Computer Vision · Hackathon

Aorta Detection

Abdominal ultrasound → Mask R-CNN segmentation → 3D mesh reconstruction → aortic diameter estimation. Built for the BitsXLaMarató hackathon.

View Source on GitHub
CV Pipeline
BitsXLaMarató hackathon · TV3 La Marató
🎬
Ultrasound video
Load study, extract frames
🧠
Mask R-CNN
Instance segmentation per frame
🗂️
Mask stack
Binary TIFFs for 3D recon
🧊
3D mesh
ISO surface with Meshlib / Open3D
📏
Diameter
Contour analysis + heuristics
Mask R-CNN (PyTorch)OpenCVMeshlib / Open3DTkinter GUICustom annotationsCUDA inference
🔬

Simulated inference

Browser mock · real B-mode frame

Mimics the real Mask R-CNN inference on a B-mode frame from the rat aorta dataset. The red overlay is a browser mock, not model output.

B-mode ultrasound frame
📏

Diameter explorer

Educational risk visualization

The app estimates aortic diameter from segmented contours. Drag to see how a measurement maps to educational risk buckets — not for diagnosis.

Max outer diameter28 mm
Typical range

Many adult abdominal aortic diameters fall well below 30 mm.

Illustration only. AAA management follows local guidelines and imaging context.

Project screenshots

3D reconstruction of the abdominal aorta from segmented frames.
3D reconstruction of the abdominal aorta from segmented frames.
Segmentation / visualization pipeline output.
Segmentation / visualization pipeline output.
GUI workflow: load video, run inference, inspect masks and 3D output.
GUI workflow: load video, run inference, inspect masks and 3D output.
Team
Pol Casacuberta · Tatiana Meyer · Pablo Vega · Ton Vilà