Purpose of document The purpose of this document is to document validation of the Strain analysis module in Segment software packages. Intended audience The intended audiences of this document are: Engineering team members from both Medviso AB and KU Leuven Management from both Medviso AB and KU Leuven External and internal auditors Users Background The analysis algorithm is developed by KU Leuven as a research project. The algorithm was integrated by Medviso AB in Segment software packages in order to have a complete image analysis package for Strain analysis purposes. Requirement requirement that should be validated is RA18.2 in the Requirement specification. Page 1/9
method The validation is performed in both phantom data and in patients. in computer phantom Phantom data were developed by Medviso as 22 cases with varying strain in both radial and circumferential direction and circumferential rotation. The phantom data simulates one shortaxis slice of a time resolved MR tagging image stack with tagging lines in the left ventricle. The LV contraction was designed in the range of normal values from cine MR images [1], as presented in Table 1. Circumferential strain Radial strain Circumferential rotation Phantom 1 0 0 0 Phantom 2 5,0 0 0 Phantom 3 0 0 0,3 Phantom 4 10,0 0 0 Phantom 5 15,0 0 0 Phantom 6 10,0 0 0,3 Phantom 7 20,0 0 0 Phantom 8 25,0 0 0 Phantom 9 20,0 0 0,3 Phantom 10 30,0 0 0 Phantom 11 35,0 0 0 Phantom 12 30,0 0 0,3 Phantom 13 0,5 4,1 0 Phantom 14 1,0 8,1 0 Phantom 15 1,5 12,2 0 Phantom 16 2,0 16,3 0 Phantom 17 2,5 20,4 0 Phantom 18 3,0 24,4 0 Phantom 19 3,5 28,5 0 Phantom 20 4,0 32,6 0 Phantom 21 4,5 36,6 0 Phantom 22 5,0 40,7 0 Table 1. Design of phantom data The simulated left ventricle was segmented manually by Helen Fransson, CEO of Medviso AB. Helen Fransson is PhD and has experience in analysis of cardiac MR images. The automatic strain analysis algorithm was then applied to the tagged image slice. The known simulated strain data was used as reference method. Page 2/9
in patients Ten patients provided by KU Leuven were considered for enrollment in the validation study. The patients were randomly selected from a study with data acquired in Linköping hospital by a Philips 1.5T MR scanner during 2010 to 2011. One patient was excluded from the validation study due to limited image quality in data. For LV volume quantification, cine short-axis image stacks were used. The LV segmentation in the cine image stack was performed by Jan Engvall by manual delineation in the software Segment. Jan Engvall has many years of experience in analysis of cardiac MR images. LV volumes for the nine patients are presented in Table 2. The mean ± SD for the nine patients were LVM 100 ± 23 g, EDV 164 ± 38 ml, ESV 66 ± 27 ml, and EF 60 ± 12 %. LVM (g) EDV (ml) ESV (ml) EF (%) Patient 1 120 162 55 66 Patient 2 148 176 108 38 Patient 3 91 154 51 67 Patient 4 86 130 34 74 Patient 5 84 239 100 58 Patient 6 72 124 29 76 Patient 7 106 199 82 59 Patient 8 105 163 75 54 Patient 9 86 125 63 50 Table 2. LV volumes for patient population The strain analysis was performed in MR tagging image stack in three short-axis slices (basal, midventricular and apical). The strain analysis starts by loading the cine image stack and the tagging image stack into. The next step is to perform LV segmentation in all slices in the first time frame (end-diastole) in the tagged MR image stack. The LV segmentation was performed by first copying the LV segmentation from the cine image stack, and then does manual correction of the LV segmentation if needed to fit the LV in the tagged image stack. The correction is needed if the position of the LV is changed between the cine and the tagged image stack, or if the slice in the cine image stack does not exactly corresponds to the slice in the tagged image stack. The LV segmentation in the tagging image stack was performed by Helen Fransson and reviewed by Robert Jablonowski. Robert Jablonowski is MD and has experience in MR strain tagging analysis. The final step in the analysis is to run the automatic strain analysis algorithm on the tagged image stack. For comparison, circumferential strain values in one short-axis slice from 2D Doppler echocardiography were used as reference method. Radial strain from 2D echo is known to be the most difficult component and has high observer variability. We therefore did not compare the radial strain component. For each patient, the tagged MR short-axis slice closest with respect to cardiac position to the Doppler short-axis slice, was selected for comparison. Page 3/9
Strain quantification The output from the Strain module is global and segmental strain values over time. The global strain value is the mean of strain for each slice. The segmental strain value is the mean of strain in each LV segment divided according to AHA 17 segment model [2]. Peak strain value is then defined as strain in the time frame corresponding to the peak global strain value in each slice. Figure 1 shows the Strain module interface with strain analysis result for one patient. Figure 1. Illustration of the Strain module interface. The left image panel shows the cine image stack and the right image panel shows the tagged image stack. Strain values are presented to the right in the interface as graphs and bullseye plot. Page 4/9
result in computer phantom Figure 2 and 3 illustrate the relationship between the simulated strain values and the strain values quantified by the Strain module in Segment software packages. Bias = -0.0 ± 1.0 Figure 2. Relationship between simulated global circumferential strain by the computer phantom and global circumferential strain quantified by the Strain module. In the left panel the dashed line is the line of identity. In the right panel the solid line is the mean bias and the dashed lines 2 SD. Please note some symbols have been superimposed. Bias = -1.0 ± 2.0 Figure 3. Relationship between simulated global radial strain by the computer phantom and global radial strain quantified by Strain module. In the left panel the dashed line is the line of identity. In the right panel the solid line is the mean bias and the dashed lines 2 SD. Please note some symbols have been superimposed. Page 5/9
in patients Figure 4 and Table 3 present the relationship between global strain from Doppler echocardiography and global strain quantified by the Strain module in Segment software packages. Figure 5 and Table 4 present the relationship between segmental strain by Doppler echocardiography and segmental strain quantified by the Strain module in Segment software package. The maximum difference on segmental level was 17 (circumferential strain -36 by Doppler and -19 by Strain module in anterior septal for patient 1). Figure 4. Relationship between global circumferential strain by Doppler echocardiography and global strain quantified by the Strain module. In the left panel the dashed line is the line of identity. In the right panel the solid line is the mean bias and the dashed lines 2 SD. Strain by Doppler Strain by Strain module Patient 1-32 -26 Patient 2-18 -16 Patient 3-30 -26 Patient 4-22 -23 Patient 5-30 -24 Patient 6-26 -22 Patient 7-29 -23 Patient 8-16 -20 Patient 9-16 -15 Table 3. Comparison between global circumferential strain by Doppler and by the Strain module in Segment software packages. Page 6/9
Figure 5. Relationship between segmental circumferential strain by Doppler echocardiography and segmental strain quantified by the Strain module. In the left panel the dashed line is the line of identity. In the right panel the solid line is the mean bias and the dashed lines 2 SD. Please note some symbols have been superimposed. Page 7/9
Strain by Doppler Strain by Anterior Patient 1-39 -35 Patient 2-20 -19 Patient 3-32 -26 Patient 4-23 -28 Patient 5-33 -24 Patient 6-30 -19 Patient 7-35 -25 Patient 8-24 -25 Patient 9-22 -14 Anteroir septal Patient 1-36 -19 Patient 2-24 -18 Patient 3-32 -27 Patient 4-32 -29 Patient 5-24 -18 Patient 6-31 -22 Patient 7-35 -25 Patient 8-17 -23 Patient 9-14 -12 Septal Patient 1-29 -27 Patient 2-20 -15 Patient 3-29 -29 Patient 4-33 -26 Patient 5-29 -28 Patient 6-30 -22 Patient 7-26 -18 Patient 8-11 -26 Patient 9-14 -9 Inferior Patient 1-23 -30 Patient 2-15 -14 Patient 3-27 -28 Patient 4-25 -25 Patient 5-36 -30 Patient 6-26 -22 Patient 7-25 -27 Patient 8-8 -18 Patient 9-11 -18 Posterior Patient 1-29 -30 Patient 2-15 -18 Patient 4-12 -25 Patient 6-20 -28 Patient 7-22 -24 Patient 8-13 -20 Patient 9-13 -25 Lateral Patient 1-37 -27 Patient 2-16 -20 Patient 4-9 -15 Patient 6-22 -27 Patient 7-28 -23 Patient 8-21 -14 Patient 9-21 -15 Table 4. Comparison of segmental circumferential strain by Doppler and by the Strain module. Page 8/9
Conclusion The Strain module in Segment software packages allows quantification and visualization of myocardial strain from tagged MR images. The proposed strain module can quantify both radial and circumferential strain with good agreement to simulated strain values, and quantify global circumferential strain with good agreement to strain in Doppler echocardiography. On segmental basis, there was a moderate correlation for circumferential strain between strain from Strain module and Doppler echocardiography. Hence, the proposed Strain module can provide clinically relevant quantification and visualization of myocardial global strain in tagged MR images. References [1] Duncan AE, Alfirevic A, Sessler DI, Popovic ZB, and Thomas JD, Perioperative Assessment of Myocardial Deformation. Anesth Analg 2014 Mar;118(3):525-44 [2] Cerqueira MD, Weissman NJ, Dilsizian V, Jacobs AK, Kaul S, Laskey WK, et al. Standardized myocardial segmentation and nomenclature for tomographic imaging of the heart: a statement for healthcare professionals from the Cardiac Imaging Committee of the Council on Clinical Cardiology of the American Heart Association. Circulation. 2002;105:539-42. Page 9/9