Stage de M2 pour Interne de Radiologie
A few words about Incepto and you !
What we do
We co-create and distribute AI applications for medical imaging – Bridging the gap between Physicians and Engineers.
Why we do it
With more and more data, Medical Imaging is becoming more and more complex. We use Artificial Intelligence technology to transform Medical Imaging. This is a fantastic opportunity to empower physicians, saving them time, bringing them closer to their patients, and helping most of the population get access to the best modern care.
How we do it
A journey is not only about what we do but also about how we do and look at things. We place our values at the heart of our work. Centered on inter-disciplinarity and clinical relevance, the foundation of our unique team resolve around expertise, humility, and integrity.
Take on the challenge of AI in Medical Imaging!
During your internship within the data science team, you will:
- Work with other experienced data scientists, radiologists, and developers of the team.
- Manage the production of an algorithm from the definition of a clinical question to the optimization of performance through the management of the annotation strategy and the training of the model Deliver tools that can fit into a global cloud platform.
Ideally, a good candidate should be interested in the articulation between both the clinical and the algorithmic aspects of the internship.
We currently offer 3 internship topics (6 months minimum), all of which deal with different clinical and algorithmic problems. The content of each of them is presented below.
Knee MRI Pathologies Detection Model Improvement
You will own the improvement of machine learning models to detect pathologies in Knee MRI examinations. In collaboration with datascientist, you will participate to the creation of high quality annotated databases to address the clinical problematic. In particular you will deal with typical transfer learning problems with the goal of maintaining our current model’s high performances on our database while enhancing them on newer databases (coming from machines not included in our training sets).
Additionally, you should propose from scientific literature reviews, some implementations and tools to automatically detect menisci and ligament pathologies in imbalanced datasets. In particular, research topics will include targeted data augmentation techniques and hard mining strategies.
High Precision Aorta Measurements using Model Based Approaches and Attention Models
You will own the design and development of a precise segmentation model for extremely varying clinical situations of the aorta on contrasted CT scans. The challenge is to include a priori information to incorporate high resolution information and do better than our baseline model. Particular attention will be given to the transcription of segmentation models into clinically exploitable measurements.
Additionally, you should :
- Propose from scientific literature reviews, some implementations, and tools to target deformed tubular structures
- Participate in the writing of a scientific paper if your results can be published.
Localization Model Development and Classifier Improvement through an Interpretability-Based Approach
You will own the adaptation of a novel learning based interpretability method to the specific problem of pathology detection on several anatomical structures of the knee (menisci, ligaments and cartilage). The new interpretability method involves state of the art generative and adversarial models. As a second, more research oriented part of the internship, you will be tasked to propose a method to exploit interpretability during the training of an evolved version of the pathology detection algorithms.
How to apply ?
If you are interested in one of the topics apply here!