The accuracy of setting the null TI in a clinical setting depends on operator expertise, contrast agent clearance rate, and patient tolerance to additional breath-hold acquisitions. This exhibits a wide variability in practice. Using phase-sensitive reconstruction in IR delayed hyperenhancement studies, it is possible to use a nominal value of TI, eliminate several breath-holds otherwise needed to find the optimal TI, and achieve a consistent contrast. The phase-sensitive reconstructed images have polarity restored, and after the window and level are adjusted a consistent contrast is achieved over a wide range of TIs, without artifacts due to incorrect polarity. As a result of surface coil intensity normalization, the window and level of displayed intensities may be adjusted after the fact to null infarct, blood, or normal myocardium across the entire heart.
The phase-sensitive reconstruction method dramatically reduces the variation in apparent infarct size (and appearance) that is observed in the magnitude images as TI is changed. This is believed to be a result of partial volume effects, which have a manifestly different appearance for the two methods, and which alter the apparent position of the infarct boundary in magnitude images. For the magnitude images, the profile of the infarcted region is also altered to a slight degree by the noise bias, whereas the phase-sensitive images are unbiased. A typical eight-slice, short-axis stack acquisition may take as long as 5-7 min, during which time the optimum null time will typically increase by approximately 15-25 ms (at a nominal 15 min from a double dose of contrast agent). The loss in CNR (MI-to-normal myocardium) for conventional magnitude detection with a TI set 15 ms earlier than the null TI is approximately 25%. With phase-sensitive reconstruction the polarity is restored, avoiding this severe loss. Loss in contrast due to loss of polarity in magnitude images was verified by acquiring patient data at various TIs and times from dose ().
For the purpose of detecting and sizing regions of MI, the contrast and CNRs between the MI and normal myocardium, and between MI and blood are the most relevant metrics. The contrast of the MI to blood and normal myocardium is affected by a number of variables, such as the TI, pulse sequence parameters (5
), and clearance rate of the contrast agent, which depends on the specific patient and elapsed time from dose. Generally, the CNR between MI and normal myocardium is much greater than that between the MI and blood, as seen in by noting the difference between the Mz
values for a given TI. The low CNR between MI and blood, particularly at a short elapsed time from dose, makes it difficult to accurately detect and size subendocardial infarcts. With TI set to null the normal myocardium, the MI and blood have a fairly high SNR; thus, the MI-blood CNR is approximately the same for both magnitude and phase-sensitive images, as may be seen by the distributions (). In situations where the MI and blood have lower SNR, such as setting TI for infarct nulling (6
), the improved SNR using the phase-sensitive method leads directly to an improved CNR. The SNR advantage of the phase-sensitive method may also be more pronounced at lower dosages of contrast agent, or after stress studies that have a more rapid washout of contrast agent.
The accuracy of CNR measurements is limited by small ROI sizes and inhomogeneous intensity. Therefore, the SNR performance of this method was validated using phantom data and corresponding simulations. The intent in this work was to characterize any significant alteration (gain or loss) in contrast due to phase-sensitive image reconstruction and/or surface coil intensity normalization, rather than to characterize the contrast or contrast mechanisms in delayed hyperenhancement imaging. Due to the low SNR regime of the reference, particularly in the inferior region of the myocardium, the B1-weighted phased array combining, which was performed prior to the phase-sensitive detection, provided an advantage over performing phase-sensitive detection on a coil-by-coil basis prior to array combining. It was found that despite the low SNR of the reference image, errors in the background phase estimate contributed negligibly to SNR loss. The phantom experiment reasonably emulated the characteristics that are important for validating the SNR of the phase-sensitive reconstruction method. The SNR values of the phantom image bracketed the SNR values for the cardiac application, and the T1 was a mid-range value between that of blood and normal myocardium tissue at 15 min after administration of a double dose of contrast agent. The anterior and posterior RF receive coils were spaced somewhat closer than the human torso; however, this did not affect the significance of the results.
The most significant loss in SNR was due to surface coil intensity normalization. Nevertheless, the SNR of the intensity normalized phase-sensitive images exceeded the SNR of the corresponding magnitude image under most circumstances. It is noted that while the noise component due to intensity normalization is correlated over a small region (due to spatial smoothing of the reference image), the CNR of regions spaced greater than approximately 7 pixels incur this loss. However, most importantly, there is negligible loss in contrast between blood and MI. The loss in SNR due to a noisy reference image is reduced by means of smoothing the reference image. The degree of smoothing was limited to avoid significantly altering the image. Breath-hold registration is less critical for intensity normalization than for background phase, since the surface coil sensitivity varies relatively smoothly across the heart region while strong susceptibility-induced gradients cause rapid phase variation. It may be possible to use additional reference data acquired separately to further improve the intensity normalization; however, it was deemed that the improvement would be slight, with a moderately large increase in complexity.
The contrast (average value of 23% between blood and myocardium) of the reference image, due to several mechanisms, slightly reduced the image contrast between MI and normal myocardium of the normalized image. The contrast mechanisms included proton density, T1
, and inflow. The contrast due to T1
-weighting was observed primarily at increased heart rates. At an increased heart rate the magnetization may not be fully recovered even after two heartbeats, thereby causing additional T1
-weighting of the reference and a slight shift in the null time for normal myocardium in the IR image. Inflow effects may contribute to the contrast, particularly when there are timing errors caused by heart rate changes. These result in acquisition at cardiac phases where there is increased flow. The CNR between blood and myocardium is typically quite high (5
). Furthermore, there is already a CNR gain due to phase-sensitive detection (10-20%). Therefore, loss in contrast due to surface coil intensity correction is traded for improved display characteristics.
For magnitude images, there is little to no contrast between the nulled normal myocardial and the lung, with the exception of instances in which there is a layer of fat, fluid, or fibrous tissue. However, the myocardium-lung boundary is fairly well defined in the reference image. As a result, the intensity normalized phase-sensitive images have better definition of the lung tissue boundary, since the lung is much noisier relative to the myocardium. This can be observed by comparing (magnitude) with (phase-sensitive). Improved contrast between the epicardium and lung should facilitate determining the transmural extent of anterior, anterolateral, and posterior infarcts. This may also be realized by measurements using both IR and reference magnitude images.
Using the method described here, the conventional magnitude image, as well as the phase-sensitive and reference images, is produced for each breath-hold acquisition. This is useful for comparison purposes. In a few cases for which there were severe microvascular obstructions, leading to a dark core in the MI of the magnitude image, the reference image had a corresponding T1-weighted dark area. In these cases, the reference is useful as an aid to interpreting the resultant magnitude image, and if the obstruction is so severe as to cause a sign error in the phase-sensitive image (which is rarely observed), the reference image may easily be used to resolve the ambiguity. This has only been observed in one out of 25 cases studied.
It is worth noting the difference between the point-spread functions for magnitude and phase-sensitive detection methods. The IR (combined with a multishot readout) leads to a nonuniform k-space weighting, which can give rise to image artifacts. The k-space weighting is nonuniform, since each inversion (heartbeat segment) has 16 phase-encodes with the pulse sequence parameters used. The phase-encode order is interleaved such that the overall weighting is generally increasing in a smooth manner. Note that the variation from segment to segment is quite low, since the magnetization has almost fully recovered after two heartbeats. The asymmetrical weighting leads to a complex point spread function, for which the primary distortion is an artifact caused by the imaginary component. The artifact appears as spatial differentiation along the phase-encode direction, which enhances the edges and causes ringing at regions with rapid amplitude or phase changes in the complex image. For phase-sensitive detection, the imaginary component is discarded, thus eliminating the edge artifact, provided that the background phase estimate is reasonably accurate. Using magnitude detection (root sum of squares), the quadrature component (imaginary point spread function) is suppressed such that the artifact is generally in the noise. In practice, it has been found that there is little artifact at boundaries such as the myocardium and LV blood pool, because the slope of the IR is quite similar despite the difference in values of T1. There may be circumstances, such as longer segment duration, in which the artifact is not sufficiently suppressed by magnitude detection. In these cases the real point spread function of the phase-sensitive method may have an advantage. Experimentally, there were no image artifacts attributed to the nonuniform k-space weighting.
Finally, since the gated, segmented acquisition uses two heartbeats between inversion pulses for nearly complete magnetization recovery, the acquisition of the additional reference image during alternate beats does not increase the overall breath-hold duration.