Early prediction of remodeling after myocardial infarction (MI) is key in the fight against heart failure. Cardiac MR (CMR) allows functional assessment (volumetric parameters, wall thickening…) and tissue characterization (scar location and morphology) to be obtained and monitored over time. The analysis of these data is challenging due to the inconsistent breath-hold positions throughout the examination and due to shape changes in follow-up scans, thus requiring advanced image analysis.¹ In this work, a registration framework is proposed to: (i) correct for slice-to-slice misregistration in 2D sequences (cine and late gadolinium enhancement, LGE); (ii) register LGE onto cine data; (iii) register the follow-up data onto the baseline data. This approach is expected to make the association between local function and scar more accurate and less user-dependent than a conventional segment-wise analysis. The framework was evaluated in terms of registration accuracy and was applied to study relations between scar transmurality and local function in the setting of reperfused MI.

Patient population: 114 patients with first ST-segment elevation MI (REMI study, Gov Identifier: NCT01109225) were included in this work. All patients underwent CMR within 3 days after MI (t₁) and a follow-up scan at 6 month (t₆). All patients were treated by primary percutaneous coronary angioplasty and optimal medication.

MRI data: CMR scans were conducted on a 3T Signa HDxt scanner (General Electric, USA). They included functional assessment by cine imaging (2D SSFP sequence, short-axis coverage of the left ventricle) and scar assessment by LGE imaging (2D IR-FGRE, short-axis coverage of the left ventricle). Images were manually segmented by a cardiologist in order to delineate left ventricular (LV) endocardial and epicardial contours (cine and LGE), scar contours (LGE) and anterior right ventricular (RV) junction points.

Registration (see Fig. 1): Firstly, a slice-to-slice correction was applied to account for breath-hold inconsistencies. This was done by searching for translations that best register the contours onto a template surface created with a polynomial fit. Then LGE contours were registered onto the cine (end-diastolic) contours using a point cloud rigid registration technique (iterative closest point ²), after upsampling in the slice direction. Finally, data at t₆ were registered onto data at t₁ (based on the cine contours) using the same point cloud registration technique. Rigid registration was used in order to preserve shape changes due to remodeling. Eventually a projection was computed to map the data from one mesh to another. Registration errors were assessed by computing (i) the mean distance between the registered endocardial surface meshes in mm and (ii) the mean angular error between these meshes, as defined by the angle formed by the two anterior RV junction radii (expected to be close to 0° after registration).

Data analysis (see Fig. 2): From the registered contours, a 3D reconstruction was applied in order to map all parameters onto the LV cavity surface mesh including regional function (wall thickening) and scar (transmurality). For each given patient the parameter of interest (here wall thickening) was averaged according to scar transmurality defined at a 5-point scale (0, 0-25%, 25-50%, 50-75% and 75-100%).After the registration steps, the mean distance (respectively angular error) between endocardial surface meshes was 2.2 mm +/- 0.8 mm (4.8° +/ -2.4°) between LGE and cine datasets (intra-session registration), and 2.9 mm +/- 1.2 mm (5.1° +/- 2.4°) between cine datasets at t₁ and t₆ (inter-session registration). Regional function at t₆, as assessed by the wall thickening in %, was inversely associated with scar transmurality at t₆, but also with scar transmurality at t₁ as shown in Fig. 3.Registration errors were acceptable considering the variability between acquisition parameters (different trigger windows in cine and LGE) and possible shape changes between t₁ and t₆. The analysis confirmed the relation between scar transmurality and regional function.³ The fact that acutely assessed scar transmurality is inversely related to 6-month local function might be useful for early stage function prediction after an MI. In future work, other parameters related to remodeling will be investigated such as cavity dilation or diastolic wall thinning.