Identification of multimodal biomarkers from a homogeneous, multicentre cohort of imaging and clinical data from patients with amyotrophic lateral sclerosis

Download the internship proposal

Internship type

Master internship (M1 or M2) in computing, data analysis, statistics and or computer vision.

Date and duration of the internship

March to August 2027 maximum

Context

Amyotrophic lateral sclerosis (ALS), or Charcot’s disease, is a rare and progressive neurodegenerative disorder, characterised primarily by the degeneration of central and peripheral motor neurons. Its incidence in France is estimated at around 2.5 new cases per 100,000 inhabitants per year [2]. Despite typically presenting with motor symptoms, ALS is now regarded as a multisystemic disorder, which may also be accompanied by cognitive and behavioural impairments and involves, to varying degrees, different structures of the brain and spinal cord [3]. One of the main characteristics of ALS is its significant clinical and biological heterogeneity. Patients differ in particular in terms of age and site of onset, the distribution of motor symptoms, the possible presence of cognitive and behavioural abnormalities, genetic background, rate of progression and survival.

This variability poses a major challenge for diagnosis and prognosis, but also for the design of therapeutic trials: insufficiently stratified populations may limit the ability to identify the efficacy of a treatment in a particular subgroup of patients [4]. Identifying biomarkers that can better characterize the different ALS profiles and predict their progression is therefore a major challenge today. Brain and spinal cord MRI represents a particularly promising source of non-invasive biomarkers in this regard. To date, studies have mainly focused on predefined parameters, such as the volumes of regions of interest or measures of atrophy. However, the information directly contained within the images (spatial distribution of abnormalities, textures, relationships between different regions of the central nervous system) remains under-exploited [9]. Furthermore, data analysis is limited by the relative scarcity of the disease and by the difficulty in establishing large, homogeneous cohorts. Constraints related to the progression of the disease limit the possibility of clinical assessment and the repetition of imaging examinations. The available databases are therefore often small, single-centre or heterogeneous, which constitutes an obstacle to the development and validation of robust and generalisable machine learning methods.

From the end of 2026 to the end of 2027, several existing multicentre cohorts of patients with ALS, some of which have already been the subject of research and publications within the LIB, are brought together and harmonised within our team. The creation of this harmonised cohort thus offers a unique opportunity to move beyond the analysis of small case series and to study the heterogeneity of ALS using an integrated approach combining clinical phenotype, disease progression, and imaging.

Objective

Identify and characterize multimodal profiles of patients with ALS, based on clinical data and brain and spinal cord imaging, in order to identify multimodal biomarkers that could improve patient staging and, ultimately, predict their prognosis.

Missions

  • to interact both with the engineering and clinician supervision in two different sites Cordeliers and APHP;
  • to describe the cohort using descriptive statistics and visualizations;
  • to analyze clinical factors using various dimensionality reduction analyses (PCA, t-SNE, UMAP, etc.) on these clinical data compared with control data, and to calculate intra-and inter-individual and intra- and inter-cluster metrics [10];
  • to use correlation, regression and explainability methods (LIME [7], XGBoost [1]) to identify the factors underlying each profile and compare them with those found in the literature;
  • to use saliency maps (XRAI [6], LayerCAM [5], IG [8], for example) on images (particularly MRI) to identify regions of interest and compare them with those of healthy subjects for various regions;
  • to produce representations of profiles of groups of individuals within the cohort (cerebrospinal disorders and identified clinical factors);
  • to validate these descriptive analyses with clinical experts in ALS, the clinical literature and previous analyses carried out within the laboratory on local or smaller cohorts;
  • to write an internship manuscript and, present at the LIB Master day and possible seminar/conferences/GT-IA from the LIB.

Skills

In your cover letter you must precise your skills in these domains and how you acquired them:

  • Statistics and development (Python, but if you prefer Matlab/Java/C/C++/R, or other you must mention your preference and in which circonstance you already used it);
  • Data analysis, machine learning and some basis in deep learning (if you have already practiced in an other project or internship, mention it by specifying the models and frameworks you already used and or conceived);
  • Multimodal data analysis (on tabular, and or molecular, and or images) interest and/or already used in other internship/project;
  • Interest in biomedical domain and/or rare disease and previous works on these related subjects (if you are interested to do a PhD, please mention it).

Monetary compensation

Internship compensation (according to current regulations).

Contact

For application send CV + motivation letter (french or english) with a description of your previous projects/internships to

  • Clara BRÉMOND clara.bremond[at]inserm.fr (assistant professor « MCF » in computing at Sorbonne University attached to the LIB)
  • Giorgia QUERIN giorgia.querin[at]aphp.fr (hospital practitioner at APHP and clinical researcher attached to the LIB).

References

1. Chen, T. (2015). Xgboost: extreme gradient boosting.
2. Corcia, P., et al. (2025). Epidemiology, disease evolution and economic burden of amyotrophic lateral sclerosis in France using the French national health data system. Brain Communications.
3. Feldman, E. L., et al. (2022). Amyotrophic lateral sclerosis. The Lancet.
4. Gil, J., et al. (2008). Causes of death amongst French patients with amyotrophic lateral sclerosis: a prospective study. European journal of neurology
5. Jiang, P. T., et al. (2021). Layercam: Exploring hierarchical class activation maps for localization. IEEE transactions on image processing.
6. Kapishnikov, A., et al. (2019). Xrai: Better attributions through regions. In Proceedings of the IEEE/CVF international conference on computer vision.
7. Ribeiro, M. T., et al. (2016). Model-agnostic interpretability of machine learning.
8. Sundararajan, M., et al. (2017, July). Axiomatic attribution for deep networks. In International conference on machine learning.
9. Turner, M. R., et al. (2009). Biomarkers in amyotrophic lateral sclerosis. The Lancet Neurology.
10. Wang, Y., et al. (2021). Understanding how dimension reduction tools work: an empirical approach to deciphering t-SNE, UMAP, TriMAP, and PaCMAP for data visualization. Journal of Machine Learning Research.