1356342 - ECG-free 2D cardiac cine MRI with data-driven clustering across a range of heart rates

2D cardiac cine imaging is widely used to evaluate cardiac morphology and function. Self-gating has emerged as an alternative approach to ECG gating and typically assumes periodic cardiac motion, but may not be robust to cardiac motion irregularities. We have previously shown [1] that a segmented k-space, Cartesian golden-step balanced steady-state free precession (bSSFP) sequence with motion navigators can be used with k-means clustering to alleviate the dependency on regular cardiac motion assumptions. We aim to evaluate the performance of k-means clustering across a wide spectrum of heart rates.

Methods:

Fifteen subjects (9 volunteers and 6 patients, 8 female and 7 male) with sinus rhythm were scanned on a clinical 3.0T scanner (Skyra, Siemens) using a modified segmented Cartesian golden-step bSSFP [2] sequence with ESPIRiT [3] reconstruction (Figure 1) and the conventional ECG-gated, segmented bSSFP cine sequence. Parameters for the golden-step bSSFP sequence were: FA=67°, matrix size=208*172, spatial resolution=1.8mm x 1.8mm x 8mm, TE/TR = 1.7ms /3.4ms, segments=20. A ventricular short axis stack was acquired during breath-holding. To compare the ability of k-means clustering to distinguish different motion states, the data were evenly divided into three groups (n=5 per group) with three heart rate ranges: (1) group 1: 50-60 bpm, (2) group 2: 61 -75 bpm, and (3) group 3: 76-100 bpm. The right and left ventricular (RV/LV) ejection fraction (EF), end-diastolic volume (EDV), end-systolic volume (ESV), and LV mass for both the cluster-based and reference cine images were computed. Bland-Altman analyses were performed to assess agreement among the quantitative metrics.

Results:

The average age was 49.3±15.1 years. The heart rate range was 50 to 100 beats/min (average heart rate 63.6±10.2 bpm). Paired t-tests found no significant difference (p < 0.05) between the reference cine bSSFP and our cluster-based reconstruction for quantification of cardiac function and volumetry (Table 1, Bland-Altman analysis). The limits of agreement were narrow and the mean bias values for all metrics were negligible. Most of the coefficients of variation were < 5%. All data points were within the 95% limits of agreement. Figure 2 shows the generated cluster matrix values from part C of Figure 1 and auto-correlation maps along the time dimension. The multiple diagonal lines in each auto-correlation matrix suggest local periodicity of sinus rhythm, but the vanishing correlation between distant timepoints suggests differences in cardiac phase images during the breath-hold, which may be related to respiratory drift. Strict ECG gating would not preserve these differences.

Conclusion:

An ECG-free data-driven clustering approach to discriminate and bin cardiac phases is feasible. For heart rate ranging from 50 to 100 bpm, the proposed cluster-based framework can obtain accurate and precise ventricular ejection fraction and volume values compared to the reference method.