Radiology AI Research Intern — Lesion Profiling

StartX; Paris; Hybrid; EUR 12,000 - 21,000; Full time

Benefits offered by this job

Competitive salary

Premium computing hardware

Healthcare access

Job summary

Raidium is developing a radiological foundation model, the GPT of radiology, to enable an imaging biomarker factory for clinical practice and research. This internship focuses on learning lesion-level characterisation from report-derived labels using OncopilotDetect detections as lesion candidates.

You will structure training data, design a weakly supervised method, handle hedged wording, and build a lesion-level reference set.

Qualifications

• MSc in Mathematics, Computer Science or Electrical Engineering.
• Strong background in statistics and machine learning.
• Proficiency with Python.

Responsibilities

• Structure a training dataset from CT exams, radiology reports, and outputs of our detection models.
• Design a weakly supervised method that learns to characterise each lesion from these labels, using the size and location constraints given by the report.
• Handle the specifics of radiology reports: hedged wording (e.g. “probable hemangioma”), findings left unreported when a more severe one dominates, and label noise.
• Learn how radiologists characterise lesions (guideline categories such as Bosniak or LI-RADS) and build a lesion-level test set with a reliable reference standard.
• Benchmark your approach against fully supervised baselines, with the goal of a MICCAI publication.

Skills

Statistics

Machine Learning

Deep Learning

Computer Vision

Natural Language Processing

Python

Healthcare experience

Education

MSc in Mathematics / Computer Science / Electrical Engineering

Tools

HPC clusters

GPU-based infrastructures

Large Language Models

Job description

Raidium is developing a radiological foundation model, the GPT of radiology, to enable an imaging biomarker factory for clinical practice and research. This internship focuses on learning lesion-level characterisation from report-derived labels using OncopilotDetect detections as lesion candidates.

You will structure training data, design a weakly supervised method, handle hedged wording, and build a lesion-level reference set.