forms the blood–cerebrospinal fluid barrier, is of growing interest in psychiatric and neurodegenerative neuroimaging. However, most volumetric studies have relied on automated segmentation tools that agree poorly with manual tracing, and manually segmented samples have been small, single-site, and narrow in age range. Methods: We analyzed DataPlex, a dataset of 700 manually segmented lateral ventricle choroid plexus labels from 3T T1-weighted MRI scans (ages 6–90 years) drawn from 29 studies and three scanner manufacturers (35 study-by-manufacturer sites). Generalized least squares models accounting for within-site dependence tested associations of age, sex, scanner manufacturer, image resolution, a registration-based image-quality metric, and intracranial volume with choroid plexus and lateral ventricle volumes, including nonlinear (spline) and sex-interaction terms. Exploratory models tested associations between these volumes and cognitive performance. Results: In 699 scans, choroid plexus volume increased linearly with age (63–69 mm³ per decade) and was 56–65 mm³ larger in males than females after covariate adjustment; the age association did not differ by sex. Lateral ventricle volume increased nonlinearly with age, with steeper age-related enlargement in males than females. Scanner manufacturer and image resolution were associated with left lateral ventricle volume; scanner manufacturer was associated with choroid plexus volume only in a sensitivity model. The image-quality metric was not associated with any outcome. In exploratory analyses, larger choroid plexus volume was associated with lower inhibitory control, working memory, and episodic memory scores and with higher emotion recognition scores (uncorrected p < 0.05). Conclusions: In the largest and most diverse manually segmented choroid plexus dataset to date, choroid plexus volume differed by age and sex and was less sensitive than lateral ventricle volume to acquisition differences. These estimates provide reference benchmarks for automated segmentation tools and support adjustment for demographic and technical covariates.
Open AccessAge and sex differences in lateral ventricle choroid plexus volume in a large, manually segmented, multi-site MRI dataset
Background: The choroid plexus, which produces cerebrospinal fluid and forms the blood–cerebrospinal fluid barrier, is of growing interest in psychiatric and neurodegenerative neuroimaging. However, most volumetric studies have relied on automated segmentation tools that agree poorly with manual tracing, and manually segmented samples have been small, single-site, and narrow in age range. Methods: We analyzed DataPlex, a dataset of 700 manually segmented lateral ventricle choroid plexus labels from 3T T1-weighted MRI scans (ages 6–90 years) drawn from 29 studies and three scanner manufacturers (35 study-by-manufacturer sites). Generalized least squares models accounting for within-site dependence tested associations of age, sex, scanner manufacturer, image resolution, a registration-based image-quality metric, and intracranial volume with choroid plexus and lateral ventricle volumes, including nonlinear (spline) and sex-interaction terms. Exploratory models tested associations between these volumes and cognitive performance. Results: In 699 scans, choroid plexus volume increased linearly with age (63–69 mm³ per decade) and was 56–65 mm³ larger in males than females after covariate adjustment; the age association did not differ by sex. Lateral ventricle volume increased nonlinearly with age, with steeper age-related enlargement in males than females. Scanner manufacturer and image resolution were associated with left lateral ventricle volume; scanner manufacturer was associated with choroid plexus volume only in a sensitivity model. The image-quality metric was not associated with any outcome. In exploratory analyses, larger choroid plexus volume was associated with lower inhibitory control, working memory, and episodic memory scores and with higher emotion recognition scores (uncorrected p < 0.05). Conclusions: In the largest and most diverse manually segmented choroid plexus dataset to date, choroid plexus volume differed by age and sex and was less sensitive than lateral ventricle volume to acquisition differences. These estimates provide reference benchmarks for automated segmentation tools and support adjustment for demographic and technical covariates.
- 12 September 2026
- Published by Springer Nature
- p. 498-517
Rethinking Real-World MRI Denoising: Learning from Physical Noise
- 12 September 2026
- Published by Springer Nature
- p. 498-517
Rethinking Real-World MRI Denoising: Learning from Physical Noise
- Open Access
- 2 June 2026
- Published by MIT Press in Imaging Neuroscience
- Vol. 4
Learning-based segmentation of diffusion-weighted MR images with arbitrary q -space samplings
Segmenting anatomical regions is a crucial step in many diffusion-weighted MRI (dMRI) workflows, such as region-of-interest analysis or anatomically-constrained tractography, which enable in vivo studies of brain microstructure and connectivity. However, convolutional neural networks (CNNs)—the foundation of most state-of-the-art segmentation models—require structured inputs with a fixed number of channels. This makes them ill-suited for dMRI, where acquisition protocols vary widely in q-space sampling—the number of measurements as well as their directions (b-vectors) and weightings (b-values)—resulting in unstructured data with inconsistent dimensionality across studies. As a consequence, the applicability of CNN-based methods is generally limited to the dataset on which they were trained. To address this, existing methods like DeepAnat and DDParcel rely on diffusion model fits, such as the diffusion tensor, to convert raw data into structured representations compatible with CNNs. While this enables broader applicability, it introduces lossy compression that can degrade performance. In this work, we propose a novel method that combines the geometric deep learning-based reconstruction framework DISCUS with the segmentation network VINN to directly map unstructured dMRI data to anatomical segmentations. Our segmentation approach is the first to achieve robust generalization across heterogeneous acquisition schemes using a single neural network without requiring diffusion model fits. Our approach generates the segmentation in minutes, whereas DeepAnat relies on the external FreeSurfer software, which runs for several hours. Additionally, we demonstrate generally superior segmentation performance of our approach across multiple datasets and acquisition settings with respect to DeepAnat, DDParcel, and SynthSeg.
- 2 June 2026
- Published by MIT Press in Imaging Neuroscience
- Vol. 4
Open AccessLearning-based segmentation of diffusion-weighted MR images with arbitrary q -space samplings
Segmenting anatomical regions is a crucial step in many diffusion-weighted MRI (dMRI) workflows, such as region-of-interest analysis or anatomically-constrained tractography, which enable in vivo studies of brain microstructure and connectivity. However, convolutional neural networks (CNNs)—the foundation of most state-of-the-art segmentation models—require structured inputs with a fixed number of channels. This makes them ill-suited for dMRI, where acquisition protocols vary widely in q-space sampling—the number of measurements as well as their directions (b-vectors) and weightings (b-values)—resulting in unstructured data with inconsistent dimensionality across studies. As a consequence, the applicability of CNN-based methods is generally limited to the dataset on which they were trained. To address this, existing methods like DeepAnat and DDParcel rely on diffusion model fits, such as the diffusion tensor, to convert raw data into structured representations compatible with CNNs. While this enables broader applicability, it introduces lossy compression that can degrade performance. In this work, we propose a novel method that combines the geometric deep learning-based reconstruction framework DISCUS with the segmentation network VINN to directly map unstructured dMRI data to anatomical segmentations. Our segmentation approach is the first to achieve robust generalization across heterogeneous acquisition schemes using a single neural network without requiring diffusion model fits. Our approach generates the segmentation in minutes, whereas DeepAnat relies on the external FreeSurfer software, which runs for several hours. Additionally, we demonstrate generally superior segmentation performance of our approach across multiple datasets and acquisition settings with respect to DeepAnat, DDParcel, and SynthSeg.
- Top Cited PapersOpen Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 45 (5) , 2156-2172
Regression Is All You Need for Medical Image Translation
While Generative Adversarial Nets (GANs) and Diffusion Models (DMs) have achieved impressive results in natural image synthesis, their core strengths – creativity and realism – can be detrimental in medical applications, where accuracy and fidelity are paramount. These models instead risk introducing hallucinations and replication of unwanted acquisition noise. Here, we propose YODA (You Only Denoise once – or Average), a 2.5D diffusion-based framework for medical image translation (MIT). Consistent with DM theory, we find that conventional diffusion sampling stochastically replicates noise. To mitigate this, we draw and average multiple samples, akin to physical signal averaging. As this effectively approximates the DM’s expected value, we term this Expectation-Approximation (ExpA) sampling. We additionally propose regression sampling YODA, which retains the initial DM prediction and omits iterative refinement to produce noise-free images in a single step. Across five diverse multi-modal datasets – including multi-contrast brain MRI and pelvic MRI-CT – we demonstrate that regression sampling is not only substantially more efficient but also matches or exceeds image quality of full diffusion sampling even with ExpA. Our results reveal that iterative refinement solely enhances perceptual realism without benefiting information translation, which we confirm in relevant downstream tasks. YODA outperforms eight state-of-the-art DMs and GANs and challenges the presumed superiority of DMs and GANs over computationally cheap regression models for high-quality MIT. Furthermore, we show that YODA-translated images are interchangeable with, or even superior to, physical acquisitions for several medical applications.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 45 (5) , 2156-2172
Top Cited PapersOpen AccessRegression Is All You Need for Medical Image Translation
While Generative Adversarial Nets (GANs) and Diffusion Models (DMs) have achieved impressive results in natural image synthesis, their core strengths – creativity and realism – can be detrimental in medical applications, where accuracy and fidelity are paramount. These models instead risk introducing hallucinations and replication of unwanted acquisition noise. Here, we propose YODA (You Only Denoise once – or Average), a 2.5D diffusion-based framework for medical image translation (MIT). Consistent with DM theory, we find that conventional diffusion sampling stochastically replicates noise. To mitigate this, we draw and average multiple samples, akin to physical signal averaging. As this effectively approximates the DM’s expected value, we term this Expectation-Approximation (ExpA) sampling. We additionally propose regression sampling YODA, which retains the initial DM prediction and omits iterative refinement to produce noise-free images in a single step. Across five diverse multi-modal datasets – including multi-contrast brain MRI and pelvic MRI-CT – we demonstrate that regression sampling is not only substantially more efficient but also matches or exceeds image quality of full diffusion sampling even with ExpA. Our results reveal that iterative refinement solely enhances perceptual realism without benefiting information translation, which we confirm in relevant downstream tasks. YODA outperforms eight state-of-the-art DMs and GANs and challenges the presumed superiority of DMs and GANs over computationally cheap regression models for high-quality MIT. Furthermore, we show that YODA-translated images are interchangeable with, or even superior to, physical acquisitions for several medical applications.
- Open Access
- 23 December 2025
- Published by Elsevier in EBioMedicine
- Vol. 123, 106090
Brain atrophy staging in spinocerebellar ataxia type 3 for clinical prognosis and trial enrichment
- 23 December 2025
- Published by Elsevier in EBioMedicine
- Vol. 123, 106090
Open AccessBrain atrophy staging in spinocerebellar ataxia type 3 for clinical prognosis and trial enrichment
- Open Access
- 18 May 2023
- Published by Elsevier in NeuroImage
- Vol. 275, 120176
Quantifying MR head motion in the Rhineland Study – A robust method for population cohorts
- 18 May 2023
- Published by Elsevier in NeuroImage
- Vol. 275, 120176
Open AccessQuantifying MR head motion in the Rhineland Study – A robust method for population cohorts
- Open Access
- 27 October 2022
- Published by Elsevier in NeuroImage
- Vol. 264, 119703
CerebNet: A fast and reliable deep-learning pipeline for detailed cerebellum sub-segmentation
- 27 October 2022
- Published by Elsevier in NeuroImage
- Vol. 264, 119703
Open AccessCerebNet: A fast and reliable deep-learning pipeline for detailed cerebellum sub-segmentation
- Open Access
- 29 September 2020
- Published by Springer Nature
- p. 375-384
AutoSNAP: Automatically Learning Neural Architectures for Instrument Pose Estimation
- 29 September 2020
- Published by Springer Nature
- p. 375-384
Open AccessAutoSNAP: Automatically Learning Neural Architectures for Instrument Pose Estimation
- Open Access
- 21 May 2020
- Published by Springer Nature in International Journal of Computer Assisted Radiology and Surgery
- Vol. 15 (7) , 1137-1145
i3PosNet: instrument pose estimation from X-ray in temporal bone surgery
Purpose: Accurate estimation of the position and orientation (pose) of surgical instruments is crucial for delicate minimally invasive temporal bone surgery. Current techniques lack in accuracy and/or line-of-sight constraints (conventional tracking systems) or expose the patient to prohibitive ionizing radiation (intra-operative CT). A possible solution is to capture the instrument with a c-arm at irregular intervals and recover the pose from the image. Methods: i3PosNet infers the position and orientation of instruments from images using a pose estimation network. Said framework considers localized patches and outputs pseudo-landmarks. The pose is reconstructed from pseudo-landmarks by geometric considerations. Results: We show i3PosNet reaches errors $$<\,0.05$$ mm. It outperforms conventional image registration-based approaches reducing average and maximum errors by at least two thirds. i3PosNet trained on synthetic images generalizes to real X-rays without any further adaptation. Conclusion: The translation of deep learning-based methods to surgical applications is difficult, because large representative datasets for training and testing are not available. This work empirically shows sub-millimeter pose estimation trained solely based on synthetic training data.- 21 May 2020
- Published by Springer Nature in International Journal of Computer Assisted Radiology and Surgery
- Vol. 15 (7) , 1137-1145
Open Accessi3PosNet: instrument pose estimation from X-ray in temporal bone surgery
Purpose: Accurate estimation of the position and orientation (pose) of surgical instruments is crucial for delicate minimally invasive temporal bone surgery. Current techniques lack in accuracy and/or line-of-sight constraints (conventional tracking systems) or expose the patient to prohibitive ionizing radiation (intra-operative CT). A possible solution is to capture the instrument with a c-arm at irregular intervals and recover the pose from the image. Methods: i3PosNet infers the position and orientation of instruments from images using a pose estimation network. Said framework considers localized patches and outputs pseudo-landmarks. The pose is reconstructed from pseudo-landmarks by geometric considerations. Results: We show i3PosNet reaches errors $$<\,0.05$$ mm. It outperforms conventional image registration-based approaches reducing average and maximum errors by at least two thirds. i3PosNet trained on synthetic images generalizes to real X-rays without any further adaptation. Conclusion: The translation of deep learning-based methods to surgical applications is difficult, because large representative datasets for training and testing are not available. This work empirically shows sub-millimeter pose estimation trained solely based on synthetic training data. - Open Access
i3PosNet: Instrument Pose Estimation from X-Ray in temporal bone surgery
Purpose: Accurate estimation of the position and orientation (pose) of surgical instruments is crucial for delicate minimally invasive temporal bone surgery. Current techniques lack in accuracy and/or line-of-sight constraints (conventional tracking systems) or expose the patient to prohibitive ionizing radiation (intra-operative CT). A possible solution is to capture the instrument with a c-arm at irregular intervals and recover the pose from the image. Methods: i3PosNet infers the position and orientation of instruments from images using a pose estimation network. Said framework considers localized patches and outputs pseudo-landmarks. The pose is reconstructed from pseudo-landmarks by geometric considerations. Results: We show i3PosNet reaches errors less than 0.05mm. It outperforms conventional image registration-based approaches reducing average and maximum errors by at least two thirds. i3PosNet trained on synthetic images generalizes to real x-rays without any further adaptation. Conclusion: The translation of Deep Learning based methods to surgical applications is difficult, because large representative datasets for training and testing are not available. This work empirically shows sub-millimeter pose estimation trained solely based on synthetic training data.
Open Accessi3PosNet: Instrument Pose Estimation from X-Ray in temporal bone surgery
Purpose: Accurate estimation of the position and orientation (pose) of surgical instruments is crucial for delicate minimally invasive temporal bone surgery. Current techniques lack in accuracy and/or line-of-sight constraints (conventional tracking systems) or expose the patient to prohibitive ionizing radiation (intra-operative CT). A possible solution is to capture the instrument with a c-arm at irregular intervals and recover the pose from the image. Methods: i3PosNet infers the position and orientation of instruments from images using a pose estimation network. Said framework considers localized patches and outputs pseudo-landmarks. The pose is reconstructed from pseudo-landmarks by geometric considerations. Results: We show i3PosNet reaches errors less than 0.05mm. It outperforms conventional image registration-based approaches reducing average and maximum errors by at least two thirds. i3PosNet trained on synthetic images generalizes to real x-rays without any further adaptation. Conclusion: The translation of Deep Learning based methods to surgical applications is difficult, because large representative datasets for training and testing are not available. This work empirically shows sub-millimeter pose estimation trained solely based on synthetic training data.