WO2017001208A1

WO2017001208A1 - Method for estimating a displacement of an structure of interest and magnetic resonance imaging system

Info

Publication number: WO2017001208A1
Application number: PCT/EP2016/063970
Authority: WO
Inventors: Erkki Tapani VAHALA; Jukka Ilmari TANTTU
Original assignee: Koninklijke Philips N.V.
Priority date: 2015-06-30
Filing date: 2016-06-17
Publication date: 2017-01-05

Abstract

It is an object of the invention to provide for a method and system for estimating a displacement of an structure of interest. This object is achieved by a method according to a method for estimating a displacement of a structure of interest comprising the following steps performing a first and a second magnetic resonance imaging sequence on an structure of interest and acquiring a corresponding first and second signal, wherein the second magnetic resonance imaging sequence is a gradient echo sequence and reconstructing a first image from the first signal and; reconstructing a second image from the second signal and performing an image registration between the first and second imageby using both magnitude and phase data of the second image estimating a displacement of the structure of interest between the acquisition of the first and second signal based on the registration between the first and second image.

Description

Method for estimating a displacement of an structure of interest and magnetic resonance imaging system

FIELD OF THE INVENTION

This invention relates to the field of magnetic resonance imaging (MRI) and more specifically to the field of MRI guided therapy. BACKGROUND OF THE INVENTION

One of the targets of an MR-Linac system (MRL) is to provide tools to estimate the movement of selected organs in near real time manner. Typically the organs of interest are the target (e.g. a tumour) to be radiated, or organs at risk (OAR). OAR's are tissues sensitive to radiation which are close to a target volume, e.g. tumour and treatment margin.

The definition of the temporal resolution and latency in near real time imaging depend on the application. In the case of e.g. abdominal organs like kidney the requirement may mean that movement estimate shall be updated 5 times in second or more.

Especially, in an MRI guided therapy application, like e.g. MR-Linac one has to be able to detect motion based on images which have limited signal to noise ration (SNR) and contrast. In such conditions e.g. edge detection capabilities of organ tracking software may easily fail. Methods to separate fat and water containing tissues could improve this since many internal organs are at least partly surrounded by fat. Existing MR methods for fat water separation typically increase the imaging time substantially.

Typical tools for motion detection to image either ID, 2D or 3D MR imaging.

ID MR images are often called navigators. Differentiation of fat and water containing tissues may be important in motion detection. Many internal organs are water containing and surrounded by fat layer. There are known methods for fat/water differentiation, like DIXON, PROSET or spectral fat suppression. The problem is that these methods easily increase the scanning time by factor of 2 or more. SUMMARY OF THE INVENTION

It is an object of the invention to provide for a method and system for estimating a displacement of an structure of interest. This object is achieved by a method according to claim 1.

This object is also achieved by a system according to claim 12 and a computer program product according to 14.

According to a embodiments of the invention the first magnetic resonance imaging sequence results in a reconstructed image with higher resolution, signal to noise ratio and/or contrast than reconstructed second image. This is advantageous, because the first image with higher resolution, signal to noise ratio and/or contrast may improve image segmentation and thereby the estimation of the displacement. The (complex) signal with higher resolution, signal to noise ratio and/or contrast may for example be acquired during a period of expected limited motion of the structure of interest (e.g. during full inspiration or expiration) more time is available to acquire an image with higher resolution, signal to noise ratio and/or contrast and the second magnetic resonance imaging sequence may be applied during a period wherein more motion of the structure of interest is expected (e.g. between full inspiration and expiration). Alternatively, the first magnetic resonance sequence is performed in an interleaved manner.

When the first and second image have been acquired at the same position such as respiratory cycle or respiratory cycle position, one can assume that the organs in image sets produced by both images are in the same position. An offset can be determined between a tissue border of the structure of interest displayed in the first image and a feature in the second image. This feature could for example be comprised in a signal void as will be explained below. This can be achieved in multiple ways. If for example scan-specific auto- segmentation algorithms are applied on image the first and second image, a result is two segmentation meshes where the difference in position is explained by the differing contrast, e.g., a thick void signal in the second image, which could be a balanced gradient echo image like a bFFE image vs. a clinically meaningful tissue border elicited in the first image. This embodiment allows determination of an offset which needs to be applied when calculating the structure displacement with the help of the subsequent second image. The subsequent second image is of the same type as the second image (e.g. in terms of resolution, contrast etc. ) Alternative ways may be used to determine the offset. For example image registration between the first and second image may be performed to determine an offset between a feature in the second image and a tissue border of the structure of interest in the first image. By determining the offset one can benefit from the fast acquisition rate of the second image and still calculate the position of the clinically meaningful tissue edge (instead of the apparent tissue edge given by the balanced gradient echo image). The first and second sequence could be interleaved to reduce a chance of motion in between and so to improve a reliability of the determined offset.

According to a further embodiment of the invention after performing the first magnetic resonance imaging sequence a plurality of second magnetic resonance imaging sequences is performed and the image registration is performed between data derived from the first image and a plurality of second images. This is especially advantageous when the the first magnetic resonance imaging sequence results in reconstructed image with higher resolution, signal to noise ratio and/or contrast than the reconstructed second image. The plurality of second magnetic resonance imaging sequences can be performed during a period wherein more motion of the structure of interest is expected. Since the second magnetic resonance imaging sequence is fast compared to the first magnetic resonance imaging sequence, the second magnetic resonance imaging sequences can be used for real-time tracking. By image registration of the images resulting from the second magnetic resonance imaging sequences with the image resulting from the first magnetic resonance imaging sequence the displacement estimation may be more accurate, as the image resulting from the first magnetic resonance imaging sequence has a higher resolution, signal to noise ratio and/or contrast. The first and the second image do not necessarily need to be acquired on the same MRI system. Also the second and subsequent second image do not necessarily need to be acquired on the same MRI system. For example the first image and potentially the second image may be acquired on 1.5 or 3T system, whereas the second image and / or subsequent second image may be acquired at a system with a lower magnetic field strength such as a 0.2T system.

It is an insight of the inventors that for estimating a displacement of an structure of interest by means of magnetic resonance imaging (MRI) it would be a good choice to have an MRI sequence which has high signal to noise ratio (SNR) within unit time, useful contrast, and small spatial distortions. A structure as used herein could for example be (part of) an organ or tumour. Gradient echo sequences (preferably steady-state gradient echo sequences) like, Balanced Fast Field Echo (bFFE) or Fast Field Echo (FFE) would be good candidates for this purpose. These sequences can be made relatively fast. Especially bFFE has high SNR per unit time. Image contrast of both FFE and bFFE is somewhat limited. FFE is typically Tl weighted. It can be made more proton density weighted or even T2* weighted with loss of SNR and speed. bFFE has a specific contrast that is proportional to the ratio T2/T1.

Both in FFE and bFFE fat has relatively high signal, but the signal intensity alone is not always enough to make distinction from other tissues. It is an insight of the inventors that phase of fat signal in both FFE and bFFE may be used to improve the specificity. The phase behavior of the MRI signal is dependent on the choice of sequence parameters. Fortunately, fast sequences at 1.5T are naturally such that the phase behavior is favourable for tissue differentiation. The invention describes a segmentation process where a segmentation shape detection uses the complex MRI data instead of the normally used magnitude or phase images. By means of the invention phase imaging in combination with magnitude imaging can be used by an MR Linac to enhance tissue differentiation capabilities of the MR Linac near real time. In addition, it describes a general segmentation process where the segmentation shape detection uses the complex MRI data instead of the normally used magnitude or phase images. Using information about the phase on top of information of the magnitude of an MRI signal is advantageous as it does not increase scanning time.

According to embodiments of the invention the first magnetic resonance sequence is a gradient echo sequence. Preferably, the first magnetic resonance sequence is the same as the second magnetic resonance sequence. The use of two the same sequences leads to images with a similar contrast, which in turn may lead to an improved image registration result and a better estimation of the structure displacement.

According to a further embodiment of the invention, using magnitude and phase data for the image registration means that the first and second image used for image registration comprise complex data. This is advantageous, because in this way valuable phase information can be used in a relatively fast way.

According to a further embodiment of the invention, the image registration is only performed on a region of interest within the first and/or second image. This is advantageous, because it may make the method faster.

According to a further embodiment of the invention prior to the image registration a phase correction is performed on the second image and the image registration is performed between the first image and the phase corrected second image. A phase correction could for example be phase unwrapping or other methods for correcting the phase as known in the art. This is advantageous because it may improve the registration result. After phase correction for image registration, the second image could either comprise complex data or a real part of the phase corrected second image. This embodiment is advantageous, because it may improve an immunity of the images against noise. This embodiment is slower than directly using the complex signal. Therefore, this embodiment is especially advantageous when it is performed on a region of interest within one or both of the images.

According to a further embodiment of the invention, a set of balanced gradient echo (preferably bFFE) sequence parameters are chosen such that the complex signal coming from fat and the complex signal coming from water have a phase difference about pi. The consequence of this is that fat in a real image has a negative intensity compared to water. Hereby edges of the structure of interest may be easier to detect.

According to a further embodiment of the invention a metric used for the registration is cross-correlation or mutual information.

According to a further aspect, the invention is a magnetic resonance imaging system configured for performing one or more of the methods described above. This is especially advantageous when real-time tracking is required. According to an embodiment the invention the magnetic resonance imaging system further comprising a radiotherapy delivery system configured for irradiating a patient based on images acquired by the magnetic resonance imaging system. This is advantageous, because real-time tracking during radiotherapy may lead to a decrease in treatment margins.

These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

BRIEF DESCRIPTION OF THE FIGURES

Figure 1 illustrates diagrammatically a magnetic resonance imaging system in which the invention is used and

Figure 2 shows a typical bFFE image obtained from kidneys 204 with a scanning time of about 150 ms and

Figure 3 demonstrates how the phase information may improve the immunity of the images against noise and

Figure 4 shows an example of the results achieved by means of the invention and

Figure 5 illustrates an example of how embodiments of the invention can be used during MRI guided radiotherapy. DETAILED DESCRIPTION OF THE INVENTION

Figure 1 illustrates diagrammatically a magnetic resonance imaging system in which the invention is used. The magnetic resonance imaging system comprises a main magnet 10 which generates a steady homogeneous main magnetic field within the examination zone 14. This main magnetic field causes a partial orientation of the spins in the patient to be examined along the field lines of the main magnetic field. An RF system 12 is provided with one or more RF antennae to emit an RF excitation electromagnetic field into the examination zone 14 to excite spins in the body of the patient to be examined. The relaxing spins emit magnetic resonance signals in the RF range which are picked up by the RF antennae, notably in the form of RF receiving coils 12. The RF system may be coupled to an Tx/Rx switch (TRSwitch) 11, which in turn is coupled to an RF amplifier (RFamp) 13. Further, gradient coils 16 are provided to generate temporary magnetic gradient fields, notably read gradient pulses and phase encoding gradients. These gradient fields usually are orientated in mutual orthogonal directions and impose spatial encoding on the magnetic resonance signals. Gradient amplifiers 18 (GradAmp) are provided to activate the gradient coils to generate the magnetic gradient encoding fields. The magnetic resonance signals picked up by the RF receiver antennae 12 are applied to an MRI data acquisition system (MRacq) 19. The MRI data acquisition system 19 provides the data to a host computer (HC) 20, which in turn provides it to a reconstructor (Recon) 22, which may reconstruct an image from the data. These data may be displayed on a display (Disp) 17.

The magnetic resonance imaging system comprises a radiotherapy delivery system. The radiotherapy delivery system (RT) 32 includes a housing 30 or other support or body supporting a radiation source arranged to move or revolve around the subject. The radiotherapy delivery system 32 may contain a multi-leaf collimator (MLC). The

combination of the multi-leaf collimator with the motion of the radiation source around the subject allows the delivery of complex dose distributions by means of for example arc therapy or intensity modulated radiation therapy. Motion of the structure of interest can be compensated for by means of motion compensation software and/or hardware 40. Examples of motion compensation that can be performed by means of hardware are movement of an imaging table (IM) 34 or movement of the leaves in the MLC. An example of motion compensation by means of software could be online recalculation or updating of the radiotherapy plan, e.g. by means of choosing from an atlas of precalculated radiotherapy plans, by means of a radiotherapy plan calculator (RPC) 36. The MRI system is configured to perform an MRI sequence. A magnetic resonance imaging sequence is a preselected set of RF and/or magnetic gradient pulses and time spacing between these pulses; used in conjunction with magnetic field gradients and MRI signal reception to produce MRI images. A gradient echo sequence is an example of a MRI sequence. In a gradient echo sequence, a gradient is applied for a limited time in a readout direction, following an excitation pulse. This causes spins to precess at different rates, according to their position along a read-out direction (x-axis). Hereby, dephasing occurs and a signal from the spins drops. The gradient is then reversed , so that the spins that were in a positive field are now in a negative field and vice versa. This reversal leads to a rephasing of the spins, and an echo signal. The gradient echo sequence is compatible with a use of smaller flip angles, shorter echo times (TE) and shorter repetition times (TR) compared to spin echo sequences. When a shorter TR is used, the excitation pulse will be less than 90 . Following a limited flip angle excitation (e.g. 15°-45°), a time rquired to restore normal longitudinal magnetization is reduced. Under these conditions , a spin system reaches a steady state after a few pulses in which a significant measurable transverse magnetization component is generated by each pulse, while most of the magnetization remains in the longitudinal direction. These gradient echo sequences are known under different names depending on a vendor and thereby an implentation on the MRI system. The gradient echo sequence is for example known as fast field echo (FFE), fast low angle single shot (FLASH) or gradient recalled acquisition in the steady state (GRASS).

The gradient echo sequence can be performed with a balanced gradient.

"Balanced" means that a net gradient-induced dephasing over a TR interval is about zero. The balanced gradient will act on stationary spins on resonance between two consequtive RF pulses and return them to the same phase as they had before the gradients were applied. An example of a balanced gradient echo sequence is a balanced fast field echo sequence (bFFE).

The MRI signal has a magnitude and a phase. From this signal an MRI image can be reconstructed. This image reconstruction is well known in the art and will not be discussed in further detail. Figure 2 shows a typical bFFE image obtained from kidneys 204 with a scanning time of about 150 ms. Typically in clinical practise only a magnitude image 201 will be used. A corresponding phase image 202 shows different appearance of water 210 and fat 220 containing tissues. The phase image 202 also contains errors 230. Phase images are sensitive to various errors including B0 and RF inhomogeneities. Furthermore, since phase is only defined in a range of ±π there is a possibility for ambiguity whenever an actual phase exceeds this range. This is seen as a so called phase wrap artifact. MRI data can also be represented as a real 203 and imaginary image.

Balanced gradient sequences like bFFE have a unique feature that within certain sequence parameter choice the phase of fat and water signal differ about π. The consequence is that fat 220 in the real image 203 has high negative intensity (negative signal shown as dark gray scale) compared to water 210 containing tissues. This makes image segmentation easier. For π phase difference, signal void detection and segmentation and edge detection algorithms may pick up the strong contrast difference. E.g., Atlas-based meshes could be trained to snap to the signal voids surrounding organs.

Actually the MRI signal in 203 is already manipulated by removing most prominent phase errors. This correction can be easily done for individual images in small field of view. 203 illustrates the inherent power of phase data. In practice the phase itself is not essential, but rather the local phase differences.

Figure 3 demonstrates how the phase information may improve the immunity of the images against noise. The top row show real images 303 and the bottom row shows magnitude images 301. In Figure 3, different amounts of artificial random noise are added in the original complex image (increasing noise level in direction from left to right images). Even with the highest amount of extra noise (on the right) the real image 303 shows the basic structure of the original image, especially when comparing to the corresponding magnitude image (bottom row).

In principle in an complex bFFE image majority of the signal can be arranged to be in the real part of the image. For that one has to correct the phase errors e.g. in the magnet, gradients, RF field and in the signal chain. This can be done, but it is challenging to create a correction which would be fast and robust in all possible clinical conditions.

In FFE images the phase data would also be beneficial. In FFE the phase difference between fat and water is proportional to the echo time. Phase difference can be anything within ±π. With a reasonable choice of sequence parameters a suitable phase behavior can be found.

An example of a suitable TR for bFFE to provide fat and water signal in opposite phases at 1.5T would be around 4.6 ms, but it can vary at least +/- 1.5 ms if one takes care of an fO setting. Some MRI machines allow change of the fO setting, e.g. there is possibility to have fO offset retrieved from an fO calibration result for bFFE. Unlike in FFE, if those conditions are valid, the phase difference between fat and water is 180°, even if there are some B0 inhomogeneities. In FFE one can have a phase difference of 180° between fat and water, if the TE is 2.3 ms and if the BO inhomogeneities are small. With phase corrections this is possible, but the robustness of the method is more challenging than for bFFE.

At 3T TR could be about 2.3 ms for bFFE. Also the tolerance is half of that at 1.5T. This is possible but requires more effective gradients.

If the π phase difference between water and fat is utilized in creating an easily detectable magnitude image signal void, the exact, physical locations of tissue edges can be confirmed with an almost concurrent scanning of a protocol where the contrast is known to highlight clinically meaningful tissue edges with sufficient accuracy. In some cases, the same scan can be used, for example, by utilizing the real 203 or phase image 202.

In one embodiment of the invention, almost concurrent scanning is arranged of two scans with different protocols, where the scans share the geometric information. This is particularly useful for imaging tissue that does not move within the timescale of the acquisitions (e.g., where beam-on imaging is used for exception gating or tracking of peristaltic motion).

In another embodiment of the invention the almost concurrent scanning is arranged by interleaved 4D imaging. This is particularly useful for bFFE in the presence of periodic motion, where the steady-state condition can only be relaxed for a short time wherein the bFFE image quality is compromised.

Below one possible implementation of the invention is described.

Motion estimation with complex conjugate

a. Obtain the first image using bFFE sequence.

b. Define a region of interest (ROI) in the first image. This step can be manual or automatic based on some segmentation algorithm.

c. Obtain series of second images from the same imaging plane and with the same imaging parameters as the first image.

d. Calculate a complex correlation c between the first image and individual second images. Do a summation over the ROI. Assume translational motion between scans. Translate the new image until c is maximized.

e. The result of step d can be considered the best estimate for the translational motion between the first image and the second image.

The motion estimation step d is computationally rather light. Using IDL and an old laptop computer the calculation time is in the order of 50 ms/image. Figure 4 shows an example of the results achieved by means of the invention. The first image 403 with the defined ROI 405 around the kidney is on the left. On the right there is the second image 404 with moved kidney, the original ROI 405 and the new ROI 406 with best correlation with the contents of the original ROI in the first image.

Figure 5 illustrates an example of how embodiments of the invention can be used during MRI guided radiotherapy. Time axis 501 represents a state of the radiotherapy system, wherein the radiotherapy system is switched on during period 504. Time axis 502 represents a state of the MRI system. During period 506 the first signal is acquired and during period 505 multiple second sequences are being performed resulting in multiple second signals. A single second sequence has a shorter duration than the first sequence performed during 506. As a result the resolution and SNR of the first signal are higher. As a result, an edge detection algorithm can provide reliable estimates about a tissue border of the structure of interest. Line 503 represents a breathing signal. As can be seen MRI imaging is only performed during full inhalation. Also radiotherapy is performed only during full inhalation. The second signal could be acquired during irradiation of the patient by means of a radiotherapy system, e.g., by means of echo planar imaging (EPI) or FFE sequence with coarse resolution for maximal temporal resolution. The edge detection algorithm output is compared with the results from the first signal and border offsets calculated, e.g., with locally affine registrations algorithms. In subsequent beam-on imaging, the offsets are applied to on- the-fly calculated contours.

In 4D imaging, a portion of the periodic motion is dedicated for acquiring the first signal in an interleaved manner. For example, the maximum exhalation peak is used to give a timeslot to the first signal k-space line acquisitions that is small enough compared to magnetization relaxation times (e.g., < 100ms), so that after the timeslot, bFFE steady-state can be recovered in an accelerated mode. The resulting images from the peak exhalation, both with the first sequence and the second bFFE sequence, can then be compared and offsets applied to on-the-fly calculated contours. The method can be further improved by

incorporating other measurement points to the respiration cycle. E.g., comparable images are acquired both at maximum inhalation and exhalation points, and offsets can be interpolated to intermediate respiratory points.

An example of this is explained below. When the first and second image have been acquired at the same respiratory cycle, one can assume that the organs in image sets produced by both images are in the same position. If scan-specific auto-segmentation algorithms are applied on image the first and second image, a result is two segmentation meshes where the difference in position is explained by the differing contrast, e.g., a thick void signal in the second image, which could be a balanced gradient echo image like a bFFE image vs. a clinically meaningful tissue border elicited in the first image. This embodiment allows determination of an offset which needs to be applied when we calculating the structure displacement with the help of subsequent second image. Hereby one can benefit from the fast acquisition rate of the second image and still calculate the position of the clinically meaningful tissue edge (instead of the apparent tissue edge given by the balanced gradient echo image).

Whilst the invention has been illustrated and described in detail in the drawings and foregoing description, such illustrations and description are to be considered illustrative or exemplary and not restrictive; the invention is not limited to the disclosed embodiments and can be used for estimation of structure displacement.

Claims

CLAIMS:

1 A method for estimating a displacement of a structure of interest comprising the following steps,

performing a first and a second magnetic resonance imaging sequence on an structure of interest and acquiring a corresponding first and second signal, wherein the structure of interest is at the same position during acquisition of the first and second signal and and

reconstructing a first image from the first signal and;

reconstructing a second image from the second signal wherein the first image has a higher resolution, signal to noise ratio and/or contrast to noise ratio than the second image and

determining an offset between a tissue border of the structure of interest displayed in the first image and a feature in the second image and

acquiring a subsequent second image and

estimating a displacement of the structure of interest by using a position of the feature in the subsequent second image and the determined offset.

2. A method for estimating a displacement of a structure of interest according to claim 1 , wherein the offset is determined by means of a registration between the first and second image or by means of segmentation of the structure of interest in the first and second image.

3. A method for estimating a displacement of a structure of interest according to claim 1 or 2, wherein the first and second magnetic resonance sequence are performed in an interleaved manner

4. A method for estimating a displacement of a structure of interest as claimed in in any of the preceding claims, wherein the first magnetic resonance imaging sequence and / or second magnetic resonance imaging sequence is a gradient echo sequence..

5. A method for estimating a displacement of a structure of interest according to any of the preceding claims, wherein for the image registration the magnitude and phase data of the first image are used

6. A method for estimating a displacement of a structure of interest according to any of the preceding claims, wherein image registration is only performed on a region of interest within the first and/or second image.

7. A method for estimating a displacement of a structure of interest according to any of the preceding claims, wherein prior to image registration a phase correction is performed on phase data of the second image and wherein the image registration is performed between the first image and the phase corrected second image.

8. A method for estimating a displacement of a structure of interest according to claim 5, wherein the second image used for image registration comprise either complex data or a real part of the phase corrected second image.

9. A method for estimating a displacement of a structure of interest as claimed in any of the preceding claims, wherein the gradient echo sequence is a balanced gradient echo sequence and wherein a set of balanced gradient echo sequence parameters are chosen such that the complex signal coming from fat and the complex signal coming from water have a phase difference about pi.

10. A method for estimating a displacement of an structure of interest as claimed in any of the preceding claims, wherein after performing the first magnetic resonance imaging sequence a plurality of second magnetic resonance imaging sequences is performed and the image registration is performed between first image and a plurality of second images.

11. A method for estimating a displacement of a structure of interest as claimed in any of the preceding claims wherein a metric used for the registration is cross-correlation or mutual information.

12. A magnetic resonance imaging system configured for performing a method as described in any of the preceding claims.

13. A magnetic resonance imaging system as claimed in claim 12, further comprising a radiotherapy delivery system configured for irradiating a patient based on images acquired by the magnetic resonance imaging system.

14. Computer program product comprising program code means for causing a computer to control an apparatus as claimed in claim 12 to carry out the steps of the method as claimed in claims 1-11 when the computer program is carried out on the computer.