Scribble-Guided Interactive 3D Myotube Instance Segmentation

Research topic/area

Deep Learning 3D Segmentation

Type of thesis

Master

Start time

-

Application deadline

15.07.2026

Duration of the thesis

-

Description

Accurate instance segmentation of 3D microscopy data is a key challenge in
biomedical image analysis. In particular, myotubes are elongated, highly
interconnected structures whose correct separation is essential for studying
muscle development, disease progression, and drug response. While recent
deep learning approaches have improved segmentation quality, they still
struggle with ambiguous boundaries, complex 3D morphology, and
overlapping structures.
In previous work, a baseline model was developed based on a U-Net
architecture. This approach struggles to predict coherent boundaries for
individual instances at intersection points.
Interactive segmentation methods, such as scribble- or prompt-based
approaches, have shown strong performance in 2D natural images by
incorporating minimal user input to refine predictions. However, these methods
are not directly applicable to dense 3D biomedical data and typically do not
leverage prior model predictions. This creates the opportunity to develop new
methods that combine automated segmentation with efficient user guidance
tailored to 3D myotube data.

This thesis aims to develop a scribble-guided segmentation method inspired
by prompt-based models. The approach will use an initial prediction from the
existing model and refine it based on sparse user-provided scribbles. Different
strategies will be explored, such as fixing the foreground prediction or jointly
predicting instances, as well as suitable ways to encode scribble information.
The method will be trained on synthetic data with simulated annotations and
evaluated on its ability to improve instance segmentation quality with minimal
user interaction.

Requirement

Requirements for students

• • Basic knowledge of deep
• learning methods / AI
• • Experience or interest in
• segmentation
• • Basic programming skills,
• preferably in Python or
• another language with deep
• learning libraries

Faculty departments

• Engineering sciences
  Electrical engineering & information technologies
  Informatics
  Mechanical engineering
  Mechatronics & information technologies
  Mechanical Engineering
  Computer Science
  Electrical Engineering and Information Technology
  Mechatronics and Information Technology

Supervision

Title, first name, last name

M. Sc. David Exler

Organizational unit

Institut für Automation und angewandte Informatik

Email address

david.exler@kit.edu

Link to personal homepage/personal page

Website

Application via email

Application documents

• Grade transcript

E-Mail Address for application
Senden Sie die oben genannten Bewerbungsunterlagen bitte per Mail an david.exler@kit.edu


Back