WO2006004933A2

WO2006004933A2 - Adaptive digital subtraction for verification of intensity modulated radiation therapy

Info

Publication number: WO2006004933A2
Application number: PCT/US2005/023284
Authority: WO
Inventors: Frank Van Den Heuvel
Original assignee: Wayne State University
Priority date: 2004-06-29
Filing date: 2005-06-29
Publication date: 2006-01-12
Also published as: WO2006004933A3

Abstract

An adaptive digital subtraction system (10) for verifying intensity modulated radiation therapy. The system (10) uses a subtraction technique similar to digital subtraction angiography (DSA), but adapted to work with images obtained from a commercially available amorphous silicon (a-Si) electronic portal-imaging device (EPID) (18).

Description

ADAPTIVE DIGITAL SUBTRACTION FOR VERIFICATION OF INTENSITY MODULATED RADIATION THERAPY

FIELD OF THE INVENTION

[0001] The present invention relates generally to intensity modulated radiation therapy and, more particularly, to an adaptive digital subtraction system for verification of intensity modulated radiation therapy.

BACKGROUND OF THE INVENTION

[0002] Intensity modulated radiation therapy (IMRT) is a type of three- dimensional radiation therapy which improves the targeting of radiation therapy treatments in a way that is likely to decrease damage to normal tissues. The verification of the delivery of intensity modulated radiation therapy treatment (IMRT) using multi-leaf collimator-based delivery systems is not a straightforward task. The methodology for verification essentially splits up the process into three independent verifications: (1 ) absolute dose verification using a phantom plan; (2) relative measurement of the intensity maps using film; and (3) patient position verification using pre-ports, ultrasound alignment or fiducial marker measurements. These steps are labor intensive and require frequent cross-checks by the physics and dosimetry staff. The added workload means that these verifications are generally only carried out once over a full patient treatment.

[0003] Recently, other verifications procedures have been proposed, such as replacing the absolute dosimetry with an independent monitor unit calculation combined with increased multi-leaf collimator quality assurance. Additionally, some researchers use portal imagers as a dosimetry device to perform a task, without the patient in place. Most of the algorithms used in these verification procedures require elaborate calibration sequences, using the imager as a dosimetric device or a device to measure the way leaves are set or the way they travel.

SUMMARY OF THE INVENTION

[0004] In accordance with the general teachings of the present invention, a verification system is disclosed for use in conjunction with intensity modulated radiation therapy treatment. In one embodiment, the verification system combines the verification of the patient's position as well as an intensity map on a daily basis.

The system uses a subtraction technique similar to digital subtraction angiography (DSA) adapted to work with images obtained from a commercially available amorphous silicon (a-Si) electronic portal imaging device (EPID).

[0005] Further areas of applicability of the present invention will become apparent from the detailed description provided hereinafter. It should be understood that the detailed description and specific examples, while indicating the preferred embodiment of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 is a general block diagram of a system for verifying intensity modulated radiation therapy, according to an embodiment of the present invention;

[0007] Figure 2 illustrates a graphical view comparing a patient image and an open image during an IMRT treatment as a function of pixel value vs. pixel number, in accordance with the teachings of the present invention;

[0008] Figure 3a illustrates a photographic view of the patient image of figure 2, in accordance with the general teachings of the present invention;

[0009] Figure 3b illustrates a photographic view of the open image of figure 2, in accordance with the general teachings of the present invention;

[0010] Figure 3c illustrates a photographic view of an image formed by a two-parameter adaptive subtraction system, in accordance with the general teachings of the present invention;

[0011] Figure 4 illustrates a photographic view of the image of figure 3c, in accordance with the general teachings of the present invention;

[0012] Figure 5 illustrates a graphical view of the images depicted in figure 3 as a function of image response vs. pixel number, in accordance with the general teachings of the present invention;

[0013] Figure 6a illustrates a photographic view of a patient image during IMRT prostate treatment, in accordance with the general teachings of the present invention;

[0014] Figure 6b illustrates a photographic view of a subtracted image of figure 6a, in accordance with the general teachings of the present invention; [0015] Figure 6c illustrates a photographic view of a patient image during IMRT prostate treatment, in accordance with the general teachings of the present invention;

[0016] Figure 6d illustrates a photographic view of a subtracted image of figure 6c, in accordance with the general teachings of the present invention;

[0017] Figure 6e illustrates a photographic view of a patient image during IMRT prostate treatment, in accordance with the general teachings of the present invention;

[0018] Figure 6f illustrates a photographic view of a subtracted image of figure 6e, in accordance with the general teachings of the present invention;

[0019] Figure 7a illustrates a photographic view of a patient image during IMRT prostate treatment, in accordance with the general teachings of the present invention;

[0020] Figure 7b illustrates a photographic view of a subtracted image of figure 6a, in accordance with the general teachings of the present invention;

[0021] Figure 7c illustrates a photographic view of a patient image during IMRT prostate treatment, in accordance with the general teachings of the present invention;

[0022] Figure 7d illustrates a photographic view of a subtracted image of figure 7c, in accordance with the general teachings of the present invention;

[0023] Figure 7e illustrates a photographic view of a patient image during IMRT prostate treatment, in accordance with the general teachings of the present invention;

[0024] Figure 7f illustrates a photographic view of a subtracted image of figure 6e, in accordance with the general teachings of the present invention;

[0025] Figure 8a illustrates a photographic view of a patient image during IMRT prostate treatment, in accordance with the general teachings of the present invention;

[0026] Figure 8b illustrates a photographic view of a subtracted image of figure 8a, in accordance with the general teachings of the present invention;

[0027] Figure 9a illustrates a photographic view of an open field image at a 315° gantry angle field, in accordance with the general teachings of the present invention;

[0028] Figure 9b illustrates a photographic view of an open field image at a 45° gantry angle field, in accordance with the general teachings of the present invention;

[0029] Figure 10 illustrates a photographic view of an image wherein the X² value is calculated, in accordance with the general teachings of the present invention;

[0030] Figure 1 1 a illustrates a photographic view of an error field image, in accordance with the general teachings of the present invention;

[0031] Figure 1 1 b illustrates a photographic view of the detected error field image of figure 11a, in accordance with the general teachings of the present invention; and

[0032] Figures 12a-12g illustrate a source code operable to practice the present invention, in accordance with the general teachings of the present invention.

DETAILED DESCRIPTION OF THE EMBODIMENTS [0033] The following description of the embodiments of the invention directed to a verification technique for intensity modulated radiation therapy is merely exemplary in nature and is in no way intended to limit the invention, its application, or uses. [0034] Figure 1 is a general block diagram of a verification system 10 for intensity modulated radiation therapy, according to an embodiment of the present invention. The system 10 includes a radiation source 12 for generating a beam of radiation 14 to generate an image of a patient 16. The radiation source 12 can be any suitable radiation source for the purposes discussed herein. The beam of radiation 14 propagates through an IMRT device 22 that provides beam processing for therapy, as is well understood to those skilled in the art. In one embodiment, the IMRT device 22 is a multi-leaf collimator delivery device. An imager 18 generates the image, which can also be any imager suitable for the purposes discussed herein, such as an EPID. The image is sent to a controller 20 that provides the verification process of the image as will be discussed in detail below. As will be discussed in detail below, the system 10 generates an actual image by the imager 18 with the patient 16 in the beam of radiation 14 and an image without the patient 16 in the beam radiation 14, where the controller 20 subtracts the two images to provide the verification. [0035] An image from an EPID can be thought of as the response of the imager 18 to a particular signal (e.g. variation in dose) over the detector surface. A number of researchers have pointed out that the response of modern flat-panel amorphous silicon (a-Si) detection is linear with respect to this signal. Furthermore, a-Si EPIDs have a large dynamic range compared to film and first generation camera-based imagers, the response also remains linear within this range. This means that the response to a linear superposition of signals will be represented as such in the response (e.g. image) of the EPID. The high dynamic range guarantees, that small signals superimposed on larger signals will be preserved unchanged.

[0036] In the case of images obtained under an IMRT treatment using a cumulative mode i, the signal reaching the EPID is the combination of the following components: (1 ) the intensity modulation (I); (2) patient attenuation (A); (Z) scattered radiation (S); and (4) internal anatomy including implanted markers (D).

[0037] The patient anatomy produces a small signal (contrast on the order of 3%), superimposed on a larger (i.e., slower varying) signal: the intensity modulation (contrast > 5%), patient attenuation and scatter contribution. Figure 2 is a graph with pixel numbers on the horizontal axis and pixel value on the vertical axis showing a cross section of a field obtained with the patient 16 in the beam 14, graph line 30, and without a patient in the beam 14, graph line 32. Specifically, one image was obtained during an IMRT-treatment with the patient 16 in place, and the second image was without the patient 16. The major difference between the two signals is a result of increased scattered radiation reaching the imager 18 and decreased direct signal through patient attenuation of the primary dose. Figures 3a and 3b show the original images used to construct the graph shown in figure 2.

[0038] The total signal T(iJ) with / and j pixel coordinates can be written as the superposition of all contributions, as set forth in Equation (1) below:

T(U) =I(i,j) + P(U) +A0J) + S(iJ) (1) [0039] Because of the linearity, this translates directly to the response of the imager 18 and hence the portal image itself.

[0040] The present invention proposes that the contribution of the patient attenuation and scatter do not change the IMRT signal, but for a change in scale and, can be modeled as a linear transformation of the latter signal as set forth in Equation (2) below:

I' = a *I(i,j) + b (2) where a € R⁺ and b € R. The parameters a and b can be thought of as a gain and offset. [0041] The parameters a and b, defined above, can be determined by minimizing the X² distance between the P(i,j) and O'(iJ), with the X² -goodness of fit, defined by the following expression:

** _ «, |τ⁽i,j⁾ ra^j)!² ,₃> ^x -^{∑ kM} _{σ (}i,j) ⁽³⁾ where σ (i,j) is the variation per pixel for different measurements. This was chosen to be the variance of a flood field image looking at a small area. The value found was 2.48. The minimization can be performed using a variety of algorithms. In one non-limiting embodiment, a Nelder and Mead simplex minimization algorithm was used because it was the simplest to implement. The resulting value is an indication for the goodness of fit and provides a probability for this fit. [0042] Equation (1 ) can be manipulated to:

^A(I^j)P(U⁾ = ^4 ^{+ a} W

Where R(i,j) is the resultant image from the subtraction of the original image and the open image.

[0043] It is important to note that a number of assumptions were made above. First, it was assumed that the images are fairly small and that the variations of patient attenuation in the field are small. This is a fair assumption to make in the case of intensity-modulated treatments. Furthermore, it was assumed that the image geometry was stable and constant between the acquisition of the open field (O) and the acquisition of the portal image (/). The latter assumption is not always the case. day-to-day and from field-to-field due to operator error or sag of the arm supporting the EPID. This is a well-known problem that has been solved in different ways by groups studying registration problems.

[0044] The use of the intensity map as a fit function yields a number signifying how well a proposed or reference intensity map matches the acquired image. Because of the limited impact of the patient on the imaging information, it turns out that the chi²- value is sensitive to any mismatch of the intensity maps, thus providing a tool to verify the correctness of the delivered intensity during treatment.

[0045] To demonstrate the effectiveness of the present invention, a commercially available aS500 flat panel portal imager from Varian Medical Systems of Palo Alto, California was used. It had a 40 x 30cm² sensitive area divided in 512 x 384 pixels. A maximal spatial resolution of 0.78mm was available. The imager was used in an acquisition mode allowing all frames acquired to be averaged. The setting also corrected for a frame-reset error. This acquisition setting was available as clinical software. Every frame was acquired using a setting described by Vetterli et al. "Medical Physics," 31 (4):828-831 (2004), which allows the optimization of the image quality by holding the treatment beam during processing of the imager 18.

[0046] The beam 14 was delivered using a nominal dose rate of 400MU/minute, rather than 600MU/minute, to insure that the imager 18 is not saturated, particularly during the open field procedure.

[0047] An intensity-modulated field from a patient treated for prostate carcinoma was randomly chosen. The patient was part of an IRB-approved study for detection of gold markers during treatment and was imaged every day.

[0048] An open field, i.e., no absorbers in the field, using the intensity- modulated field, was taken with the portal imager as shown in figure 3b. Subsequently, an image was taken with a 10 x 10 x 10 cm³ plexi phantom in the beam as shown ni figure 3a. The phantom had gold seeds embedded in the center. The gold seeds, measuring 2mm in length and 1.2mm diameter, had a contrast of about 3%. [0049] A cross-section was selected in such a way that it did not pass through any of the gold markers featured. The data was cropped to represent data only within the treatment field. A minimization procedure was performed using the objective function described for equation (3).

[0050] A usable implementation needs to correct for the changes in imaging geometry. There are many methods to do this which are all nearly equivalent. In this case, the simplest approach was chosen, assuming that all geometrical differences were due to translations of the imager 18 with respect to the source, i.e., no or negligible rotations. The problem was then reduced to magnification and an in-plane (with respect to the imager) shift. This was solved as follows: [0051] (1) perform segmentation on the reference image O as well as the portal image P using a variance image based procedure yielding two binary images O" and P" having an on-bit in the field and off-bit outside;

[0052] (2) calculate the magnification of P' with respect to O' by taking the square root of the ratio of on pixels; [0053] (3) apply the magnification to P';

[0054] (4) calculate the difference of the centroid of O' and P'; and

[0055] (5) apply the magnification and calculated shifts to P and P'.

[0056] The binary image P' was reduced using mathematical morphological erosion to fit within the treatment field and used as a mask to apply equation (3) to a region of interest. Equation (3) was minimized, obtaining optimal values for a and b, and the subtraction (P - O¹) was performed. The image was displayed after renormalization.

[0057] Fig. 3c illustrates a two-parameter adaptive subtraction, using identical image geometry as figures 3a and 3b. An open image was obtained using the intensity map, the phantom placed on the table was positioned in the beam 14 and a new image was acquired. Figure 4 is a blown-up version of figure 3c. Note that even through the blocks the outline of the square phantom is visible.

[0058] The treatment plan with which the test data was generated was a standard prostate IMRT-plan. The planning was performed using Corvus 5.0, available from Nomos Corp., Sewickly, Pennsylvania, and delivered on a Varian accelerator (2300EX) using dynamic multileaf collimation (DMLC). The treatment plan consists of seven different intensity modulated beams at angles: 150°, 100°, 55°, 0°, 305°, 260° and 210°, see Table I below.

TABLE I

[0059] The intensity modulation was calculated using 1 x 1cm pixels and delivered with a 120-leaf Millenium MLC having 5mm leafwidths. The patient 16 was treated at 400MU/minute nominal dose rate.

[0060] The minimization technique described above can be used to assess the correctness of the delivered field. Indeed, the value yielded by equation (3) is a measure of the goodness of fit and as such is a measure for the correct delivery. Instead of using the full field as in the visualization algorithm, the goodness of fit is determined on a line in the image. This was applied column-by-column, yielding a vector AQ) and line-by-line (BQ)). The image was rotated over 45° and the procedure was repeated yielding C(i') and DQ'). Performing the inverse rotation, new lines C(i,j) and DQJ) were determined. This provides X² values along 4 lines. This process is repeated for all pixels in the image. An "error" map is generated by multiplying the four values, after which thresholding is performed. Pixels that remain highlighted indicate errors in the intensity map.

[0061] Within the ROI, a map of goodness of fit (MQJ)) was obtained over the entire image using MQJ) = AQ)BQ)CQJ)DQJ). [0062] Note that equation (3) needs to be normalized for the number of pixels contributing to the goodness of fit and that the erosion of the ROI should be minimized. [0063] Using a threshold on the map M(Ij) yields a flag for errors during a treatment. This threshold needs to be sensitive enough to detect errors and robust enough to rule out false positives.

[0064] The 0° angle intensity modulated field was used as a test case. In one single segment of the treatment, the position of one leaf was changed by 2mm, 4mm, 6mm, 8mm and 10 mm, respectively. The treated segment comprised 10.1 1 % of the total treatment that was delivered with 77MU. A Las Vegas phantom was used to provide patient scatter.

[0065] After minimization, the X²-goodness of fit was 364.1 for 140 degrees of freedom. This value yielded a p-value of p<0.001. Values for a and b were 0.542 resp. 16,160. After application of the scaling factors, subtraction and renormalization of a clear image of the markers in the phantom was presented as shown in figure 3c. The cross-section after adjustment is shown in figure 5, where the graph line 34 is the image and the graph line 36 is the adjusted open field. It should be noted that figure 5 depicts the same cross-section of the two images shown in figure 2. The open field is scaled linearly to minimize the X²-goodness of fit. The fit is performed on the pixels inside of the field only. The good agreement outside of the field is likely due to chance and the fact that a small phantom will not change the out-of-field scatter significantly. [0066] Patient treatment images obtained before and after ADS are shown in figures 6, 7 and 8. On some of the images, implants and markers can be observed easily, for example, see figure 8. For comparison, open field images used for marker positioning of the same patient are shown in figure 9, i.e., the images for the example patient without IMRT-field. [0067] A non-limiting example of the detection process of the invention is shown generally in figure 1 1. In this view, direct subtraction of a field of the Las Vegas phantom with a 6mm introduced in 10.11 % of the intensity map, where the detection is performed using 2 of the four possible lines, is shown. A threshold value of 1 x 10⁷ on the goodness of fit map detected the presence of errors in the images 4mm leaf error and higher. For 2mm, no error was detected with this threshold, nor was an error detected in the image without error. With a threshold of 1 x 10⁶, false positives started to occur (e.g., errors were detected in the 2mm and reference image (no errors introduced), but not at the position of the error).

[0068] The response of the imager 18 is linear, and the information from patient anatomy is a small signal compared to the intensity modulation, increased scatter and patient attenuation. This allows one to use a direct approach for the patient visualization. The means of both signals is calculated and subtracted from the original, yielding (P-< P>) and (O' - < O' >). The ratio of standard deviations of the resulting signals yields the scaling factor. This approach works well in most cases, but does not give information on the correctness of the intensity map. [0069] The present invention employs an acquired image to subtract from the portal image. As an alternative, using an intensity map from a planning system, the response of the imager 18 can be modeled to obtain an intensity signal that can be subtracted instead of an acquired field.

[0070] A treatment could then be verified as follows: (1) the treatment planning system generates an intensity map per field (/); (2) / is transformed to reflect the sequencing, leakage and response of the portal imager: / → R; (3) R is stored on the portal imager and linked to the treatment by the Record and Verify module; (4) at time of treatment, R is used to generate patient positioning images and verify the intensity map; (5) this can be done at low does rates enabling IMRT- preporting (e.g., about 5MU) or at higher rates generating verification images.

[0071] Alternatively, the "open field" images can be obtained at the time of a dosimetric check. The present invention provides the first description of direct verification of an intensity modulated radiation therapy treatment, verifying patient position and intensity maps. [0072] The source code for a software program operable to practice the present invention is shown in figure 12a-12g.

[0073] The description of the invention is merely exemplary in nature and, thus, variations that do not depart from the gist of the invention are intended to be within the scope of the invention. Such variations are not to be regarded as a departure from the spirit and scope of the invention.

Claims

CLAIMS What is claimed is:

1. A system for verification of intensity modulated radiation therapy, said system comprising: a radiation source for providing a beam of radiation, said source directing the beam of radiation towards the patient; an imager responsive to the beam of radiation from the patient, said imager providing an actual patient image of the patient; and a controller for analyzing the image generated by the imager, said controller storing an open field image without the patient, said controller subtracting the open field image from the actual image to generate a subtracted image from which the verification is performed.

2. The system according to claim 1 wherein the imager is a portal imager.

3. The system according to claim 2 wherein the portal imager is a flat panel portal imager.

4. The system according to claim 2 wherein the portal imager is an electronic portal-imaging device.

5. The system according to claim 4 wherein the electronic portal-imaging device is amorphous silicon electronic portal-imaging device.

6. The system according to claim 1 wherein the actual image includes patient attenuation, scattered radiation and internal anatomy including implanted markers.

7. The system according to claim 1 wherein the beam of radiation has a nominal dose rate of 400 MU/minute.

8. The system according to claim 1 wherein the controller linearly scales the open field image to minimize a goodness of fit process.

9. The system according to claim 1 wherein the controller generates an error map between the actual image and the open field image.

10. The system according to claim 1 wherein the controller obtains the open field image using an intensity map.