Evaluation of Various Free Software Options for Catphan 504 Phantom Analysis

.


INTRODUCTION
The use of computed tomography (CT) imaging has been on the rise for several years, making quality assurance of CT exams a crucial requirement.This is in line with international recommendations and regulations in Brazil, which mandate that these exams must be performed and evaluated by trained and legally qualified professionals.[1][2].
When high-resolution imaging is necessary for accurate diagnosis, adjusting parameters in data processing can be an effective solution.Data processing is done using convolution filters (FCs), which can reduce artefacts and noise without increasing the radiation dose for the patient [3].
The assessment of image quality in CT scans involves the use of appropriate phantoms, such as the Catphan 504 (The Phantom Laboratory, Salem, NY, USA) or those provided by the CT manufacturer.These phantoms can include structures for measuring various image quality parameters such as uniformity, noise, slice thickness, low-contrast and high-contrast resolution, CT number linearity for different materials, and others.When these parameters do not meet the limits recommended by regulations or literature, it may indicate poor image quality [1,2,4].
To evaluate the images, there are several free software available for processing the Catphan 504 phantom [5,6,7,8], such as ImageJ, Pylinac, Spice-CT, and CTQA-cp.These software can help standardize the analysis, minimize errors, and increase efficiency for the user.However, it remains to be determined if the results obtained by these software are consistent with each other and if they match the manual analysis of the phantom.
The objective of this study is to investigate the use of free software (ImageJ, Pylinac, Spice-CT, CTQA-cp) for analysing the Catphan 504 phantom, and to assess their agreement with each other in different convolution filters for thorax and bone imaging.

A) B)
C) The images were processed by the software mentioned above.The following modules were evaluated in the phantom: CTP404, for CT number linearity, CTP 528 for MTF (Modulation Transform Fourier), and CTP 486 for uniformity and noise measurement, as shown in Figure 2.
For the comparisons between the software with different FCs, the t-student statistics were performed for each test and software proposed.

Catphan® 504 phantom
There are several models and variations of the Catphan phantom for CT.The Catphan

Spice-CT software
The Spice-CT Package (V.1.8.7) is a plugin developed to be used in conjunction with the ImageJ software.This plugin enables the analysis of several quality parameters in Catphan CT images, including CT number linearity and slice thickness, uniformity and noise, helical sensitivity in the z-direction, noise between slices, noise power spectrum (NPS), modulation transfer function (MTF), low contrast, and geometry and alignment [9].
The software does not automatically choose the image slices to be calculated, it is up to the executor to select the best image for calculation.

ImageJ
ImageJ is a free software, and it has tools for medical image analysis, but this software does not have automatic functions for analysing the Catphan phantom.In this sense, it will be described how to obtain the analysis of the Catphan 504 phantom [12].

Uniformity
Uniformity testing can identify artefacts such as beam hardening artifacts.It may vary depending on the x-ray spectrum for imaging and phantom centring in the gantry [13][14].
Uniformity is evaluated in the uniformity module (CTP486).ImageJ was placed a ROI in the centre and four ROIs in the periphery, located at 3, 6, 9 and 12 o'clock, with a diameter of 10% of the phantom's diameter [13][14].Uniformity was calculated for all software, with the average values of the number of CTs of their respective ROIs, by the largest difference between the average value of the central ROI and a peripheral ROI [13].
According to the IEC 61223-3-5, uniformity values for water should not be greater than ± 4 HU [14][15].For materials other than water, IPEM suggests a level of ± 10 HU from the baseline number of other materials [15].

Noise
In the uniformity module (CTP486) the noise could be measured by the random variation of CT number values, and it can be estimated through the equation below.In ImageJ software, for the measurements of noise, an ROI with a diameter of approximately 40% of the diameter of the image of the phantom was obtained [1].And the noise was obtained by equation 1: Where   is the standard deviation of ROI, CT  is the average number of ROI CTs, and CT air is the mean air CT number, by definition is -1000 on the Hounsfield scale [1].

CT number linearity
Linearity is measured in the sensitometry module (CTP404); this test aims to evaluate the linearity of CT numbers for different materials.
In the ImageJ software, ROIs were aligned within the seven materials structures to be investigated on the image.The mean CT number for each material was obtained from this information [1,4].

Slice thickness
The level value (L) was obtained to calculate the slice thickness for the analysis with the ImageJ software.To do this, the window (W) was set to its minimum value (W:0).Then, the value of L was adjusted until the ramp disappeared.The registered value of L is the maximum CT number of the ramp (  ).An ROI adjacent to the ramp was selected, and the background's mean CT number (  ) was obtained.The CT number without background (  ) was calculated by subtracting   from   .Next, the   was divided by 2 to obtain  50% .The CT number of half the maximum of the window is  ℎ = p. 9 50% +   .Finally, the value of L was replaced with the value of  ℎ and W was left at its minimum value (W:0) [4].
In ImageJ, a line was drawn over the ramp and the profile curve was plotted, obtaining the FWHM (full width at half maximum).Finally, the slice thickness (Z in mm) was calculated using equation 2 [4]:

Python Script
A Python script was built to calculate the MTF to evaluate the high contrast resolution for comparison with the software codes.

High contrast resolution
To quantify the high contrast resolution in the CT image, the use of Fourier Transform Modulation (MTF) can be adopted.In module 528 there are two impulse sources (beads) aligned with the y-axis.The beads are positioned 20 mm from the centre, and 2.5 and 10 mm from the centre of the module on the z-axis [4,16].
The bead at the top was chosen to calculate the resolution by the MTF.The MTF measurement was performed using Python code (V.3.6) and the pydicom and numpy libraries.Initially, the central image of the beads was chosen, and a square ROI around the bead was performed.Then the background noise was removed to homogenize the ROI [17].
Performed the two-dimensional Fourier transform, the values of the Fourier transform norm were summed in the y-axis direction and then normalized.The plotted curve resulted in the determination of 10% and 50% MTF.

RESULTS AND DISCUSSIONS
The results present the data obtained by the software for each proposed test with different FCs, and thus are compared by the t-student statistics.
Figure 4 shows the values of the CT numbers for the materials air, Acrylic, Delrin, PMP, Teflon, LDPE and Polystyrene, depending on each filter chosen, for each program used. Figure 5 shows the Uniformity and Noise values.The MTF values are presented in Figure 6. Figure 7 shows the slice thickness values.
As can be seen in Figure 4, there is no coincidence in the values of the CT number between the software, but their values were in agreement with each other with  > 0.05.
The smallest p-value among all FCs for each software and its respective test are presented in Table 1.A y-axis was added to the right of the figures to visualise the differences between the values.On this axis, the value of zero represents the minimum value of the data, and the maximum value of the axis is the greatest difference that exists between the minimum and maximum values of each figure.Uniformity data are shown in Figure 5 A

p. 12
The Pylinac software does not generate the Noise result.Therefore, the Noise values were analysed only for the ImageJ, Spice-CT and CTQA-cp software, and the results coincided with each other with p > 0.05, as shown by the Figure 5 B).
The 50% MTF values presented in Figure 6 A).The results obtained values of  > 0.05 for the Python script, Pylinac, Spice-CT and CTQA-cp.And for the 10% MTF, Pylinac obtained  < 0.05 to filter FC52, FC83, FC84, FC85, and FC86.This significant difference may be due to the 10% MTF value returned to be an extrapolation or calculation the MTF by the Line pair per centimetre high resolution gauge method, and the Python script, Spice-CT and CTQA-cp analyse by the bead [4].
For the slice thickness values, shown in Figure 7,  > 0.05 was obtained for ImageJ, Spice-CT and CTQA-cp programs, and a significant difference of  < 0.05 for Pylinac.In Figure 8, it is possible to observe the comparisons of some characteristics of the software used.The program that had the best performance in processing the tests was Spicect, and the one with the lowest performance was Pylinac.
504 phantom was developed by Varian Medical Systems, shown in Figures1 B) and C)[4].The Catphan 504 phantom is subdivided into modules: the CTP404 module, for CT number linearity, scan slice geometry (slice width and slice sensitivity profile), pixel size, circular symmetry, phantom position verification, patient alignment system check, scan incrementation; the CTP 528 module for measurement of high resolution with 21 line pair per cm gauge and point source, bead point source for point spread function and MTF; and the CTP 515 low contrast module with supra-slice and subslice contrast targets; and the module CTP 486 measurements of spatial uniformity, mean CT number and noise value[4].
).It is noted that Pylinac obtained uniformity values outside the results obtained by the other software, with a value of  < 0.05.And the ImageJ, Spice-CT and CTQA-cp software obtained  > 0.05.p. 11

Figure 4 :Fernandes
Figure 4: Number of CT for A) air; B) Acrylic; C) Delrin; D) PMP; E) Teflon; F) LDPE; G) Polystyrene.A) B) Using images with different FCs made it possible to observe different results and test the limitation for data capture by the software.Figures1 to 7show some results of FCs outside the standard deviation range.

Figure 5 :Fernandes 13 Figure 6 :Figure 7 :
Figure 5: Graph of the values of A) Uniformity and B) Noise for FCs.A) B)

Table 1 :
The smallest p-value among all FCs for each software and test evaluated with 95% confidence.