From Supervised to Self-Supervised: A JEPA Encoder for Coronary Artery Calcification Detection in 2D Angiography

Research topic/area

Medizinische Bildgebung für Modellierung und Simulation

Type of thesis

Master

Start time

-

Application deadline

30.04.2027

Duration of the thesis

-

Description

Motivation
Coronary artery calcification (CAC) is a key determinant of the complexity and risk of percutaneous coronary interventions. However, in the catheterisation laboratory, CAC is still primarily assessed through visual inspec- tion of 2D coronary angiography (2DCA) images, which can be subjective and challenging.
Our group recently developed a deep learning workflow for automatic CAC detection based on a double-head ResNet-18 encoder and a U-Net-like decoder *. However, the encoder is trained from scratch on a relatively small annotated cohort, which may limit the robustness and generalisability of the resulting model.
Joint Embedding Predictive Architectures (JEPA) offer an alternative approach by learning meaningful visual representations from large collections of unlabelled images, without relying on negative sampling. This makes them particularly promising for medical imaging, where large annotated datasets are often difficult to obtain. In this thesis, a JEPA-based vision encoder will be pretrained on large public coronary angiography datasets and subsequently integrated into the existing CAC detection workflow, replacing its supervised backbone.

Student Project
The project aims to assess whether self-supervised pretraining on public angiography data improves CAC identification over the current supervised baseline.

• Collection and harmonisation of public 2DCA datasets (ARCADE, CADICA, AngioCAD) • Self-supervised pretraining of a ViT encoder with a LeJEPA / SIGReg objective
• Integration of the pretrained encoder into the existing double-head workflow
• Benchmarking against the published baseline (F1-score, IoM, Hausdorff distance)

*: De Rose & Raggio et al. “A multi-model deep learning approach for the identification of coronary artery calcifications within 2D coronary angiography images”. In: International Journal of Computer Assisted Radiology and Surgery 20 (2025), pp. 1273–1281.

Requirement

Requirements for students

• General knowledge in Computer Science and Biomedical Engineering.

Faculty departments

• Engineering sciences
  Biomedical Engineering
  Electrical Engineering and Information Technology
  Medical technology
• Natural sciences and Technology
  Biology
• Other e.g. teaching degrees
  Computer Science

Supervision

Title, first name, last name

M.Sc. Ciro Benito Raggio, Dr.-Ing. Edoardo De Rose

Organizational unit

Institut für Biomedizinische Technik

Email address

ciro.raggio@kit.edu

Link to personal homepage/personal page

Website

Application via email

Application documents

• Cover letter
• Curriculum vitae
• Grade transcript

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


Back