The 2nd ThrombUS+ data challenge is now available in Kaggle and as part of the EAI CloudComp 2026 conference.
Overview
DVT diagnosis relies on experts manually assessing vein compressibility in ultrasound videos. This data challenge aims to foster the development of AI models for DVT detection without the need for exhaustive pixel-wise annotations, reducing the burden of manual labeling.
Participants are invited to submit their responses to one or both of the data challenges in Kaggle, and submit a full paper describing their methodology and results in the respective EAI CloudComp 2026 workshop.
There will be a monetary prize for the first 3 most successful submissions!
The Dataset
Compression ultrasound videos of lower limbs collected during a multi-center cohort study across European hospitals. Patients suspected of DVT were scanned using conventional ultrasound machines according to a dedicated protocol, with full ethics approval and informed consent.
Videos capture four key anatomical sites: Near the inguinal ligament (common femoral vein); a few centimeters distally (great saphenous vein junction); Mid-thigh (femoral vein); Below the knee (popliteal vein).
All videos have been anonymized and tagged with anatomical site and limb laterality. Medical experts have assessed each video for vein compressibility (No/Partial/Yes) and thrombosis presence (Yes/No), providing ground truth annotations for model training.
Training data set of 2341 videos acquired from 594 patients from 5 hospitals
Testing data set of 578 videos acquired from 594 patients from 5 hospitals
Challenge 2.1: Detect DVT Presence
Your task is to build a machine learning model that analyzes compression videos and calculates the probability that the video corresponds to a thrombosis case, evaluated on Log Loss.
Challenge 2.2: Anatomical Site Identification
Your task is to build a machine learning model that can look at compression videos and accurately identify the anatomical site where scan was taken from (choice out of 4 anatomical sites).
The data challenge is an official component of the 2026 EAI CloudComp conference which will take place in Derby, UK, 29th June – 1st July 2026. Challenge participants are expected to submit a paper to the conference workshop “Benchmarking Weakly Supervised, Segmentation-Free AI Models for Autonomous Deep Vein Thrombosis Detection”
Related Links
EAI CloudComp Workshop Important Dates:
Submission deadline: 30th April 2026
Notification deadline: 15th May 2026