Longitudinal assessment of tissue properties and cardiac diffusion metrics of the ex vivo porcine heart at 7 T: Impact of continuous tissue fixation using formalin

In this study we aimed to assess the effects of continuous formalin fixation on diffusion and relaxation metrics of the ex vivo porcine heart at 7 T. Magnetic resonance imaging was performed on eight piglet hearts using a 7 T whole body system. Hearts were measured fresh within 3 hours of cardiac arrest followed by immersion in 10% neutral buffered formalin. T2* and T2 were assessed using a gradient multi‐echo and multi‐echo spin echo sequence, respectively. A spin echo and a custom stimulated echo sequence were employed to assess diffusion time‐dependent changes in metrics of cardiac diffusion tensor imaging. SNR was determined for b = 0 images. Scans were performed for 5 mm thick apical, midcavity and basal slices (in‐plane resolution: 1 mm) and repeated 7, 15, 50, 100 and 200 days postfixation. Eigenvalues of the apparent diffusion coefficient (ADC) and fractional anisotropy (FA) decreased significantly (P < 0.05) following fixation. Relative to fresh hearts, FA values 7 and 200 days postfixation were 90% and 80%, while respective relative ADC values at those fixation stages were 78% and 92%. Statistical helix and sheetlet angle distributions as well as respective mean and median values showed no systematic influence of continuous formalin fixation. Similar to changes in the ADC, values for T2, T2* and SNR dropped initially postfixation. Respective relative values compared with fresh hearts at day 7 were 64%, 79% and 68%, whereas continuous fixation restored T2, T2* and SNR leading to relative values of 74%, 100%, and 81% at day 200, respectively. Relaxation parameters and diffusion metrics are significantly altered by continuous formalin fixation. The preservation of microstructure metrics following prolonged fixation is a key finding that may enable future studies of ventricular remodeling in cardiac pathologies.


| INTRODUCTION
Cardiac diffusion tensor imaging (cDTI) has become an emerging application in the characterization of cardiac tissue and its functional and structural integrity. It has since been applied in a variety of cardiovascular pathologies, such as hypertrophic 1-3 and dilative cardiomyopathy 3,4 as well as myocardial infarction. 5 In recent years the number of in vivo studies in humans [6][7][8] and animals 3,9 has increased. But, due to cardiac motion, breath holds, intrinsically low SNR, [10][11][12] and limited scan times, spatial and angular resolution as well as spatial coverage remain low. Many cardiac studies using DTI have therefore been performed on ex vivo specimens. Usually rapid chemical fixation is applied to prevent autolytic effects after the organ harvest, 13,14 preserving physiological diffusion properties and tissue microstructure as best as possible. This tissue fixation, often performed using formalin, enables long scan times and, thus, ex vivo measurements remain an important scientific tool which provides high resolution and high fidelity, ground truth data. Historically, such data allowed validation of DTI against histology, 9,[15][16][17] proof of concept studies, [18][19][20] as well as validation of in vivo results. 3,21 Furthermore, cardiac tissue samples can often be gained as a byproduct of other scientific or biomedical studies.
The site of cardiac experimentation and fixation, and the site for subsequent DTI measurement may be separated by hundreds of miles. 19 Thus, tissue specimens may be shipped before the MR measurements.
In order to allow an accurate comparison between in vivo and ex vivo measurements it is paramount to assess fixation-induced changes of tissue properties. While there are studies analyzing the impact of tissue fixation on diffusion metrics of the heart, 18,20,22 methodology in studies throughout the years remained highly heterogeneous with regard to the procedure of fixation, storage times and temperature, the fixative itself, and the overall fixation duration.
There is a lack of data on variations of cardiac MRI and DTI parameters following excision of the heart combined with subsequent formalin fixation. For brain tissue, studies have shown that continuous formalin fixation shortens T 1 and T 2 relaxation times compared with the in vivo values [23][24][25][26][27][28][29] and that diffusivity is significantly decreased following fixation. 13,30,31 In a study comparing different fixation methods on the porcine heart, Agger et al 18 concluded that immersion fixation using formalin yielded the diffusion profile most similar to that of "fresh" tissue. In addition, they state that the process of perfusion fixation, in dependence of the perfusion pressure, may possibly inflict tissue damage, resulting in changes of both mean diffusivity and components of the diffusion tensor.
Therefore, the main aim of this study was to assess the impact of formalin immersion fixation on diffusion metrics of the ex vivo porcine heart at 7 T. Diffusion properties for various fixation durations are assessed in dependence of the diffusion time using a stimulated echo diffusion sequence. The secondary aim of this study was to assess fixation-induced alterations in T 2 and T 2 * at ultrahigh field strength and their impact on SNR in diffusion MRI. Results of this study will benefit future ex vivo diffusion experiments regarding fixation, measurement and sequence protocols. In addition, changes in diffusion metrics in dependence of the fixation duration will ease comparison of existing and upcoming studies.

| Porcine heart samples
Hearts (n = 8) were kindly provided by the Translational Center for Regenerative Therapies Wuerzburg, following the animals approved use (55.2 2532-2-256, District Government of Lower Franconia, Germany) in another study. All experiments were performed according to the German Animal Welfare Act and the EU Directive 2010/63/EU. Euthanasia was achieved as described in 32 and hearts were collected, rinsed and stored in physiological saline solution. The eight hearts were harvested in four experiments, where two hearts each were excised with~20 minutes temporal difference. All samples in this study were harvested from male German Landrace piglets, which were obtained from the same breeder. Body weights ranged from 20-22 kg. Details regarding age and body weight are shown in Table S1.

| SAMPLE PREPARATION
Prior to MR measurements and subsequent tissue fixation, both atria were removed, easing the release of air trapped in the ventricles. Sample transport and preparation required~45-60 minutes before measurement of the first heart (set of two). We always started with MRI measurements of the heart that was excised second, in order to minimize the time differences from excision to measurement between hearts within one set. Hearts were centered in a plastic container filled with physiological saline solution and the sample position was fixed using surgical threads. Directly after the initial MR measurement of the fresh heart, each heart was fixed via immersion in 10% phosphate buffered formalin solution. Before MR measurements of fixed specimens, hearts were briefly (~3-5 minutes) rinsed to remove excess formalin. Throughout the study, hearts were stored immersed in formalin in the scanner operating room. This room is temperature-controlled, allowing storage at a consistent 21 ± 1 C.

| General experimental setup
MRI was performed at bore temperature on a 7 T whole body MRI system (Siemens MAGNETOM Terra, Erlangen, Germany) using a 1/32 Tx/Rx head coil (Nova Medical). Data for the two hearts were acquired in consecutive measurements. Figure 1 shows an illustration of the whole study protocol. Initial measurements on the day of excision were performed on fresh hearts.
For MRI measurements, three short axis slices were positioned in relation to the valves in order to reproduce the slice position in subsequent measurements. The distance between 5 mm slices was set to 15 mm, resulting in single basal, midcavity and apical slices. All scans were performed with an in-plane resolution of 1 mm. Prior to measurements we applied third-order B 0 -shimming using the scanner-integrated 3Dshimming algorithm covering the whole three-slice volume. The scan protocol consisted of relaxometry (T 2 , T 2 * ), diffusion (SE and STE) and SNR (diffusion) measurements. An overview of all sequence parameters is displayed in Table S3. Total acquisition time for the protocol was 63 minutes.
Scans for both hearts were therefore completed within 3 hours after excision. Data was postprocessed using MATLAB (MathWorks, Natick, MA) and DSI Studio. 33 Measurements were repeated 7, 15, 50, 100 and 200 days after immersion.

| SNR
In order to evaluate peak SNR in diffusion measurements for the parameters used, we acquired 30 b = 0 images with the spin-echo sequence as well as the stimulated echo sequence (t Mix = 100 ms) for the basal, midcavity and apical slice. Due to a longer diffusion time, the applied spoiler gradients amount to a diffusion weighting of b~18 s/mm 2 in the stimulated echo sequence. Other sequence parameters are described in more detail in the diffusion section below. F I G U R E 1 Study protocol for the assessment of changes in tissue properties and diffusion metrics in the ex vivo porcine heart due to continuous formalin fixation

| Diffusion
Diffusion data were acquired using a Stejskal-Tanner spin-echo sequence and an in-house written stimulated echo pulse sequence. Both pulse sequences used a monopolar diffusion preparation, EPI readout and a GRAPPA acceleration factor of R = 3. In order to minimize TE we used a bandwidth of 2442 Hz/pixel, leading to a bigger FOV. Further measurement parameters were: TR: 3500 ms, FOV: 208 mm x 256 mm, bandwidth: 2440 Hz/pixel, vendor-supplied diffusion directions (b = 1000 s/mm 2 ): 6 (averages: 12), reference images (b = 150 s/mm 2 ): 6 (averages: 12). In order to gain additional information about changes in diffusivity and microstructure, we used varying mixing times (t Mix = 50, 100, 200, 400 and 600 ms) in the stimulated echo sequence, probing diffusion metrics in dependence of the diffusion time. Prolonging the diffusion time allowed shortening of the diffusion gradients, which resulted in shorter echo times. TE for the spin-echo sequence was 44 ms, while TEs for the stimulated echo sequences (t Mix ) were 37, 36, 35, 32 and 32 ms.

| Data processing
All postprocessing was based on DICOM-images created using the vendor's reconstruction pipeline and MATLAB, if not indicated otherwise.

| SNR
Myocardial contours of the left ventricle (LV) were manually segmented and SNR calculated according to the multiple acquisition method. 34 The 30 images were evaluated as a "pseudo" time-dependent dataset, where SNR can be calculated for each voxel (r) using the mean ( x) and standard deviation (σ) over time (t): The resulting values were averaged for the previously created LV ROI.
In addition, we assessed background noise and SNR in the saline solution as a reference, ensuring coil functionality. As described by Reeder et al 34 noise, assessed in a signal-free area, was corrected for Rayleigh distribution.

| Relaxometry
T 2 * and T 2 maps were reconstructed on a pixel basis using an in-house developed MATLAB script. Here a mono-exponential model fit: with the free parameters (T 2 * and S 0 ) was applied to a preliminary normalized data vector S (TE). T 2 decay was fit the same way. Margins for the fitting parameters of the nonlinear solver were set to 1-50 ms for T 2 * and 1-80 ms for T 2 . Providing manually segmented epicardial contours accelerated the fitting process.

| Diffusion
Motion correction was applied in order to correct for eddy current-induced geometrical distortions and DSI Studio 33 was used for reconstruction of the diffusion tensor as described by Jiang et al. 35 All the following processing steps were performed using MATLAB. Eigenvalues (λ 1 , λ 2 , λ 3 ), fractional anisotropy (FA) and the apparent diffusion coefficient (ADC) were derived from the diffusion tensor using eigenvalue analysis and the following two equations: where λ corresponds to the mean eigenvalue of the diffusion tensor.
In addition, we assessed cardiac diffusion metrics such as the primary eigenvector or helix angle (HA) and secondary eigenvector angle (E2A), which correspond to the fiber bundle and sheetlet orientation, respectively. Changes in HA or |E2A| distribution due to fixation were assessed using histograms. Endocardial and epicardial contours of the LV were segmented for subsequent data analysis of diffusion metrics. This also enabled determination of the LV center, which was used to calculate primary and secondary eigenvector angles as described by Ferreira et al. 2 The LV center was recalculated for each time point. Changes in these metrics were also assessed in dependence of the diffusion time up to 600 ms. In order to demonstrate temperature independence, we placed a 20 x 20 pixel ROI in the saline solution for all ADC maps based on the spin echo sequence. This was done for all time points of this study (0, 7, 15, 50, 100, 200 days).

| Statistics
All statistical testing was done using MATLAB. Prior to statistical tests, data were checked for normal distribution using Shapiro-Wilk tests with a significance level of P < 0.05. Fixation-induced changes in T 2 , T 2 * , eigenvalues of the diffusion tensor, ADC and FA were then assessed using a paired t-test with P < 0.05. Bonferroni correction was applied to account for multiple testing.  Table S4 and histograms of the SNR distribution are shown in Figure S1. Using the STE sequence resulted in narrower histograms with sharper modes. Destructive B 1 interference, as illustrated in Figure S2, was present in the saline solution in all diffusion scans. Despite this signal variation, the saline SNR reference was consistent throughout this study. In addition, there were no variations in the noise floor.

| Relaxometry
Representative image contrasts for the varying echo times in T 2 * and T 2 measurements are displayed in Figures S3 and S4 for a midcavity slice of one heart. Alterations of the transverse and the effective transverse relaxation times due to continuous tissue fixation are shown in Figure 3. As visualized by the parameter maps of a representative midcavity slice, both T 2 (see Figure 3A) and T 2 * (see Figure 3B)

| Diffusion
Mean temporal changes in the eigenvalues of the diffusion tensor, the ADC and FA are shown in Figure 4. All eigenvalues and therefore the ADC decreased, similarly to T 2 and T 2 * , within 7 days following fixation. Changes in λ 2 and λ 3 were rather small. After this initial drop, eigenvalues slightly increased with fixation duration. After 100 days of fixation this led to significant differences compared with first week λ 1 and ADC values. Reference values for the ADC are depicted in Figure S5. The average ADC in saline solution was 1.740 ± 0.022 [10 -3 mm 2 /s], with the standard deviation being lower than 2%.
Representative fixation-induced alterations of the main eigenvector orientation and therefore microstructure for heart #5 are displayed in    Figure S10.
While λ 1 continually decreased with increasing diffusion time in the unfixed heart, this trend was not apparent after fixation. Values of λ 1 appear rather consistent with increasing diffusion time. As described above, the main eigenvalue of the diffusion tensor drops following fixation.
In addition, there is a slight increase with increasing fixation duration. Thus, the difference between λ 1 in fixed and unfixed cardiac tissue is smallest for longer diffusion times and tissue fixation durations (<10%).
Differences of the secondary and tertiary eigenvalues following fixation were small compared with the main eigenvalue. Both λ 2 and λ 3 initially decreased with increasing diffusion time, resulting in a rather consistent state for longer diffusion times (>200 ms). This diffusion timedependent decrease was highest in unfixed hearts and on average became less impactful for longer tissue fixation durations. ADC curves exhibit the same diffusion time dependency with smaller differences between fixed and unfixed tissue.

| DISCUSSION
We have demonstrated tissue fixation and fixation duration-dependent alterations of various MRI (T 2 , T 2 * , SNR) and diffusion metrics (ADC, λ 1 , λ 2 , λ 3 , FA, HA, |E2A|) in the LV of the porcine heart at 7 T. Our results show that T 2 and T 2 * relaxation times were continuously altered by immersion fixation in formalin and that SNR values strongly correlated with the respective changes in T 2 and T 2 * . In addition, we found that tissue fixation significantly changed the main eigenvalue of the diffusion tensor, ADC and FA, and that continuous changes in the main eigenvalues, ADC and FA after a fixation duration of 15 days were not significant. By contrast, no systematic changes in structural parameters, such as HA and | E2A|, were introduced by fixation, independent of the immersion duration. This is an important finding for studies assessing ventricular remodeling of cardiac microstructure in various cardiac pathologies.
SNR in diffusion MRI directly influences the data fidelity 11,12 of derived diffusion parameters. Most ex vivo studies use a Stejskal-Tanner diffusion preparation and EPI readout. Thus, SNR strongly depends on both T 2 and T 2 * . Since ex vivo studies are usually performed to achieve high quality and high fidelity data, study protocols should be optimized with respect to SNR. Our results show that SNR decreases (mean:~32%) initially after tissue fixation and that SNR was somewhat restored (mean:~20%) with prolonged tissue fixation. This change in SNR is mainly due to the changes observed in T 2 and T 2 * relative to the used echo times as well as subsequent k-space filter effects. Special care needs to be taken in order to avoid further reduction of T 2 * due to increased B 0 inhomogeneity at ultrahigh field strengths (≥7 T).
Many studies have reported a decrease in T 2 in nervous tissue or other organs following tissue fixation using formalin. [23][24][25][26][27][28][29] To our knowledge such changes in relaxation parameters have not been yet reported for the heart. Initial T 2 changes following tissue fixation observed in this study fit well to the literature values for brain tissue listed above. T 2 shortening in nervous tissue has been described in many studies and has been attributed to multiple causes, such as increased tissue rigidity due to protein cross-linking, interactions of water molecules and myelin lipids in F I G U R E 6 Representative secondary eigenvector angle distribution for one heart prior to fixation and 7, 15, 50, 100 and 200 days after fixation. The diffusion tensor for a basal slice is visualized as a superquadric glyph for each pixel.
Color-coding of each glyph corresponds to |E2A| values. Dotted line: example section at day 0 and respective zoomed images for the various time points. Threshold for visualization of a pixel as superquadric glyph was set to a FA value of 0.2 tissue compartments, 36 changes in diffusion rates, 37 intra/extra-cellular water components, 25 tissue dehydration 28,38 and regular replacements of formalin solution. 27 The analysis of the mechanisms of T 2 shortening and their application to heart tissue was outside the scope of this study.
We rather focused on changes in MR and diffusion metrics induced by continuous tissue fixation in order to (1) ease optimization of future study protocols and (2) enable comparisons of past and future studies regarding fixed versus unfixed tissue, fixation duration, and spin echo versus stimulated echo data.
Dawe et al 23 reported a gradual increase (20%, 60-90 days postmortem) of T 2 values in "deep" brain tissue for longer fixation durations, following the initial decrease after formalin immersion. We observed an increase (~13%) in T 2 between 15 and 50 days of fixation. Dawe et al 23 argue that this recovery in T 2 values is connected to latent tissue decomposition.
In our study, we removed the atria from all hearts in order to ease the removal of trapped air and fixative distribution, allowing penetration of the myocardial wall from both sides. Aldehyde fixation penetration occurs with a rate of 0.5 to 1 mm/hour at room temperature and we observed the pale yellow color change, which indicates penetration, shortly after immersion. Multiple studies have shown that autolytic effects on cardiac tissue are very slow. 14,32,39 It is therefore unlikely that the gradual increase in T 2 at later stages of fixation is related to latent effects of tissue decomposition. During the fixation process cells move through stages of (1) shrinkage, (2) prolonged swelling and (3) secondary shrinkage. As mentioned above, changes in T 2 have been attributed to changes in intra-and extra-cellular volume components. 25 In this study we apply formalin, a neutral buffered, aqueous solution of formaldehyde (systematic name: methanal), with a formaldehyde concentration of 4%. Formaldehyde concentrations below 5% only cause very minimal initial shrinkage followed by extensive swelling 40 and later on shrinkage. Observed decreases and increases in T 2 may thus be caused by cell swelling and shrinkage, respectively. A dedicated study resolving the fixation process itself will be necessary to completely exclude influences of tissue decomposition and variations in both heart size and contraction state. In addition to SNR, alterations in T 2 may possibly change T 2 weightings in the diffusion-weighted images acquired.
F I G U R E 7 Average diffusion time-dependent changes of λ 1 , λ 2 , λ 3 , ADC and FA in the midcavity slice in dependence of the fixation duration Since we did not observe any interference patterns in images for T 2 quantification due to imperfect inversions, normalized the signaltime curves and checked for mono-exponential decay via goodness of fit, we believe that B 1 imperfections did not lead to first-order approximation errors in measured T 2 values. In addition, the goodness of fit showed that the effect of B 1 inhomogeneity on SNR was negligible for the analyzed tissue. Our results show that all eigenvalues decrease following fixation. Alterations of the main eigenvector of diffusion are the most pronounced, and the second and third eigenvalues are less affected. Both Agger et al, 18 for heart tissue, and D'Arceuil and de Crespigny, 13 for white matter, showed that the effects on the main eigenvector are most pronounced. In order to improve our understanding of changes in the ADC and FA it is paramount to report changes in the individual eigenvalues. Future studies analyzing diffusion metrics in highly structured tissue, such as the brain or the heart, should therefore report all eigenvalues.
Agger et al 18 demonstrated that immersion fixation using formalin does not affect main fiber orientation and therefore tractography results. This is in agreement with studies by Scollan et al 16 and Helm et al, 44 who compared the main eigenvector orientation in fixed hearts to myofiber orientation in histology. Results obtained in this study support this observation. Even prolonged, continuous tissue fixation had little impact on the helix angle and therefore main eigenvector orientation. Furthermore, the smooth gradient from endocardium to epicardium was consistent in the presence of scan-to-scan slice offsets.
To our knowledge this is the first study analyzing effects of formalin fixation on the secondary eigenvector orientation within cardiac tissue.
Recent developments in cardiac DTI, particularly in vivo applications, have shown that the secondary eigenvector angle holds high diagnostic value for myocardial remodeling in cardiac pathologies. [2][3][4] In this study we demonstrate that continuous tissue fixation using formalin does not affect the average secondary eigenvector orientation. Sheetlets generally run perpendicular to myocytes, which means through-plane in the short axis slice, and thus scan-to-scan slice offsets are more apparent in the visualization of |E2A|.
Kim et al 45 report a consistent main eigenvalue and decreasing secondary and tertiary eigenvalues for increasing diffusion times in refrigerated and thawed heart tissue of the calf. While this is in good agreement with the changes we found in heart tissue postfixation, we also found that diffusion time-induced alterations in fixed and unfixed tissue follow a similar trend, but with a different scale. Longer diffusion times and therefore increased anisotropy have been shown to decrease uncertainty in the determination of the main eigenvector orientation. 46 In this study we have shown that the dependence of FA from diffusion encoding time is reduced following fixation using formalin, indicating that formalin fixation introduces diffusion barriers along the main direction of diffusion.
Reference values, such as SNR in the saline solution and noise in the background, were meant to remove external factors, such as temperature and coil heating, for the assessment of SNR and diffusion metrics. While the background noise was very stable, SNR in the saline solution was affected by destructive B 1 interference. However, considering the combination of the SNR and the ADC reference in saline solution, we believe that the results of this study are unaffected by temperature variation.
In order to minimize tissue decomposition-induced changes in derived metrics of fresh hearts, we set the scanning protocol to 63 minutes, trading higher SNR for limited volume coverage. The diffusion weighting of b = 1000 s/mm 2 was chosen to maintain comparability to other fixation studies on porcine hearts such as Agger et al 18  have increased the dynamic range for restricted diffusion and enabled analysis based on non-Gaussian models, they also require longer diffusion gradients, leading to longer echo times and a loss in SNR, particularly at 7 T. In this study we show that fixation leads to decreases in T 2 and T 2 * , which are already short at ultrahigh field strengths. An assessment of respective tradeoffs and optimal b-values for ex vivo cDTI of fixed porcine hearts at 7 T may be subject to future studies.
Since we received sets of two hearts per experiment, prefixation scans within one set were performed with a time difference of~40 minutes between excision and measurement. We found no systematic difference in diffusivity between hearts measured first and second. This is in agreement with prior experiments analyzing sample stability of fresh hearts following excision. 32 However, variations in heart size and times for transport, tissue preparation and sample fixation may have resulted in different tissue temperatures. Thus, prefixation scans in this study may therefore include a temperature bias.
Throughout the study, hearts were repeatedly taken from the fixation container and placed in dedicated plastic containers for measurements. Therefore, slight offsets in slice positioning were present in some repeat scans, despite slice positioning with respect to anatomical markers. In general these offsets were more pronounced for the prefixation scans and rather small for scans of fixed hearts. They may thus be related to tissue changes introduced by the fixation process. While these offsets had little impact on derived structural parameters in the LV of fixed samples, there were distinct differences at intersection points to the right ventricle. Particularly at very basal slices, structural coherence at these points was lower. Further analysis of these changes may be part of future studies, but was outside the scope of this work.
In conclusion, we have demonstrated that continuous immersion fixation of porcine hearts using formalin leads to alterations of relaxation parameters and diffusion metrics, such as eigenvalues, ADC and FA. Thus, tissue fixation durations should be kept in mind when interpreting the impact of diseases on these parameters. In addition, our results highlight the importance of proper method selection and reproducible tissue handling, preparation and storage. In recent years, metrics of microstructure been shown to be precise and specific markers for remodeling in cardiac pathologies. [3][4][5]19 The preservation of anisotropy and overall microstructure following even long periods of immersion in formalin may enable future studies assessing ventricular remodeling of cardiac microstructure in various cardiac pathologies in more detail.