Dr. Provins Céline

Highlights

Medical imaging expert with 5 years of experience in developing, testing, and optimizing scalable, high-performance, and modular data pipelines for MRI using Python, R, containers, HPC, and Git tooling, building on strong foundations in biomedical engineering, biomarker research, and mathematics. Appreciated for outstanding organizational and for working effectively both independently and in teams, demonstrating curiosity, initiative and autonomy in solving complex problems and ensuring a reliable and timely delivery of results. Recognized for excellent communication skills, whether it is to collaborate with a multidisciplinary team or to communicate the interpretation of complex medical data to a broad audience effectively.

My motto

Turning complex medical data into reliable insights

News

May 14, 2026	After years in the making, I am very happy to see our MRIQC protocol finally published! 🙌 Our goal with this paper was really to provide a complete and detailed step-by-step guide on how to perform quality control on your neuroimaging data, covering three MRI modalities: anatomical, functional and diffusion. Although the protocol is tailored to MRIQC, the broader concepts and philosophy are applicable far beyond a single tool, even a single domain: define clear criteria, inspect data systematically, document decisions, and catch quality issues early enough to act on them. My hope is that this protocol helps make QC a standard, rigorous, and thoroughly applied step in neuroimaging studies, rather than treated as thi slightly-annoying and too time-consuming step. Reliable results start with reliable data. A huge shout-out to McKenzie Paige Hagen for bringing it across the finish line!
Apr 07, 2026	Delighted to announce that I started a new position as Machine Learning Engineer at greenteg AG! 🥳 Greenteg is a leader in core body temperature monitoring, enabling cutting-edge applications in health tracking, sports performance, and work safety. I will bring my expertise in evaluating data quality, reliability, and implementing robust data pipelines to improve machine learning models predicting core body temperature from wearable sensor data. I will also support both clients and internal teams on data science initiatives. I am super happy to have the opportunity to collaborate with such a fantastic team and to continue pushing the boundaries of biomedical research. Hopefully, this is the start of a long and impactful collaboration 🚀
Mar 18, 2026	I completed the Machine Learning in Production certification by Andrew Ng! 🎓 The Machine Learning in Production course highlighted how building reliable ML systems requires far more than training models. It emphasized the importance of data-centric development, systematic error analysis, and treating ML as an iterative pipeline spanning scoping, data, modeling, and deployment. I also learned how to monitor models under data and concept drift, evaluate performance beyond average accuracy, and design robust ML pipelines. Through hands-on exercises, I deployed computer vision models with Docker and AWS and explored how dataset composition and labeling strategies affect performance. Coming from a background in medical imaging pipelines, these lessons reinforced the importance of data quality and engineering discipline in building trustworthy AI systems. I’m looking forward to continue applying these principles — and many of Andrew’s practical tips — when building robust and reproducible AI systems. Certificate Here!
Jan 23, 2026	🚀 WEF Hackathon 2026 – Turning Energy Market Data into Actionable Intelligence I participated in the WEF Hackathon 2026, where our team built an Energy Market Intelligence Agent to transform complex energy market time series into actionable insights for decision-makers. The system combined statistical signal analysis with AI agents to detect anomalies, link price movements to underlying drivers such as wind, solar, and consumption, and generate trader-oriented explanations and interactive visualizations. This experience highlighted how accessible AI tooling has become — and how the real challenge lies in understanding business needs and designing systems that support better decisions.
Dec 10, 2025	Presented two posters at the AI conference EurIPS in Copenhagen! 🔥 I attended the european counterpart of the top-tier ML conference NeurIPS in Copenhagen, where I presented my Bayesian mixture modeling framework to evaluate the reliability of repeated measurements in brain networks. The model represents each connection as a probabilistic mixture capturing the presence or absence of a true brain connection and estimates reliability across repeated measurements. We first validated the approach using synthetic repeated-measures data with controlled noise mimicking real neuroimaging variability, before applying it to real datasets. The model successfully detected connections with varying reliability and produced biologically meaningful estimates, while also revealing important limitations in recovering absent connections. Although developed for brain connectivity, the framework is broadly applicable to any network derived from repeated observations where uncertainty quantification is critical. The conference was additionally a great opportunity to learn more about the latest advances in deep learning research, reconnect with friends and former colleagues, and exchange ideas with the community. Exploring the beautiful harbor of Copenhagen was also a wonderful highlight of the trip.

Selected Publications

Front. Neuroimaging

Quality control in functional MRI studies with MRIQC and fMRIPrep

Céline Provins, Eilidh MacNicol, Saren H. Seeley, and 2 more authors

Frontiers in Neuroimaging , 2023

Abs HTML

The implementation of adequate quality assessment (QA) and quality control (QC) protocols within the magnetic resonance imaging (MRI) research workflow is resource- and time-consuming and even more so is their execution. As a result, QA/QC practices highly vary across laboratories and “MRI schools”, ranging from highly specialized knowledge spots to environments where QA/QC is considered overly onerous and costly despite evidence showing that below-standard data increase the false positive and false negative rates of the final results. Here, we demonstrate a protocol based on the visual assessment of images one-by-one with reports generated by MRIQC and fMRIPrep, for the QC of data in functional (blood-oxygen dependent-level; BOLD) MRI analyses. We particularize the proposed, open-ended scope of application to whole-brain voxel-wise analyses of BOLD to correspondingly enumerate and define the exclusion criteria applied at the QC checkpoints. We apply our protocol on a composite dataset (n = 181 subjects) drawn from open fMRI studies, resulting in the exclusion of 97% of the data (176 subjects). This high exclusion rate was expected because subjects were selected to showcase artifacts. We describe the artifacts and defects more commonly found in the dataset that justified exclusion. We moreover release all the materials we generated in this assessment and document all the QC decisions with the expectation of contributing to the standardization of these procedures and engaging in the discussion of QA/QC by the community.
Nature Methods

Reliability characterization of MRI measurements for analyses of brain networks on a human phantom

Céline Provins, Hélène Lajous, Elodie Savary, and 7 more authors

Stage 1 Registered Report accepted in principle at Nature Methods , 2023

Abs HTML

Network-based approaches are widely adopted to model functional and structural ‘connectivity’ of the living brain, extracted noninvasively with magnetic resonance imaging (MRI). However, these analyses —on functional and structural networks— render unreliable at the finer temporal, spatial, and brain-parcellation scales. Consequently, the clinical translation of these analyses has yet to materialize meaningfully, and interpretation of the skyrocketing production of scientific literature requires caution. We will characterize relevant sources of variability and assess the reliability of structural and functional networks extracted from MRI with the repeated acquisition of a single, healthy individual, whom we regard as the ‘Human Connectome Phantom’. Two comprehensive MRI protocols will be executed across three different devices ( 48, 12, and 12 sessions, respectively) while recording a wealth of physiological signals to help model corresponding spurious effects on brain networks. To maximize reuse, e.g., as a benchmark reference, a baseline for machine learning models, or a source of prior knowledge, we will openly share all data and their derivatives. By systematically assessing spurious sources of variability throughout the neuroimaging workflow, we will deliver reliability margins of brain networks that inform future research and contribute to the standardization of ‘connectivity measurement’.
PLOS Biology

Defacing biases in manual and automatic quality assessments of structural MRI with MRIQC

Céline Provins, Elodie Savary, Thomas Sanchez, and 8 more authors

Stage 1 Registered Report accepted in principle at PLOS Biology , 2023

Abs HTML

A critical requirement 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 comparing their ratings on the same set of images in two conditions: “nondefaced” (that is, 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.