A resource intended to test and validate new CMR post-processing technology and machine learning techniques.

Download the data Benchmark your performance Find out more 

The VOLUMES resource can:

• Benchmark machine learning techniques: Measuring head-to-head performance against expert clinicians.
• Measure clinical performance: quantify what the smallest change in LVEF that clinicians can be confident in detecting.
• Help researchers: Sample sizes calculations and core lab standardisation.

(A) Real world imaging biomarker analysis using 110 patients undergoing multi-center, multi-disease, multi-vendor, multi-field strength cardiovascular magnetic resonance imaging on two consecutive occasions in a short time frame. (B) Machine learning using a convolutional neural network approach, and human analysis with five different approaches are used to extract left ventricular ejection fraction (LVEF), mass, end-diastolic volume (EDV), end-systolic volume (ESV) and stroke volume (SV). (C) There is equivalent precision for all LV metrics (scan-rescan coefficient of variation, CV) between an expert and neural network approach, in addition to good inter-observer accuracy. A significant source of human error is attributable to intra-observer error (CV). Abbreviations: ICC= Lin’s intra-class coefficient; ns= non-significant.