French
English
[Winter-Spring 2026]
Scientific context
Over the past decade, deep learning have established itself as the most common and powerful for computer vision, achieving state-of-the-art performance in many tasks, from image classification and segmentation to modality registration or image reconstruction. Although label-guided supervision have been the prominent learning paradigm, it suffers from several limitations in some cases. In many fields, such as medical imaging, the cost to acquire labels can be major obstacle to the use a supervised models. Moreover, it induces a bias towards the type of anomalies to detect, which can be undesirable if one aims to spot rare pathologies.
To address these limitations, unsupervised anomaly detection (UAD) has emerged as an alternative approach to supervised learning. Instead of learning from both normal and abnormal examples, UAD models are trained exclusively on normal (i.e., anomaly-free) data and learn to capture the underlying distribution or structure of the normal data. At test time, the model computes the distance between the new sample and the learnt data distribution. Based on this measure, it is expected to differentiate between in-distribution samples and samples that deviate from the estimated distribution of anomaly-free data. A wide range of methods have been developed, employing different strategies to localize abnormal regions in images. A common family of such methods relies on the extraction of hierarchical features from a pre-trained encoder to build a memory bank, that is, a set of discrete prototypical features representing normal patterns [1, 2]. These methods have several benefits, in particular the simplicity, speed, and their interpretability compared to other existing methods. In the field of medical image analysis, UAD has also been widely studied [3, 4, 5] but many models focus on rather easy-to-spot anomalies, such as large brain tumours, and recent studies show that such UAD methods struggle to detect subtle anomalies [5] as encountered in various brain pathologies including microbleeds, epilepsy or multiple sclerosis lesions as illustrated on Figure 1. Unsupervised learning also has potential applications in the modelling of normative brain development for the early detection of various pathologies such as dementia or Parkinson’s disease (PD).
Within the MYRIAD team at CREATIS laboratory, we have developed expertise in the field of unsupervised representation learning and UAD for the detection of subtle anomalies in both industrial and brain images[2, 5, 6]. We recently proposed a prototype-based method, PRADOT [2] which can enforce a structural constraint when learning the prototypes, allowing to capture the underlying organization of the normal samples, thus improving the detection of incoherencies in images (see Figure 2). This method achieves performance on par with strong baselines on two reference benchmarks for anomaly detection on industrial images. The main goal of the project is to build on this method and explore prototype-based UAD methods for the detection of various subtle pathologies on 3D brain MRIs.
The main goal of the project is to build on this method and explore prototype-based UAD methods for the detection of various subtle pathologies on 3D brain MRIs
Objectives of the internship
The purpose of this master project is to develop and evaluate the performance of prototype-based unsupervised anomaly detection for the detection of various subtle brain pathologies. The intern will explore the following axes:
• As seen on Figure 2, many deep learning methods for image analysis rely on the extraction of relevant features from input images. In traditional computer vision, a large field of research is dedicated to finding the models to extract those features. However, most of them do not transfer well to the medical image field. Based on preliminary work, a first objective of the internship is to select and train a (variational) autoencoder on 3D brain images in order to obtain an encoder capable of extracting relevant features from input images. Following previous works, both patch-based and whole-brain approaches can be tested.
• In a second stage, the intern will adapt prototype-based UAD methods, such as PRADOT [2] and PatchCore [1], for pathology detection on 3D brain MRIs. He/she will conduct the training of the models and their qualitative and quantitative evaluations, as well as the comparison with state-of-the-art methods.
• Depending on the progress of the internship, a final objective may be to adapt prototype anomaly detection methods to take into account clinical variables in the model, such as age to better account for those confounding effects in population image analysis.
The candidate will work closely with other students involved in the project, in particular with Robin Trombetta, a PhD student who will prepare the general pipeline analysis (data formatting, image preprocessing, etc.) before the arrival of the student and co-supervise the master project. The work carried out during this internship could lead to a publication in a national or international conference.
Skills and working environment
The candidate should have a background either in machine learning and/or deep learning or image processing, as well as good programming skills. Experience with deep learning libraries such as PyTorch would be appreciated. We are looking for an enthusiastic, autonomous and rigorous student with strong motivation and interest in multidisciplinary research (image processing and machine learning in a medical context).
He/she will have access to computing resources (CREATIS and/or CNRS supercomputer) as well as several public datasets: ADNI [7], a large-scale database of 3000 MR exams with controls and patients with various levels of cognitive decline, WHM [8] , a dataset containing 60 3D MR images of patients with hyperintensities of the white matter, as well as two datasets [9, 10] for the study of epilepsy with respectively 170 (85 with epilepsy and 85 controls) and 542 subjects (442 with epilepsy and 100 controls).
The candidate will join a research project gathering specialists with complementary expertise in image processing and machine learning (CREATIS) and medicine (HCL). The project is funded by the ANR research grant SEIZURE (https://anr-seizure.github.io/). The candidate will benefit from a stimulating research environment, as he/she will have the opportunity to interact with clinicians and members of the MYRIAD team working in the field of deep machine learning for medical image analysis.
Application
Interested applicants are required to send a cover letter, CV and any other relevant documents (reference letter, recent transcripts of marks, ...) to: carole.lartizien@creatis.insa-lyon.fr and robin.trombetta@creatis.insa-
lyon.fr
References
[1] Karsten Roth et al. “Towards total recall in industrial anomaly detection”. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 14318–14328.
[2] Robin Trombetta and Carole Lartizien. “Learning local and global prototypes with optimal transport for unsupervised anomaly detection and localization”. In: arXiv preprint arXiv:2508.12927 (2025).
[3] Julian Wyatt et al. “Anoddpm: Anomaly detection with denoising diffusion probabilistic models using simplex noise”. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 650–656.
[4] Walter HL Pinaya et al. “Unsupervised brain imaging 3D anomaly detection and segmentation with transformers”. In: Medical Image Analysis 79 (2022), p. 102475.
[5] Nicolas Pinon, Robin Trombetta, and Carole Lartizien. “One-Class SVM on siamese neural network latent space for Unsupervised Anomaly Detection on brain MRI White Matter Hyperintensities”. In: Medical imaging with deep learning. PMLR. 2024, pp. 1783–1797.
[6] Zaruhi Alaverdyan et al. “Regularized siamese neural network for unsupervised outlier detection on brain multiparametric magnetic resonance imaging: Application to epilepsy lesion screening.” In: vol. 60. 2020, p. 101618.
[7] R C Petersen et al. “Alzheimer’s Disease Neuroimaging Initiative (ADNI): clinical characterization”. en. In: Neurology 74.3 (Jan. 2010), pp. 201–209.
[8] Hugo J Kuijf et al. “Standardized assessment of automatic segmentation of white matter hyperintensities and results of the WMH segmentation challenge”. In: IEEE transactions on medical imaging 38.11 (2019), pp. 2556–2568.
[9] Fabiane Schuch et al. “An open presurgery MRI dataset of people with epilepsy and focal cortical dysplasia type II”. In: Scientific Data 10.1 (July 2023), p. 475. issn: 2052-4463. doi: 10.1038/s41597-023-02386-7.
[10] Peter N Taylor et al. “The Imaging Database for Epilepsy And Surgery (IDEAS)”. In: Epilepsia 66.2
(Dec. 2024), pp. 471–481.