Machine Learning Internship - Foundation model adaptation for Multiple Sclerosis

About Raidium

Raidium is developing the most advanced radiological LVLM (Large Vision Language Model); The first version of our Foundation model, Curia, is now the state-of-the-art copilot of the new generation of radiological image manipulation software.

The internship - Foundation model adaptation for Multiple Sclerosis

Multiple Sclerosis (MS) is a complex chronic neurological disease. Early and accurate diagnosis via MRI, relying on subtle, tiny lesions and specialized sequences (FLAIR, SWI), remains a significant challenge in clinical practice.

• Foundation models (like DINOv2) leveraging Self‑Supervised Learning (SSL) have revolutionized computer vision, showcasing superior generalization. While their medical adaptations like Rad‑DINO and Curia (developed at Raidium, pre‑trained on 200M+ medical images that come from 150K CT scans and MRIs) exist, adapting them for MS imaging faces diminished efficacy.
• This challenge arises because the MS‑defining features—specifically the tiny, sub‑millimeter lesions and the data from specialized sequences (SWI‑m, SWI‑ph)—constitute a highly granular and underrepresented domain within large‑scale medical pretraining datasets.

In this internship, you will:

• Adapt or pre‑train a vision transformer with SSL (e.g. Curia or DINOv2 from scratch) on brain MRI images leveraging large public and private datasets.
• Investigate and apply parameter‑efficient fine‑tuning (PEFT) methods, such as LoRA, to adapt the pre‑trained model for specific MS tasks.
• Learn everything about the diagnosis of MS on brain MRI (McDonald Criteria, Central Vein Sign (CVS) and paramagnetic rim lesions (PRLs), …).
• Work directly with radiologists to gain feedback and validate model performance against clinical ground truth.
• Establish robust evaluation pipelines to quantitatively assess model performance and benchmarking against leading foundation models.

You

• Finishing MSc in Mathematics / Computer Science / Electrical Engineering.
• Proficiency with statistics / Machine Learning algorithms.
• Proficiency with Deep Learning, Computer Vision algorithms and/or NLP algorithms.
• Proficiency with Python.
• Willingness to work in the healthcare industry.
• Authorization to work in France.

Pluses

• Experience in an agile work environment.
• Publication history in top‑tier Machine Learning or healthcare journals.
• Experience in healthcare and/or with the manipulation of radiological images.
• Experience with Self‑Supervised Learning and Transformers.
• Experience with HPC clusters and GPU‑based infrastructures.

Why join us

• Have a real and positive impact on people's lives.
• Apply and improve the latest Deep Learning technologies to build an inspiring product.
• Be in contact with radiologists in our partner center.
• Be among the first employees of Raidium.
• Work with a team of internationally recognized experts.
• Work with a diverse team of engineers and medical doctors.
• Competitive salary.
• Access to premium computing hardware.

€2,000 - €2,000 a month

Duration: 6 months

Expected start date: March 2026

Location: Hôpital Cochin, Paris 14