A technique for automatically extracting useful field of view and central field of view images.

INTRODUCTION: It is essential to ensure the uniform response of the single photon emission computed tomography gamma camera system before using it for the clinical studies by exposing it to uniform flood source. Vendor specific acquisition and processing protocol provide for studying flood source images along with the quantitative uniformity parameters such as integral and differential uniformity. However, a significant difficulty is that the time required to acquire a flood source image varies from 10 to 35 min depending both on the activity of Cobalt-57 flood source and the pre specified counts in the vendors protocol (usually 4000K-10,000K counts). In case the acquired total counts are less than the total prespecified counts, and then the vendor's uniformity processing protocol does not precede with the computation of the quantitative uniformity parameters. In this study, we have developed and verified a technique for reading the flood source image, remove unwanted information, and automatically extract and save the useful field of view and central field of view images for the calculation of the uniformity parameters.
MATERIALS AND METHODS: This was implemented using MATLAB R2013b running on Ubuntu Operating system and was verified by subjecting it to the simulated and real flood sources images.
RESULTS: The accuracy of the technique was found to be encouraging, especially in view of practical difficulties with vendor-specific protocols.
CONCLUSION: It may be used as a preprocessing step while calculating uniformity parameters of the gamma camera in lesser time with fewer constraints.

INTRODUCTION

The purpose of quality control is to detect changes in the performance of a gamma camera system that may adversely affect the interpretation of clinical studies. There are a large number of factors that contribute to the final image quality, and it may not be feasible to evaluate these every day before subjecting the gamma camera for clinical studies. Therefore, only the parameters that are most significant in introducing changes in the system performance and have large potential or likelihood to impact clinical studies are monitored daily.[123]

System uniformity of the gamma camera refers to its ability to produce a true representation of the actual distribution of radioactivity within a region. It is the most important parameter that can impact system performance. Changes in photopeak location, photomultiplier tube performance, energy and linearity correction, and so on, all affect image uniformity. Therefore, uniformity test is usually performed on daily basis.

It is difficult to assess visually the variations in the counts corresponding to different areas of the field of view of gamma camera that reflect a lack of uniformity of the camera system. Therefore, comparisons between different cameras or variations in uniformity of a single camera over a period are visually difficult to estimate. Therefore, two different uniformity parameters that are usually measured during the test are: Integral uniformity and differential uniformity. These are calculated for both the central field of view (CFOV) and useful field of view (UFOV) of the gamma camera respectively.[4]

Vendor specific acquisition and processing protocol to perform uniformity test provide both the flood images and integral and differential uniformity in UFOV and CFOV. However, these protocols are specific to the factory designated number of counts that must be acquired and do not show required parameters if the acquired counts in the image does not fulfill criteria. For example, 10,000k counts per head, and 4000k counts per head are required for Symbia T6 (Siemens) and Discovery NM/CT 560 (GE Health Care), respectively.

In this study, we have present a procedure that reads the flood source image independent of vendor's single photon emission computed tomography gamma camera system on which the image was acquired and also independent of the total number of counts acquired in the flood source images, remove unwanted information that is pixels having zero counts surrounding the flood source image, and automatically extract and save the UFOV and CFOV images for further processing. The results have been verified in more than one way and are certainly encouraging.

MATERIALS AND METHODS

Program design and development

Since medical physicists and nuclear medicine technologists will be the end users, therefore an interactions with these and the authors provided to get the perspective, and it was decided that the program should have the following features:

The program should run at the command prompt

It should provide option to the user to select the image file that needs to be processed

The program should accept and process digital imaging and communications in medicine (DICOM) images only, and if user selects any other file format the program should show error message

The program on a successful selection of the image file should automatically extract UFOV and CFOV images and save them in the current folder/path.

The program was implemented using MATLAB R2013b (The Math Works, Inc. 3 Apple Hill Drive Natick, MA 01760-2098). MATLAB has image processing toolbox with built-in DICOM support. Since it was software based study and did not involve the patient, therefore ethical clearance was not required.

A MATLAB script that integrates six in-built MATLAB functions (imgetfile, dicominfo, dicomread, dicomwrite, find, and imcrop) in a logical fashion to complete this task was written. The script is divided into four sections namely: User Interface to select the image file from a folder, read the DICOM images, find the coordinates of nonzero pixels, and to automatically remove zero pixels surrounding the image and automatically extract and save UFOV and CFOV images.

Section 1: User interface (select the image file from a folder)

It provides the facility to the user to select the desired *.dcm file to be processed. It opens the files and on successful completion, it returns the filename, otherwise provides error-message.

Section 2: Read the digital imaging and communications in medicine images

This section uses “dicominfo” to find the number of images in the selected file, and height and width of the image and store these information in variables for further to be used by next section. Then “dicomread” to read user selected *.dcm file, and store the images data in a matrix.

Section 3: Find the coordinates of nonzero pixels

The function find was used that takes image matrix as an input and return the coordinates of nonzero pixels stored in the two arrays one for x-axis (row) and another for y-axis (column). Figure 1 shows how the image is stored in MATLAB. Then, minimum and maximum values of both the x-axis and y-axis are determined. The minimum value will correspond to the starting point of the first nonzero pixel and maximum will correspond to the last nonzero pixel. These are stored in the variables to be used by the next section.

Figure 1

How a digital image is stored in MATLAB. The origin is on the left top. The x-coordinate increases in the bottom direction, and the y-coordinate increases in the right direction

Section 4: Extract and save useful field of view and central field of view images

The function “imcrop” with input parameter as starting x-coordinate of nonzero pixel, starting y coordinates of nonzero pixels, length in x-direction, and length in y-direction was used to extract the flood source image leaving unwanted information that was surrounding the flood source image. The length in x-direction was computed by subtracting the starting x-coordinate from the ending x-coordinate of nonzero pixel, and similarly, the length in y-direction was computed by subtracting the starting y-coordinate from the ending y-coordinate of the nonzero pixel.

UFOV is 95% of half height radius of the digitized image that is the flood source image in this case, and CFOV is the 75% of UFOV.[5] Therefore, in order to extract UFOV image, 5% pixels from the boundary of extracted flood source image were excluded and UFOV image was extracted using “imcrop” function following the same procedure as was used to extract flood source image that is mentioned in the previous paragraph. CFOV image was extracted from the UFOV images by excluding 12.5% of the pixels from the boundary of the UFOV images. Using the function “dicomwrite,” the UFOV and CFOV image was saved permanently on the hard disk in the current folder for further processing.

Verification

The program was verified by subjecting it to the simulated and real flood sources images (for 106 input images). The input images were from the following groups: The acquired images when gamma camera was not fully covered with flood source, extremely low counts image (5K counts in 256 × 256 matrix), high counts images (10,000K in 1024 × 1024 matrix) of the Cobalt-57 (Co-57) flood source image. Flow Chart 1 summarizes the characteristics of the test images. There were 424 images (4 × 106 = 424 images, input, extracted, UFOV, and CFOV images) and their workspace variables were reviewed during the verification.

Flow Chart 1

The characteristics of the test images

The accuracy of the result has been demonstrated by looking into the each pixel count of the original and the extracted image data. A 64 × 64 image matrix data having total 5K counts were selected to display the result because higher than this such as 128 × 128, 256 × 256 matrix size images are relatively cumbersome in printed form.

The above script runs by typing “autoextract” on the command prompt. Run presents an option to select input image, displaying an error message if any error occurs during the selection of the input image. Successful selection displays input image, extracted foreground image, extracted UFOV image and extracted CFOV images along with the variables and their values in workspace window [Figure 2].

Figure 2

(a) The result of typing autoextract on the command prompt and then after pressing Enter Key, file browser opens that allows user to select the image to be processed. (b) Displays the error message when no file was selected by the user. (c) Displays error message when a file having format other than digital imaging and communications in medicine was selected. (d) Displays the workspace variable when a file having format other than digital imaging and communications in medicine was selected

MATLAB function tic starts the clock and toc provide the time elapsed to execute the commands placed between tic and toc. These two functions were used to find the time taken by this technique to extract UFOV and CFOV image. The program “autoextract” was executed twenty-six times and each time different input flood image having counts ranging from 5K to 10,000K was selected.

RESULTS

Figure 3a and b shows one representative image displaying successful extraction of UFOV and CFOV image. The average, standard deviation, and range of the time taken by this technique to extract UFOV and CFOV were found to be 0.5781 s, 0.0176, and 0.556 to 0.641 s, respectively.

Figure 3

(a) Input image of matrix size 1024 × 1024 displayed at 50% while the extracted foreground image (643 × 887) displayed at 100%. (b) The extracted useful field of view (579 × 799) and extracted central field of view images (435 × 599) from the input image (1024 × 1024) as shown in Figure 3a

Review of 106 × 4 = 424 images, where for each input image, one input image, one extracted foreground image, one extracted UFOV image and one extracted CFOV image that is total four images for each input images, were verified.

The accuracy of the results has been demonstrated by looking into the each pixel counts of the original and the extracted image data. A representative image data acquired in 64 × 64 matrix having total 5K counts is shown in Figure 4a. The extracted images data after removing the unwanted information that is zero-pixel counts surrounding the object (flood source) have been shown in Figure 4b. Careful observation of the Figure 4a and b, displays that there is an accurate extraction the object (i.e., flood source).

Figure 4

(a) Pixel counts of Cobalt-57 flood source image (total number of counts = 5K) in a 64 × 64 matrix. Flood source data starts from row number 13-53 and column number 05-60. The technique should remove pixel having zero count surrounding the flood source image, and extract image area defined by row number 13-53 and column number 05-60 (matrix: 41 × 56. The extracted flood source image data have been shown in b. Looking into the a and b, it is obvious that the program successfully extracted the flood source image. (b) The extracted flood source image from the image data shown in a. Note that extracted image is of the size 41 × 56 matrix

When the detector surface area was not fully covered by the flood source, the extracted flood source image looks like as if it had included the zero pixel counts surrounding the flood source image as shown in [Figure 5]. However, this is not the cases. Reviewed the input image data (as similar to as shown in Figure 4a and b), showed accurately extracted object as per the method defined in above.

Figure 5

Example of acquired flood source images when gamma camera crystal surface was not fully covered as test case

DISCUSSION

System uniformity is an important parameter that can introduce change in the system performance and is the most likely to impact clinical studies and therefore it is monitored daily. The successful completion of the uniformity test ensures that any abnormality noted in patient scan is only due to processes within the scanned organ. However, if the results are not within the acceptable limit, then remedial action must be taken.

Uniformity acquisition and processing protocol provide both flood source image as well as the uniformity parameters (UFOV, CFOV, integral, and differential uniformity). However, the vendor-specific protocols do not entertain the images that have total counts less than that specified by the protocol. In this study, we developed a simple technique to select desired flood source image independent of total counts acquired and also independent of the gamma camera on which data was acquired, remove unwanted information, and finally save the UFOV and CFOV images for further processing.

The technique successfully extracted foreground Co-57 flood source images with no information loss. When the detector surface area was not fully covered by the flood source, the extracted image had a larger area in comparison to the area of the detector actually exposed by the Co-57 flood source. This was because of the contribution of the photons in the image that were scattered inside the collimator [Figure 6]. In such case, the method does not remove unwanted information completely. Practically, this type of situation will not arise because it is prerequisite that detector area must be fully covered with the Co-57 flood source before starting the acquisition. However, uniformity parameters calculated for UFOV and CFOV images in both cases and results are useful.

Figure 6

(a) Front view. (b) Side view. The scattered gamma photons from the collimator might interact with NaI (Tl) detector area that is not covered with Cobalt-57 flood source, and contribute to pixel counts in the image. However, their percentage will be is significantly less (in comparison to the photons contributing pixel counts in the area covered with the flood source) and therefore will appear as black, when image is displayed in full gray scale range

Information loss as a disadvantage of the auto-cropping algorithm has been reported in literature while cropping the signature of a person.[6] We did not find data loss while extracting UFOV and CFOV images provided the detector was fully covered with the Co-57 flood source. The time-taken for image analysis improves significantly when it is applied to the region of interest, in comparison to the whole image. Therefore this technique can be used as preprocessing step of image analysis task.

Limitation of the method

Extracted flood source image included a few pixels of unwanted information that were not part of actual flood source, in case of real data. However in simulated data such problem is not present. If one is interested in extracting actual flood source image, the method needs modification, that is, instead of finding the coordinate of nonzero pixels, one can first determine the average counts in the area of the image of gamma camera that was not covered with the flood source, and find the coordinate of those pixels greater than that average counts. However, this needs another study.

Since MATLAB runs on several platforms such as Windows, Linux, and Mac OS, this technique can be used on several platforms too. Our software is not yet available on the public network. There exists a software application called NMQC in the literature that has been developed as an alternative or complement to the vendor specific processing software application.[7] However, this study is different since it is not specifically related to extraction of foreground Co-57 flood source image.

CONCLUSION

This technique accurately extracts UFOV and CFOV images and has been verified. It is suggested that this technique may be used as preprocessing step while calculating uniformity parameters.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest