AI Research Intern - Generative AI for Synthetic Bronchoscopy Data Generation

The Internship In Short

This internship explores data-driven generative AI for realistic synthetic bronchoscopy images. Using an existing simulator with exact 3D geometry and camera poses, conditional models will learn visual appearance from real bronchoscopy data while preserving anatomy and geometric ground truth. Generated data will be evaluated for realism, diversity, and geometric consistency.

The topic

Early detection of lung cancer significantly reduces mortality. After a suspicious lesion is detected via computed tomography (CT), a biopsy is often performed using bronchoscopy, which requires navigating complex airways to reach the lesion. Within the Naviscope project, Multitel develops AI methods to assist this navigation. Development of such methods requires datasets containing bronchoscopy images, ground-truth camera poses (positions and orientations), and CT-derived 3D airway models. However, acquiring ground-truth poses is not feasible with the thin bronchoscopes used in the project to reach deep airways.

To address this limitation, Multitel developed BronchoSim, an automated pipeline that generates synthetic bronchoscopy videos from CT-derived 3D airway models, with exact camera poses for every frame. It improves visual realism by automatically optimizing rendering parameters to maximize the similarity between the generated and real bronchoscopy images. However, image generation still relies on manually designed procedural textures and rendering rules defined in the Blender software.

Generative AI offers an alternative: learning visual appearance directly from real data, conditioned on simulator geometry such as depth, surface normals, masks, and camera pose. The internship will use Naviscope's private bronchoscopy data to develop and evaluate generative AI models. Technical and clinical feedback on the generated images will be obtained from clinicians and the bronchoscope manufacturer involved in the project.

The goal of this internship is to develop and evaluate a geometry-conditioned generative AI module that can replace or complement the current procedural rendering stage of BronchoSim while preserving its accurate ground-truth geometric information.

What you will work on:

You will explore several research questions, including:

• How can generative AI turn BronchoSim's geometric data into realistic bronchoscopy images?
• How can we preserve airway anatomy and keep the images consistent throughout a video?
• How can we assess image quality and usefulness for AI-based navigation?

Your responsibilities will include:

• Reviewing relevant methods, datasets, and pre-trained models to select a suitable approach.
• Adapting a generative model to create realistic bronchoscopy images from BronchoSim's geometric data.
• Evaluating realism, diversity, and geometry preservation, and comparing the results with the current rendering pipeline.
• Discussing the generated images with the project team and clinicians.
• Documenting and presenting the findings to Multitel's AI team.

The profile we are looking for:

We are looking for a motivated student who is curious about generative AI and enjoys applying computer vision to practical challenges in medical imaging. You should have:

• Proficiency in Python.
• Basic knowledge of deep learning and computer vision.
• Interest in or knowledge of medical imaging and generative AI.
• Experience with PyTorch, Git, Docker, or Blender is a plus.
• Strong oral and written communication skills in English.

Even if you don’t check every box in our requirements, we encourage you to apply. We value diverse backgrounds, curiosity, and the drive to learn, grow, and achieve.

Recruitment Process:

1. Application Review

We review your CV and cover letter to assess your background, skills, and interest in the internship topic.

2. Interview

Selected candidates are invited to an interview combining a technical discussion with a conversation about the internship objectives, your interests, and your expectations.

3. Following the interview

We select the candidate whose profile and interests best match the internship.