Thomas Sanchez
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Thomas Sanchez has a total of 123 co-authors. Here are the top co-authors by number of shared publications:
- Meritxell Bach Cuadra
CIBM Center for Biomedical Imaging, Lausanne, Switzerland
11 shared publications - Elisenda Eixarch
BCNatal Fetal Medicine Research Center (Hospital Clínic and Hospital Sant Joan de Déu) Universitat de Barcelona Barcelona Spain
7 shared publications - Guillaume Auzias
Institut de Neurosciences de la Timone Aix-Marseille Université Marseille France
7 shared publications - Mériam Koob
Department of Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland
7 shared publications - Gerard Martí-Juan
BCN MedTech, Department of Engineering, Universitat Pompeu Fabra, Barcelona, Spain
6 shared publications - Vincent Dunet
Department of Diagnostic and Interventional Radiology, Lausanne University Hospital and University of Lausanne, 1011 Lausanne, Switzerland
6 shared publications
Latest Publications
- Open Access
- 14 August 2026
- Published by MIT Press in Imaging Neuroscience
- Vol. 4
Data quality biases normative models derived from fetal brain MRI
Normative modeling is increasingly used to characterize typical growth trajectories and identify atypical neurodevelopment, including early brain development using magnetic resonance imaging (MRI) acquired before birth. Recent work has emphasized the importance of large sample sizes for accurate and robust centile estimation. In this study, we investigate how image quality influences fetal brain normative models, a critical factor in this context where MRI is acquired on a moving fetus in utero. Using a multi-centric cohort of 635 fetal MRI scans, we applied a standardized visual quality control (QC) protocol with continuous quality ratings. We fit normative models for multiple brain structures under progressively relaxed QC stringency, and quantified the deviations in centile estimates relative to a high-quality reference subgroup. Our results showed that including lower-quality data systematically biased normative centiles, with the strongest effects observed in the outer centiles, particularly the lower tail (1st–10th). Bias increased progressively as QC stringency was relaxed and could not be attributed solely to the number of scans used to fit the models. Quality-induced bias was structure dependent, and often not visually apparent at the segmentation level. These findings highlight that image quality is an important source of bias in normative fetal brain modeling, and that increasing sample size at the expense of quality may systematically affect centile estimates, potentially jeopardizing the utility of the model.
- 14 August 2026
- Published by MIT Press in Imaging Neuroscience
- Vol. 4
Open AccessData quality biases normative models derived from fetal brain MRI
Normative modeling is increasingly used to characterize typical growth trajectories and identify atypical neurodevelopment, including early brain development using magnetic resonance imaging (MRI) acquired before birth. Recent work has emphasized the importance of large sample sizes for accurate and robust centile estimation. In this study, we investigate how image quality influences fetal brain normative models, a critical factor in this context where MRI is acquired on a moving fetus in utero. Using a multi-centric cohort of 635 fetal MRI scans, we applied a standardized visual quality control (QC) protocol with continuous quality ratings. We fit normative models for multiple brain structures under progressively relaxed QC stringency, and quantified the deviations in centile estimates relative to a high-quality reference subgroup. Our results showed that including lower-quality data systematically biased normative centiles, with the strongest effects observed in the outer centiles, particularly the lower tail (1st–10th). Bias increased progressively as QC stringency was relaxed and could not be attributed solely to the number of scans used to fit the models. Quality-induced bias was structure dependent, and often not visually apparent at the segmentation level. These findings highlight that image quality is an important source of bias in normative fetal brain modeling, and that increasing sample size at the expense of quality may systematically affect centile estimates, potentially jeopardizing the utility of the model.
- 1 July 2026
- Published by Machine Learning for Biomedical Imaging in Machine Learning for Biomedical Imaging
- Vol. 2026 (July 2026) , 482-506
Evaluating Synthetic Data Generation for Domain Generalization in Fetal Brain MRI Segmentation
Fetal brain tissue segmentation from magnetic resonance imaging (MRI) is crucial for studying neurodevelopment, but remains challenging due to data heterogeneity and limited annotations. Domain randomization (DR) has recently emerged as a promising strategy for single-source domain generalization by synthesizing training images with randomized artifacts, contrast, and resolution. In this work, we investigate how to maximize the out-of-domain (OOD) generalization of DR-based methods using FetalSynthSeg as a case study. We show that simple Gaussian mixture-based intensity modeling outperforms more complex physics-based simulations and that intensity clustering (subdividing tissue classes by intensity) substantially improves OOD robustness. Evaluated on 348 fetal subjects from four sites spanning 0.55–3T and both T1w and T2w contrasts, FetalSynthSeg reaches state-of-the-art performance on several FeTA 2024 testing datasets (80–85 Dice score) and, for the first time, offers robust segmentation on modalities other than T2w for fetal brain segmentation (80 Dice on dHCP-T1w dataset). Compared with state-of-the-art methods such as BOUNTI, nnU-Net ensemble, and the FeTA 2024 winner, FetalSynthSeg delivers comparable or superior accuracy while maintaining strong robustness across domain shifts. Our code, model weights, and Docker image ready for easy inference are available at https://hub.docker.com/r/vzalevskyi/fetalsynthseg
- 1 July 2026
- Published by Machine Learning for Biomedical Imaging in Machine Learning for Biomedical Imaging
- Vol. 2026 (July 2026) , 482-506
Evaluating Synthetic Data Generation for Domain Generalization in Fetal Brain MRI Segmentation
Fetal brain tissue segmentation from magnetic resonance imaging (MRI) is crucial for studying neurodevelopment, but remains challenging due to data heterogeneity and limited annotations. Domain randomization (DR) has recently emerged as a promising strategy for single-source domain generalization by synthesizing training images with randomized artifacts, contrast, and resolution. In this work, we investigate how to maximize the out-of-domain (OOD) generalization of DR-based methods using FetalSynthSeg as a case study. We show that simple Gaussian mixture-based intensity modeling outperforms more complex physics-based simulations and that intensity clustering (subdividing tissue classes by intensity) substantially improves OOD robustness. Evaluated on 348 fetal subjects from four sites spanning 0.55–3T and both T1w and T2w contrasts, FetalSynthSeg reaches state-of-the-art performance on several FeTA 2024 testing datasets (80–85 Dice score) and, for the first time, offers robust segmentation on modalities other than T2w for fetal brain segmentation (80 Dice on dHCP-T1w dataset). Compared with state-of-the-art methods such as BOUNTI, nnU-Net ensemble, and the FeTA 2024 winner, FetalSynthSeg delivers comparable or superior accuracy while maintaining strong robustness across domain shifts. Our code, model weights, and Docker image ready for easy inference are available at https://hub.docker.com/r/vzalevskyi/fetalsynthseg
- Open Access
- 2 June 2026
- Published by The Open Journal in Journal of Open Source Software
- Vol. 11 (122)
Fetpype: An Open-Source Pipeline for Reproducible Fetal Brain MRI Analysis
- 2 June 2026
- Published by The Open Journal in Journal of Open Source Software
- Vol. 11 (122)
Open AccessFetpype: An Open-Source Pipeline for Reproducible Fetal Brain MRI Analysis
- Open Access
Data quality biases normative models derived from fetal brain MRI
Normative modeling is increasingly used to characterize typical growth trajectories and identify atypical neurodevelopment, including early brain development using magnetic resonance imaging (MRI) acquired before birth. Recent work has emphasized the importance of large sample sizes for accurate and robust centile estimation. In this study, we investigate how image quality influences fetal brain normative models, a critical factor in this context where MRI is acquired on a moving fetus in utero. Using a multi-centric cohort of 635 fetal MRI scans, we applied a standardized visual quality control (QC) protocol with continuous quality ratings. We fit normative models for multiple brain structures under progressively relaxed QC stringency, and quantified the deviations in centile estimates relative to a high-quality reference subgroup. Our results showed that including lower-quality data systematically biased normative centiles, with the strongest effects observed in the outer centiles, particularly the lower tail (1st-10th). Bias increased progressively as QC stringency was relaxed and could not be attributed solely to the number of subjects used to fit the models. Quality-induced bias was structure-dependent, and often not visually apparent at the segmentation level. These findings highlight that image quality is an important source of bias in normative fetal brain modeling, and that increasing sample size at the expense of quality may systematically affect centile estimates, potentially jeopardizing the utility of the model.
Open AccessData quality biases normative models derived from fetal brain MRI
Normative modeling is increasingly used to characterize typical growth trajectories and identify atypical neurodevelopment, including early brain development using magnetic resonance imaging (MRI) acquired before birth. Recent work has emphasized the importance of large sample sizes for accurate and robust centile estimation. In this study, we investigate how image quality influences fetal brain normative models, a critical factor in this context where MRI is acquired on a moving fetus in utero. Using a multi-centric cohort of 635 fetal MRI scans, we applied a standardized visual quality control (QC) protocol with continuous quality ratings. We fit normative models for multiple brain structures under progressively relaxed QC stringency, and quantified the deviations in centile estimates relative to a high-quality reference subgroup. Our results showed that including lower-quality data systematically biased normative centiles, with the strongest effects observed in the outer centiles, particularly the lower tail (1st-10th). Bias increased progressively as QC stringency was relaxed and could not be attributed solely to the number of subjects used to fit the models. Quality-induced bias was structure-dependent, and often not visually apparent at the segmentation level. These findings highlight that image quality is an important source of bias in normative fetal brain modeling, and that increasing sample size at the expense of quality may systematically affect centile estimates, potentially jeopardizing the utility of the model.
- Open Access
- 16 January 2026
- Published by Elsevier in Medical Image Analysis
- Vol. 109, 103941
Advances in automated fetal brain MRI segmentation and biometry: Insights from the FeTA 2024 challenge
- 16 January 2026
- Published by Elsevier in Medical Image Analysis
- Vol. 109, 103941
Open AccessAdvances in automated fetal brain MRI segmentation and biometry: Insights from the FeTA 2024 challenge
- Open Access
Accounting for Underspecification in Statistical Claims of Model Superiority
Machine learning methods are increasingly applied in medical imaging, yet many reported improvements lack statistical robustness: recent works have highlighted that small but significant performance gains are highly likely to be false positives. However, these analyses do not take \emph{underspecification} into account -- the fact that models achieving similar validation scores may behave differently on unseen data due to random initialization or training dynamics. Here, we extend a recent statistical framework modeling false outperformance claims to include underspecification as an additional variance component. Our simulations demonstrate that even modest seed variability ($\sim1\%$) substantially increases the evidence required to support superiority claims. Our findings underscore the need for explicit modeling of training variance when validating medical imaging systems.
Open AccessAccounting for Underspecification in Statistical Claims of Model Superiority
Machine learning methods are increasingly applied in medical imaging, yet many reported improvements lack statistical robustness: recent works have highlighted that small but significant performance gains are highly likely to be false positives. However, these analyses do not take \emph{underspecification} into account -- the fact that models achieving similar validation scores may behave differently on unseen data due to random initialization or training dynamics. Here, we extend a recent statistical framework modeling false outperformance claims to include underspecification as an additional variance component. Our simulations demonstrate that even modest seed variability ($\sim1\%$) substantially increases the evidence required to support superiority claims. Our findings underscore the need for explicit modeling of training variance when validating medical imaging systems.
- 23 September 2025
- Published by Springer Nature
- p. 3-14
Automatic Quality Control in Multi-centric Fetal Brain MRI Super-Resolution Reconstruction
- 23 September 2025
- Published by Springer Nature
- p. 3-14
Automatic Quality Control in Multi-centric Fetal Brain MRI Super-Resolution Reconstruction
- Open Access
- 8 August 2025
- Published by Springer Nature in Pediatric Radiology
- Vol. 55 (10) , 2064-2075
Biometry and volumetry in multi-centric fetal brain magnetic resonance imaging: assessing the bias of super-resolution reconstruction
Background: Fetal brain MRI is increasingly used to complement ultrasound imaging. Images are processed using complex super-resolution reconstruction pipelines, which may bias biometric and volumetric measurements. Objective: To assess the consistency of 2-dimensional (D) biometric and 3-D volumetric measurements across three hospitals using three widely used super-resolution reconstruction pipelines. Materials and methods: This retrospective multi-centric study used T2-weighted fetal brain MRI scans acquired at three hospitals between 2009 and 2023. MRIs from each subject were reconstructed with each super-resolution reconstruction method, and biometric measurements were performed by four experts. Automated 3-D volumetry was performed using a state-of-the-art segmentation method. A qualitative evaluation assessed the clinicians’ likelihood of using super-resolution reconstructed volumes in their practice. Results: Eighty-four healthy subjects were included. Biometric measurements revealed statistically significant changes that consistently remained below voxel width (0.8 mm; PPP=0.02) and sharpness of the image (P=0.01). Conclusion: Variations in 2-D and 3-D quantitative measurements did not show any large systematic bias when using different super-resolution reconstruction methods for clinical radiological assessment across centers, scanners, and raters.
- 8 August 2025
- Published by Springer Nature in Pediatric Radiology
- Vol. 55 (10) , 2064-2075
Open AccessBiometry and volumetry in multi-centric fetal brain magnetic resonance imaging: assessing the bias of super-resolution reconstruction
Background: Fetal brain MRI is increasingly used to complement ultrasound imaging. Images are processed using complex super-resolution reconstruction pipelines, which may bias biometric and volumetric measurements. Objective: To assess the consistency of 2-dimensional (D) biometric and 3-D volumetric measurements across three hospitals using three widely used super-resolution reconstruction pipelines. Materials and methods: This retrospective multi-centric study used T2-weighted fetal brain MRI scans acquired at three hospitals between 2009 and 2023. MRIs from each subject were reconstructed with each super-resolution reconstruction method, and biometric measurements were performed by four experts. Automated 3-D volumetry was performed using a state-of-the-art segmentation method. A qualitative evaluation assessed the clinicians’ likelihood of using super-resolution reconstructed volumes in their practice. Results: Eighty-four healthy subjects were included. Biometric measurements revealed statistically significant changes that consistently remained below voxel width (0.8 mm; PPP=0.02) and sharpness of the image (P=0.01). Conclusion: Variations in 2-D and 3-D quantitative measurements did not show any large systematic bias when using different super-resolution reconstruction methods for clinical radiological assessment across centers, scanners, and raters.
- Open Access
Defacing biases visual quality assessments of structural MRI
A critical step before data-sharing of human neuroimaging is removing facial features to protect individuals’ privacy. However, not only does this process redact identifiable information about individuals, but it also removes non-identifiable information. This introduces undesired variability into downstream analysis and interpretation. This registered report investigated the degree to which the so-called defacing altered the quality assessment of T1-weighted images of the human brain from the openly available “IXI dataset”. The effect of defacing on manual quality assessment was investigated on a single-site subset of the dataset (N=185). By comparing two linear mixed-effects models, we determined that four trained human raters’ perception of quality was significantly influenced by defacing by modeling their ratings on the same set of images in two conditions: “nondefaced” (i.e., preserving facial features) and “defaced”. In addition, we investigated these biases on automated quality assessments by applying repeated-measures, multivariate ANOVA (rm-MANOVA) on the image quality metrics extracted with MRIQC on the full IXI dataset (N=581; three acquisition sites). This study found that defacing altered the quality assessments by humans and showed that MRIQC ’s quality metrics were mostly insensitive to defacing.
Open AccessDefacing biases visual quality assessments of structural MRI
A critical step before data-sharing of human neuroimaging is removing facial features to protect individuals’ privacy. However, not only does this process redact identifiable information about individuals, but it also removes non-identifiable information. This introduces undesired variability into downstream analysis and interpretation. This registered report investigated the degree to which the so-called defacing altered the quality assessment of T1-weighted images of the human brain from the openly available “IXI dataset”. The effect of defacing on manual quality assessment was investigated on a single-site subset of the dataset (N=185). By comparing two linear mixed-effects models, we determined that four trained human raters’ perception of quality was significantly influenced by defacing by modeling their ratings on the same set of images in two conditions: “nondefaced” (i.e., preserving facial features) and “defaced”. In addition, we investigated these biases on automated quality assessments by applying repeated-measures, multivariate ANOVA (rm-MANOVA) on the image quality metrics extracted with MRIQC on the full IXI dataset (N=581; three acquisition sites). This study found that defacing altered the quality assessments by humans and showed that MRIQC ’s quality metrics were mostly insensitive to defacing.
- 10 October 2024
- Published by Springer Nature
- p. 46-56
Assessing Data Quality on Fetal Brain MRI Reconstruction: A Multi-site and Multi-rater Study
- 10 October 2024
- Published by Springer Nature
- p. 46-56
Assessing Data Quality on Fetal Brain MRI Reconstruction: A Multi-site and Multi-rater Study