Espree

Without any previous ventilator or MRI training or experience, on the day of the experiment each subject was allowed sufficient time (~5 minutes) to become comfortable breathing spontaneously via a facemask and connected to an MR compatible non-invasive ventilator (MR1 Hamilton Medical AG, Bonaduz, Switzerland). Subjects then acclimatised to lying on an MR couch (Siemens Espree, Siemens AG, Erlangen, Germany) whilst breathing spontaneously with the ventilator using air/oxygen mix (FiO 2 <0.4). Baseline end tidal partial pressure of carbon dioxide (P ET CO 2 ) levels, non-invasive pulse oximetry and heart rate were measured at the start and monitored throughout, with pre-defined limits set to ensure subject safety during the procedure [7] . The ventilator was then switched to take over their breathing non-invasively with positive pressure, controlling the rate and level of breathing, by causing all the usual and normal cycles of inflation and deflation. Two arbitrary ventilation rates of 20 and 25 breaths per minute were set, whilst simultaneously reducing inflation volume to match or exceed their metabolic rate such that P ET CO 2 >2.7kPa, sPO 2 >94% and heartrate <100bpm [5] [6] [7] .
Participant experience of the procedure was assessed through a non-validated questionnaire, using 5 questions each scoring between 1 and 5 points; maximum comfort / experience scoring 5 points.

Cardiac magnetic resonance imaging was completed on visit 3 using a 1.5-T magnetic resonance imaging system (Siemens Espree; Siemens, Erlangen, Germany; 70 cm bore, advanced cardiac package, TIM Matrix surface coil) and this protocol was based on the previously described Multiethnic Study of Atherosclerosis cardiac magnetic resonance protocol. 18 (link) Cine images with steady-state free precession were gated by ECG to assess LV function and mass. Short axis volumetric coverage was used to obtain LV volume and mass. Papillary muscles were excluded from LV mass but included in LV volumes. Analysis of cardiac structure and function data was performed using Cardiac Image Modeler software (University of Auckland, New Zealand). Tagged images of the basal, mid, and apical LV walls were performed using a cine radiofrequency grid-tagging sequence (field of view 400 mm, slice thickness 8 mm, 192×256 matrix, repetition time 60 ms, echo time 4 s, and fractional anisotropy of 12° [Siemens sequence: Tl2d1r5]). A harmonic phase algorithm (Diagnosoft, Morrisville, NC) was used to complete the myocardial tagging analysis. Finally, global strain was computed by using the average peak circumferential strain of the apical, mid, and basal walls of the LV.