Minseok RyuNeuroscience × Robotics

04 / Medical imaging

CCTA segmentation

A KURF research project exploring how anatomical structure can be extracted reliably from coronary CT angiography.

DomainMedical imaging
ProgrammeKURF
MethodsDetails pending
StatusArchive in progress
Placeholder illustration of layered segmentation contours on a scan

Overview

Finding anatomy in the noise.

Segmentation converts a medical image into an explicit representation of anatomy—the point where pixels begin to support measurement, modelling and clinical reasoning.

To complete this section: the full KURF context, target anatomy, supervisor/team, dataset, research objective and your exact contribution.

The problem

What makes CCTA difficult?

This section will explain the imaging context and the sources of variation that matter for your target: anatomy, acquisition, contrast, artefact, class imbalance or annotation ambiguity.

Anonymised scan / target anatomy

Pipeline

From volume to structure.

The case study will trace data selection, preprocessing, model or algorithm design, training protocol and post-processing. Every step should connect to a stated challenge rather than read like a software inventory.

Useful material: pipeline diagram, representative slices, ground-truth masks, prediction overlays, training curves and carefully anonymised examples.

Evaluation

Accuracy with context.

We’ll report the metrics actually used—such as overlap or boundary measures—along with the evaluation split, baselines and qualitative failure cases needed to interpret them responsibly.

Prediction overlay / key result

Limits & future work

Where does the method break?

Failure cases, dataset limitations and clinical relevance will shape the final discussion. This is also where the project can point toward the research questions you may pursue next.

Back to case study / 01

Unibots