University College London
Browse
- No file added yet -

SR-TEE Dataset for Regressing Simulation to Real: Unsupervised Domain Adaptation for Automated Quality Assessment in Transoesophageal Echocardiography

Download (741.4 MB)
Version 2 2023-08-03, 09:49
Version 1 2023-07-19, 14:26
dataset
posted on 2023-08-03, 09:49 authored by Jialang XuJialang Xu, Yueming JinYueming Jin, Bruce Martin, Andrew Smith, Susan Wright, Danail StoyanovDanail Stoyanov, Evangelos MazomenosEvangelos Mazomenos

This SR-TEE dataset is for our accepted paper at MICCAI2023 titled 'Regressing Simulation to Real: Unsupervised Domain Adaptation for Automated Quality Assessment in Transoesophageal Echocardiography'. Official code can be found at https://github.com/wzjialang/SR-AQA.


It includes 16,192 simulated and 4,427 real transoesophageal echocardiography (TEE) images from 9 standard views (i.e., Mid-Esophageal 4-Chamber, Mid-Esophageal 2-Chamber, Mid-Esophageal Aortic Valve Short-Axis, Transgastric Mid-Short-Axis, Mid-Esophageal Right Ventricle inflow-outflow, Mid-Esophageal Aortic Valve Long-Axis, Transgastric 2-Chamber, Deep Transgastric Long-Axis, Mid-Esophageal Mitral Commissural). 


Simulated images were collected with the HeartWorks TEE simulation platform from 38 participants of varied experience asked to image the 9 views. Fully anonymized real TEE data were collected from 10 cardiovascular procedures in 2 hospitals, with ethics for research use and collection approved by the respective Research Ethics Committees. 


Each image is annotated by 3 expert anaesthetists with two independent scores w.r.t. two automated quality assessment tasks for TEE. The criteria percentage (CP) score ranging from ‘0-100’, measuring the number of essential criteria, from the checklists of the ASE/SCA/BSE imaging guidelines, met during image acquisition and a general impression (GI) score ranging from ‘0-4‘, representing overall ultrasound image quality. 


There are significant style differences (e.g. resolution, brightness, contrast, acoustic shadowing, and refraction artifact) between simulated and real data, posing a considerable challenge to unsupervised domain adaptation.


The structure of the dataset is as follows:

  • 'real_cases_data_frames' folder: contains real TEE images.
  • 'simulated_data_frames' folder: contains simulated TEE images.
  • real_cases_data_frames.csv: ground truth of real TEE images, four columns represent image name, view class, CP value, and GI value, respectively.
  • simulated_data_frames.csv: ground truth of simulated TEE images, four columns represent image name, view class, CP value, and GI value, respectively.


Funding

Wellcome/EPSRC Centre for Interventional and Surgical Sciences (WEISS) [203145Z/16/Z and NS/A000050/1]

EPSRC-funded UCL Centre for Doctoral Training in Intelligent, Integrated Imaging in Healthcare (i4health) [EP/S021930/1]

Horizon 2020 FET [863146]

Singapore MoE Tier 1 Start up grant [WBS: A-8001267-00-00]

History