Quality Assurance of Intra-oral X-ray Images

(1)

Umeå University

Quality Assurance of Intra-oral X-ray Images

Bildbaserade kvalitetskontroller av intraoral röntgen

Dieudonne Diba Daba

Supervisor Josef Lundman

Examiner Jonna Wilén

(2)

Abstract

Dental radiography is one of the most frequent types of diagnostic radiological investigations performed. The equipment and techniques used are constantly evolving.

However, dental healthcare has long been an area neglected by radiation safety leg- islation and the medical physicist community, and thus, the quality assurance (QA) regime needs an update. This project aimed to implement and evaluate objective tests of key image quality parameters for intra-oral (IO) X-ray images.

The image quality parameters assessed were sensitivity, noise, uniformity, low- contrast resolution, and spatial resolution. These parameters were evaluated for repeatability at typical tube current, voltage, and exposure time settings by computing the coefficient of variation (CV) of the mean value of each parameter from multiple images. A further aim was to develop a semi-quantitative test for the correct alignment of the position indicating device (PID) with the primary collimator.

The overall purpose of this thesis was to look at ways to improve the QA of IO X-rays systems by digitizing and automating part of the process. A single image receptor and an X-ray tube were used in this study. Incident doses at the receptor were measured using a radiation meter. The relationship between incident dose at the receptor and the output signal was used to determine the signal transfer curve for the receptor. The principal sources of noise in the practical exposure range of the system were investigated using a separation of noise sources based upon variance.

The transfer curve of the receptor was found to be linear. Noise separation showed that quantum noise was the dominant noise. Repeatability of the image quality parameters assessed was found to be acceptable. The CV for sensitivity was less than 3%, while that for noise was less than 1%. For the uniformity measured at the center, the CV was less than 10%, while the CV was less than 5% for the uniformity measured at the edge. The low-contrast resolution varied the most at all exposure settings investigated with CV between 6 - 13%. Finally, the CV for the spatial resolution parameters was less than 5%. The method described to test for the correct alignment of the PID with the primary collimator was found to be practical and easy to interpret manually. The tests described here were implemented for a specific sensor and X-ray tube combination, but the methods could easily be adapted for different systems by simply adjusting certain parameters.

(3)

List of Abbreviations

C-D Contrast Detail

CV Coefficient of Variation

ESF Edge Spread Function

IO Intra-Oral

LSF Line Spread Function

MTF Modulation Transfer Function

QA Quality Assurance

QC Quality Control

SNR Signal-to-Noise Ratio

SSM Swedish Radiation Safety Authority

(4)

1. Introduction

1.1 Background

X-ray imaging in dental radiology is a useful tool for the diagnosis of many oral dis- eases and conditions. Intra-oral (IO) X-ray imaging is the most common modality in this area, for which bite-wing, periapical, and occlusal are the typical projections.

In an IO X-ray examination, the image receptor is placed inside the mouth of the patient and irradiation is external. Conventional radiographic techniques using X-ray films have been used over the years to acquire IO X-ray images, but advancement in digital technology has led to the widespread adoption of digital image receptors.

The various image processing techniques offered by digital radiography results in improved image quality and eliminates the use of potentially harmful photo processing chemicals. Additionally, the decreased image acquisition time with digital radiography reduces the patient’s exposure to ionizing radiation [1]. In comparison to conventional X-ray or computed tomography, the level of radiation exposure to the patient during dental X-ray is small. However, as the number of annual dental examinations with X-ray imaging is increasing, the cumulative ionizing radiation dose may be significant [2]. As the probability of developing cancer increases with increasing radiation dose [3], measures are needed to reduce radiation exposure to patients whilst maintaining high image quality during dental X-ray imaging.

Approximately four IO X-ray images are acquired during each dental diagnostic X- ray exam. This could be reduced by minimizing unnecessary retakes often caused by poor image quality. A quality assurance (QA) program that takes into account the routine quality control of the individual system components is needed to monitor the performance and image quality of dental X-ray systems.

In Sweden, the Swedish Radiation Safety Authority (SSM) is the agency responsible for the formulation of regulations and supervision of activities involving radiation.

The new Radiation Protection Act (2018:396) [4] and associated regulations (SSMFS 2018:2) [5] have outlined rules concerning the use of IO X-ray imaging in dentistry.

Important requirements on beam shaping devices and image receptors used in dental X-ray imaging systems are set in one of the regulations. Beam shaping devices used must be capable of limiting the spatial extent of the X-ray field to match the size and shape of the image receptor. Image receptors must have high sensitivity and the X-ray system must be capable of short exposure times in order to benefit from the receptor’s high sensitivity [5]. Thus existing QA programs that do not conform to these regulations need to be updated appropriately.

(7)

1.1. Background

Moreover, due to the widespread availability of IO X-ray systems, the requirement for qualified personnel to perform QA testing and evaluation is practically time- consuming and not efficient. Thus, there is a need for an automated and time- efficient QA evaluation of IO X-ray systems. This thesis deals with the implemen- tation of quantitative methods that can be used for an automated workflow for QA testing of two indicators of the performance of IO X-ray systems: imaging quality performance of the receptor, and the correspondence between the size of the X-ray field at the open end of the PID and the effective image reception area of the digital X-ray image receptor.

1.1.1 Image quality evaluation

The performance of image receptors can be assessed using image quality parameters such as noise, low-contrast resolution, uniformity, spatial resolutions, sensitivity and dynamic range [6]. Subjective methods such as manual inspection and objective approaches using mathematical algorithms have been used to evaluate the image quality of digital receptors for IO X-rays [7, 8, 9]. Objective methods remove observer influences and offer the possibility to digitize and automate the QA process. It eliminates the differences in results that are inherent from subjective assessments.

In addition, an objective approach offers the possibility of digitizing and automating the image QA process.

Regular quality control (QC) of the IO systems are needed to ensure they perform as expected. Hellén- Halme et al. employed objective methods to evaluate the performance of a large number of sensors commercially available in Sweden against sensitivity, noise, low-contrast resolution, spatial resolution and uniformity [9]. The authors found that sensors of different brands, and also individual sensors of the same model for some brands, showed a large variation in performance. Noise and low-contrast resolution were found to vary the most for sensors of different brands.

Furthermore, the detector transfer characteristics which expresses how the mean pixel intensity of the receptor over a specified region varies with exposure was found to be considerably different for sensors of different brands. This creates a serious challenge as the settings for image quality testing of each and every receptor in use must be separately determined.

1.1.2 Collimation and beam alignment

Regular checks on the alignment between the primary collimator and the position indicating device (PID) are needed to ensure that the X-ray beam is centered per- pendicularly to plane of the image receptor. Misalignment may result in the exit X-ray field size smaller than the area of the beam exit end of the PID which can make it difficult for an operator of the system to properly aim the field at the ROI.

As a consequence, the entire area of the receptor may not be covered by the X-ray beam resulting in images with partially unexposed regions. In film-based systems, test for collimation and beam alignment can be done by taking radiographs with the PID aimed at a well-defined area of film or arrangements of films. Evaluation

(8)

1.2. Aim

is conducted by measuring the size of the unexposed area inside the area that is supposed to be covered by the PID.

1.2 Aim

The aim of this project is to develop ways to improve the QA of IO X-rays systems by digitizing and automating part of the process. The specific aims are:

• To implement quantitative image quality tests for the receptor’s sensitivity, uniformity, noise, spatial resolution, and low-contrast resolution; and a semi- quantitative test for the correct alignment of the PID with the primary collimator.

• To evaluate the image quality tests for the feasibility of an automated workflow for QC testing of IO X-ray images.

1.3 Limitations

This project is limited to the digital IO X-ray system assessed but could be adapted to other IO systems.

(9)

2. Theory

2.1 Digital Intraoral X-ray systems

2.1.1 System description

An X-ray image is produced by X-rays passing through an object and interacting with the image receptor. Image receptors for dental X-ray systems are either based on film or digital technologies, with digital imaging being the standard nowadays.

Common intraoral dental systems are either fixed or mobile, consisting of a tubehead, a positioning arm, and a control panel. The X-ray tube or tubehead is similar to a conventional X-ray tube. The control panel provides means for the operator to control the tube voltage (kV), tube current (mA) and exposure time (s), while the extension arm allows for movement and positioning of the tubehead. Figure 2.1 shows the components of a typical fixed intraoral X-ray system.

Figure 2.1: Components of a typical fixed intraoral system. (a) The X-ray tube and positioning arms, (b) control panel, (c) exposure switch and (d) image receptor.

(10)

2.1. Digital Intraoral X-ray systems

2.1.2 Image receptor

Image acquisition is through either direct or indirect image capture. Direct acquisition refers to the direct capture of a latent image by the receptor that can be viewed instantly on a computer monitor. Receptors for direct digital IO imaging are based on solid state technologies, for which the two main types are charged-coupled device (CCD) and complementary metal-oxide semiconductor (CMOS). Figure 2.1(d) shows an example of a direct digital image receptor based on CMOS technology.

The most common form of indirect image acquisition is similar to film and uses a light-sensitive phosphor plate. A photo-stimulable phosphor layer coated on the plate gives it the capability of storing X-ray energy during exposure. After exposure, the plate goes through a series of processing where the captured X-ray energy is converted to an electrical signal and displayed as a digital image. The main disadvantage of this compared to direct digital acquisition is the long processing time which can take between 30 seconds to 5 minutes [10].

2.1.3 Collimation

X-rays are generated in the tubehead by bombarding a metal target with fast-moving electrons. Collimation is a means of restricting the spatial extent of the X-ray field.

The primary collimator is a metal plate with an aperture in the middle that is fixed inside the tubehead. In addition to the primary collimator, the PID acts as a secondary collimator and also as a means to maintain an acceptable focal spot to skin distance [10]. Figure 2.2 is an illustration of an intraoral tubehead showing the PID attached. There are three basic types of PIDs in use: conical, rectangular, and round. In an IO X-ray examination, the PID is aimed closed to the patient’s face and aligned with the image receptor. Collimating the X-ray beam to correspond to the size and shape of the image receptor is an effective way of reducing patient exposure. Studies have shown that patient exposure is significantly reduced when using a rectangular collimator as compared with a circular collimator. A review of thirteen studies showed that the use of rectangular collimation resulted in a reduction in radiation dose of at least 40% when compared with circular collimation [11].

2.1.4 Cone cut effect

A cone-cut artifact occurs when the X-ray beam is not properly aligned with the image receptor leaving part of the image receptor outside of the beam. This may result in unexposed portions of the image. The outline of the unexposed area in the image is determined by the type of PID used. A curved outline will result from a circular PID, while a rectangular PID will produce a straight outline [12]. Improper alignment of the PID with the primary collimator is one cause of this effect.

(11)

2.2. Image quality

Figure 2.2: Diagram of an IO X-ray tubehead showing various components.

2.2 Image quality

2.2.1 Sensitivity and uniformity

Sensitivity is a measure of the receptor’s signal amplitude or intensity as a function of absorbed dose. It depends on the transfer characteristics of the receptor. Uniformity, on the other hand, measures the variation of sensitivity over the effective area of the receptor. Sensitivity variation can be due to differences in gain among the pixels of the receptor.

2.2.2 Noise

Noise describes random variations in the pixel values of the image that does not arise from the attenuation in the object. It is an important factor that affects the quality of medical X-ray images [13] and determines the ability of the imaging system to detect small and low-contrast structures [14]. In digital X-ray images, noise originates from the statistical nature of X-ray emission and detection, scattered radiation and imperfections in the receptor and associated electronics [15]. Statistical or quantum noise is the principal source of noise in digital X-ray images. It varies with the number of photons arriving at the image receptor and has been shown to follow a Poisson distribution [16]. Increasing the intensity of the X-rays or numbers of quanta will lead to a reduction in noise, and vice versa. For situations in which the variations in each X-ray quanta is independent of the other and does not change with time, noise can be estimated by measurement of the variance in pixel values σ²_p from a specified ROI in the image, i.e

(12)

2.2. Image quality

σ²_p = 1 N − 1

N

X

i=1

I_i− ¯I2

(2.1)

where N is the number of pixels in the ROI, Ii is the value of pixel i, and ¯I is the mean pixel value of the ROI. The coefficient of variation (CV) of the pixel values in the ROI is the ratio of the standard deviation to the mean pixel value i.e

CV = σ_p

¯I . (2.2)

Relative noise which is the magnitude of image fluctuations relative to the signal present can be calculated as the CV in pixel value [13].

2.2.3 Contrast resolution

Contrast is a measure of the relative signal difference between an object of interest and the surrounding background. It is affected by X-ray intensity, scatter radiation, and detector properties [17], and can be defined by

C = ∆I

I_b (2.3)

where ∆I is the relative difference in intensity between the target or ROI and the background, and Ib is the average background intensity in the immediate surrounding of the target [18].

An important parameter that expresses quantitatively how the signal compares to noise is the signal-to-noise ratio (SNR). If the signal is expressed as a contrast in the image, then SNR is defined as

SNR = ∆I

σ_b = CI_b

σ_b, (2.4)

where, σb is the standard deviation of the background intensity due to noise.

The ability of an observer to detect a target in an image depends on contrast. The lowest contrast value for which a target can be detected with certainty is called the threshold contrast. Low-contrast resolution measures how well a very low signal originating from an object in a high noisy background can be detected. Noise is a limiting factor in the detection of low-contrast structures in radiographic images and it has been shown in CT images that a reduction in image noise improves the observation of low-contrast structures [19]. A contrast-detail analysis is a common technique used to evaluate low-contrast resolution of radiographic images and provides a quantitative evaluation [20].

(13)

2.2. Image quality

The Rose Model

The Rose model is a theory put forward by Albert Rose in 1948. The theory states that a certain minimum SNR is required for a target in an image to be distinguished from the surrounding background. Consider a target of area, A in an image with uniform background noise having a mean number of photons per unit area ¯Nb. Rose defined noise as the standard deviation σb, in the number of quanta in an equal area of the background as the target. Assuming that the background photons are uncorrelated, and follow Poisson statistics, the noise is given by σb =p

A ¯N_b. For a contrast C resulting from the target, the SNR according to the Rose model is given by [18],

SNR_Rose = Cp

A ¯N_b. (2.5)

The main focus of this theory is to determine the minimum SNR sufficient to detect a target signal. A signal is expected to be reliably detected if the measured SNR is above some threshold value, k. Theoretically, the probability that noise fluctuations in the target will exceed the mean value of the background by k = 1, 2, 3, 4 is 0.15, 0.023, 1.3 × 10⁻³, 3 × 10⁻⁵, [21]. The best value of k has been sought out by several researchers. Rose initially set the value of k to unity but investigations by others showed that for reliable detection, a value between 3 to 6 is desirable [22].

2.2.4 Spatial resolution

Spatial resolution describes the ability of an imaging system to represent close objects as distinct structures. It depends on several aspects of the imaging system, of which the X-ray focal spot size, and the image receptor elements size and separation are the main factors [16]. Digital image receptors are made up of a matrix of discrete sampling points separated by a fixed distance, called pitch. Each sampling point is surrounded by an area called a pixel. Pitch determines the sampling rate and according to the Nyquist–Shannon sampling theorem, the maximum frequency in the image that can be truly reproduced by the system is is 1/2d , where d is the pitch [16]. The system will incorrectly represent spatial frequencies that are greater than 1/2d, resulting in an aliasing effect. Aliasing degrades the system’s spatial resolution and leads to poor image quality [23].

The most widely used metric for the characterization of spatial resolution is the systems modulation transfer function (MTF) [24]. The MTF describes how well the spatial frequencies that make up the object are transferred to the image. A common practical method to measure the MTF of a radiographic system is by the determination of the system’s response to an object with a sharp edge by computing the edge spread function (ESF). The ESF is then differentiated to obtain the line spread function (LSF). The MTF is then obtained by a Fourier transform of the LSF [25]. This method is preferable to other methods such as the slit and the point methods because its accuracy is not sensitive to imperfections of the object’s edge [24].

(14)

3. Methods

The image quality parameters used were sensitivity, noise, uniformity, low-contrast resolution, and spatial resolution. The first four of these quantities were determined from a single flat-field image, while the last one was determined from an edge image.

The flat-field image was produced with no attenuating material in front of the image receptor. The edge image was obtained utilizing an edge test object. The test for the correct alignment of the collimators was performed using an alignment test device.

The system used in this study was a Focus IO X-ray machine from GE Healthcare with two tube voltage settings of 60 and 70 kV, respectively. All exposures were made at a fixed tube current of 7 mA. The image receptor used for the study was a Schick 33 sensor, based on CMOS technology. The sensor has two resolution settings of 0.015 mm and 0.03 mm but the system was set to use the lower resolution with a matrix size of 1200 × 854. The sensor has 16 bits allocated but stores data at 12 bits. Images captured were stored and later analysed. The codes for the analysis was programmed in the Python programming language.

3.1 Signal transfer curve

The signal transfer curve of the image receptor was obtained by plotting the output signal against dose at the receptor. The output signal is the calculated mean pixel value from a central ROI of size 100 × 100 pixels in the image. Flat-field images were obtained at the exposure times of 0.02, 0.025, 0.032, 0.04, 0.05, 0.063, 0.08, 0.1, 0.125, 0.16, and 0.20 s at fixed tube voltage of 60 kV. This was repeated at the higher tube voltage of 70 kV. The exposure time range was limited and did not extend to the highest exposure possible for the receptor because previous studies [9]

have shown that the Schick 33 sensor gets saturated at an exposure of about 0.2 s for similar X-ray tube settings.

Absorbed dose at the receptor was measured using an RTI Pirahna dose meter.

The receptor was replaced by the meter and the dose was measured for each of the exposure time settings above for tube voltage of 60 kV. From the acquired flat- field images, the mean pixel value, and standard deviation at each exposure were calculated from a centrally placed rectangular ROI. The transfer property of the receptor was obtained through a linear fit of the mean pixel values versus measured dose at each exposure time setting [26]. A linear model was assumed in the fit because previous studies have shown that the transfer property of Schick 33 sensor is approximately linear for exposures below the saturation exposure [9]. The method

(15)

3.2. Noise components

used is similar to that described in the work of Samei et al. [24], except for the fact that in their study they used an intercept-free linear model, and the mean pixel values and measured dose at each exposure setting were obtained by taking the average from three images.

3.2 Noise components

The exposure range for which image noise is dominated by quantum noise is of rele- vance because its level in an image is determined by the number of X-ray quanta or exposure. For example, a decrease in noise may indicate a change in exposure. The noise contribution from different sources in an image was investigated by expressing the noise variance as a sum of different components. The variance in raw pixel values recorded by an X-ray image receptor is assumed to be a combination of quantization noise, additive noise, photon noise, and structured noise [27]. Quantization noise is assumed to be very small and negligible. The variance due to additive noise which is electronic in nature is assumed to be independent of exposure and negligible. The photon noise variance σp² which is Poisson in nature is proportional to the exposure.

Finally, the structured noise variance term σb² which includes variation due to for example the heel effect, and scatter radiation, is assumed to vary as the square of exposure. The total variance in the raw pixel values is described by the equation [27]

σ²_R= σ²_p+ σ_b² = k_pX + k_sX² (3.1) Where kp and ksare proportionality coefficients for the photon noise and structured noise respectively, while X is the exposure. Flat-field images were obtained at a fixed tube voltage of 60 kV and for exposure times of 0.02, 0.025, 0.032, 0.04, 0.05, 0.063, 0.08, 0.1, and 0.125 s. The images were linearized using the signal transfer property of the receptor determined in section 3.1. From the exposure linearized images, the variance was calculated at each exposure from an ROI of size 100 x 100 pixels and fitted to equation 3.1 to obtain the parameters kp and ks.

3.3 Image quality tests

3.3.1 Sensitivity

Sensitivity was determined from a flat-field image as the mean pixel value in a centrally placed rectangular ROI covering 16% of the receptor area. This is similar to the method used in Hellén-Halme, Johansson, and Nilsson’s study, except for the fact that they used an elliptical ROI [9].

(16)

3.3. Image quality tests

3.3.2 Uniformity

Receptor uniformity was determined from a flat-field image for both the central portion of the receptor area and for the edges. From the image, five square ROIs each side equal to 25% of the shortest side of the effective area of the sensor were obtained. Central uniformity was obtained as was done in Hellén-Halme, Johansson, and Nilsson’s study [9], where one of the ROIs was placed at the center of the image and the other four placed peripherally around it as shown in Figure 3.1(a). The mean pixel value of each ROI was calculated and uniformity was expressed as the maximum percentage deviation in the mean pixel values for the peripheral ROIs to that of the central ROI. The arrangement of the ROIs for the measurement of uniformity at the edges was done as in Marshall, Mackenzie and Honey’s study [28], by placing the peripheral ROIs at the edges of the image as shown in Figure 3.1(b).

Similarly, edge uniformity was calculated as the maximum percentage deviation in the mean pixel values of the peripheral ROIs to that of the central ROI.

Figure 3.1: Arrangement of ROIs for uniformity determination. (a) central uniformity, (b) edge uniformity.

3.3.3 Noise

Noise was measured as the CV of the pixel value in a specified ROI expressed as a percentage [9]. From a flat-field image, the mean pixel value and the standard deviation of the pixel values were measured from a centrally placed rectangular ROI covering 16% of the image area.

3.3.4 Low-contrast resolution

Low-contrast resolution is determined from a flat-field image based on the Rose model [9]. A central rectangular portion of the image of size about 10% of the image area was extracted, and from which 1000 square ROIs were generated. The ROIs were placed randomly within the specified portion of the image and were allowed

(17)

to overlap. The mean pixel value in each ROI was measured and the standard deviation of the mean pixel values, σn was calculated. The procedure was repeated for 10 different square ROIs of sides in the range 0.2 − 3.0 mm. The threshold SNR, k was set to 3 as was done in [9]. For each ROI size, the threshold limit for low-contrast detectability was calculated as k · σn. A contrast-detail (C-D) curve was created by plotting the threshold low-contrast detectability limit divided by the mean as a percentage against object size in mm. The low-contrast resolution parameter was reported as the area under the C-D curve. The processing steps from the flat-field image to the low-contrast resolution parameter calculation are given in Figure 3.2. Figure 3.3 shows examples of C-D curves obtained for a tube voltage of 60 kV at two different exposures of 0.04 s and 0.08 s.

Figure 3.2: Examples of C-D curves for exposures of 0.04 and 0.08 s

(18)

Figure 3.3: Schematic diagram of the processing steps from the flat-field image to the low-contrast resolution parameter.

(19)

Objective determination of spatial resolution was done by measuring the MTF using a combination of the slanted edge technique described in the work by Samei et al.

[29] and in IEC 62220 standard [26]. The test object is a rectangular steel bar of thickness ∼ 4.93 mm. During the test, the phantom was placed directly on the detector, slanted by a small angle before exposure. The obtained image was first linearized using the signal transfer property of the receptor determined in section 3.1 before the extraction of the edge profile. The image was converted to a binary image by applying a thresholding operation based on the mean pixel value. A Sobel edge detection operator was later performed to extract the edge within the image.

The processing of the image containing the detected edge was done on a row-by- row basis, where it was assumed that the edge was oriented vertically. Figure 3.4 shows the binary edge image (a) and the result of the Sobel operation (b). A Hough transformed was then performed to determine the angle of the edge and the position of the edge line for each row in the image. A linear function was fitted across all the sub-pixel edge location estimates for each row. A square ROI with the edge line at the center was extracted on which a second Hough transformed was carried out to improve on the accuracy of the edge angle determination.

Figure 3.4: Edge image. (a) Binary image, (b) result of the Sobel operation The size of the ROI must be carefully selected to have sufficient number points for the representation of the ESF. In Samei et al. studies [29], the size of the ROI was set to a region of 5 × 5 cm² deemed sufficient for the representation of the ESF.

Another approach mentioned in IEC 62220 suggested the use of N consecutive lines across the edge, where N = 1/ tan(θ) is rounded up to the nearest integer, and θ is the edge angle. This approach has the disadvantage that the size of the projection region depends on the edge angle. In this work, the size of the ROI was set to a square region of 120 × 120 pixels. For each row of the ROI, each pixel was projected onto a location z on a line perpendicular to the edge given by

z_k = [j − φ(i) cos(θ)]p (3.2)

where i and j are the pixel row and column coordinates, φ(i) denotes the location of the edge in row i, θ is the estimated angle of the edge, and p is the pixel size in mm.

The oversampled ESF are binned into uniformly spaced bins of width, ∆s equal to

(20)

bin k was calculated by averaging all the pixel intensities of pixels whose distances from the edge si,j satisfied the condition zk≤ s_i,j < z_k+ ∆s. The obtained sub-pixel sampled ESF was then smoothed through the application of a Savitzky–Golay filter with a window size of 17 pixels. The ESF was numerically differentiated to obtain the LSF using the finite difference approximation [29],

LSFk= ESF_k+1− ESF_k−1

2∆s . (3.3)

A Hanning filter was then applied to reduce noise in the tails of the LSF. MTF was calculated by taking the Fast Fourier Transform (FFT) of the LSF and then normalized at zero frequency. Finally, spatial resolution was reported as the spatial frequency at 10% and 50% of the MTF. The processing steps involved in the determination of the MTF from the edge image are illustrated in Figure 3.5. Figure 3.6 shows the binned ESF, LSF and the MTF obtained from an image of the edge test object with the X-ray tube voltage of 60 kV and exposure time of 0.04 s.

Figure 3.5: Schematic diagram of the processing steps for the determination of the presampled MTF.

(21)

(22)

3.4. Collimator and PID agreement test

3.4 Collimator and PID agreement test

The main function of the primary collimator is to limit the spatial extent of the X-ray field, while that of the PID is to shape and accurately guide the beam onto the area under investigation. The PID can be rotated through 360° during imaging by the operator. When the PID and the primary collimator are properly aligned, the shape and size of the beam should correspond with that of the X-ray emitting end of the PID.

Overtime the PID may be misaligned with the primary collimator due to its weight or from mechanical handling. The goal of this test was to develop a simple method to test for the proper alignment of the collimators.

The idea behind the test was to expose images with the image receptor fixated in a radiographic test object. A prototype of the radiographic test object and holder was developed by Josef Lundman based on an idea from Fredrik Bryndahl. The test object contains a grid pattern with a known grid separation distance. The grid pattern is clearly visible in exposed images. By exposing images in this configuration, any misalignment can be identified and quantitatively evaluated by measuring the extent of the X-ray field in relation to the grid pattern. For this test, two separate images are needed in order to ensure that all edges are visible on the images.

Evaluation of the test images was done manually using the image processing software ImageJ [30]. This semi-quantitative approach is time-consuming and prone to human errors, but however, offers the possibility of comparing the degree of alignment of the collimators at different times during routine QC. Figure 3.7 shows a picture of the setup and two versions of the test object.

Figure 3.7: Collimator and PID agreement test. (a) The image receptor and the test object are held fixed at the X-ray emitting end of the PID by a holder, (b) &

(c) are two versions of the test object 3D printed from a bioplastic material.

(23)

3.5. Repeatability of the methods

3.5 Repeatability of the methods

Repeatability measures for the parameters sensitivity, noise, uniformity, and low- contrast resolution across different exposure time and tube voltage combinations was performed, as suggested in the Joint Committee for Guides in Metrology report (JCGM 100:2008) [31]. This was performed at exposure times of 0.032, 0.04, and 0.05 s for tube voltage of 60 kV and at 0.04 s for tube voltage of 70 kV. Ten flat-field images were obtained for each case above under a relatively short period time between exposures. The mean value and standard deviation for each of the parameters were calculated. The repeatability measure used was the CV, calculated as the ratio of the standard deviation to the mean, expressed as a percentage. For the spatial resolution parameter, repeatability evaluation was carried out at exposure times of 0.04 and 0.08 s for a fixed tube voltage of 60 kV. Four images were obtained with the same edge test object and for each exposure setting under a relatively short period time between exposures. The CV in the spatial frequencies at 10% and 50% of the MTF for each exposure time was calculated.

(24)

4. Results

4.1 Signal transfer curve

Figure 4.1(a) shows a plot of the ROI’s mean pixel value (output signal) against exposure time. The mean pixel value increases approximately linearly with exposure time up to a point where the sensor gets saturated. From the plots, it can be seen that the receptor gets saturated at an exposure time of about 0.160 s, and 0.01 s for tube voltages 60 kV and 70 kV respectively.

Figure 4.1: (a) Plots for mean pixel value versus exposure time for tube voltages of 60 and 70 kV, (b) Signal transfer property for the Schick 33 sensor.

Using the recorded doses at the receptor measured for the tube voltage of 60 kV and at exposure times 0.02, 0.025, 0.032, 0.04, 0.05, 0.063, 0.08, 0.1, and 0.125 s, the mean pixel values were plotted against dose at the receptor to give the transfer plot shown in Figure 4.1(b). The coefficient of variation of the dose at the receptor in this measurement was approximately 0.22%. The dose at the detector measured at the saturation exposure time of 0.160 s was 0.44 mGy. A linear fit of the data gave a line with intercept −166.36 and slope 10013.30 mGy⁻¹.

(25)

4.2. Noise components

4.2 Noise components

Figure 4.2 plots the separated noise components as a function of the dose at the receptor. The total variance, the quantum, and structured noise variance components are plotted as a function of dose at the receptor in Figure 4.2(a). While the fraction of each noise variance component compared to the total noise is plotted in Figure 4.2(b). From the plots, we see that quantum noise is the dominant noise source at exposures below the saturation exposure of the receptor.

Figure 4.2: Noise component variance. (a) Linearized variance representing the total detector noise, the quantum, and structured noise components plotted as a function of dose at the receptor. (b) Quantum and structured noise variance as a fraction of total noise variance plotted as a function of dose at the receptor.

4.3 Image quality

4.3.1 Sensitivity

A plot of the receptor sensitivity measured as the mean pixel value in a centrally placed ROI is shown in Figure 4.3(a). It is seen that the mean pixel value increases linearly with exposure time.

(26)

4.3. Image quality

Figure 4.3: Variation of sensitivity and noise with exposure time for tube voltage of 60 kV. (a) Sensitivity. (b) Noise.

4.3.2 Uniformity

Figure 4.4 shows the variation of uniformity with exposure time. Uniformity at the center of the receptor was below 1%, while uniformity at the detector edges was about 1% at all exposure times in the range 0.2 - 0.125 s. The lower the value, the better the uniformity of the receptor. As expected, uniformity at the center of the receptor was better than at the edges.

Figure 4.4: Variation of uniformity with exposure time for tube voltage of 60 kV.

(a) Central uniformity, (b) edge uniformity.

4.3.3 Noise

Receptor noise measured as the CV of the mean pixel value in a centrally placed ROI as a function of exposure time is shown in Figure 4.3(b). Noise is seen to decrease almost exponentially with exposure.

(27)

4.3. Image quality

4.3.4 Low-contrast resolution

The low-contrast resolution parameter was expressed as the area under the C-D curve. Figure 4.5 shows the low-contrast resolution of the receptor plotted against exposure time. The values lie between 1 and 3 units for the exposure time range investigated.

Figure 4.5: Variation of low-contrast resolution with exposure time for tube voltage of 60 kV.

Figure 4.6 shows the calculated presampling MTF from three separate images of the edge test object obtained at different combinations of the tube voltage and exposure time. The settings were 60 kV and 0.04 s for the first image; 60 kV and 0.08 s for the second image. For the third image, the setting was 70 kV and 0.08 s. The MTF had the same shape for all three images. The calculated spatial frequencies at 10%

of the MTF were 47, 49, and 46 cycles/mm and were 16, 16, and 15 cycles/mm at 50% of the MTF respectively.

Figure 4.6: MTF measured at three different exposure settings.

(28)

4.3. Image quality

4.3.6 Repeatability evaluation

Figure 4.7 shows the evaluation of repeatability for the parameters sensitivity, noise, uniformity, and low-contrast resolution for a fixed tube voltage of 60 kV and at exposure times 0.032, 0.04, and 0.05 s. The height of each bar gives the mean value of each parameter, while the error bars represent ± one standard deviation from the mean. The variation in sensitivity and noise was low, while uniformity and low-contrast resolution varied moderately across all three exposure times.

Figure 4.7: Repeatability evaluation for sensitivity, noise, uniformity, and low- contrast resolution for tube voltage of 60 kV and exposure times of 0.03 s, 0.04 s, and 0.05 s.

The influence of tube voltage on repeatability was examined at a fixed exposure time of 0.04 s and for tube voltages 60 and 70 kV as shown in Figure 4.8. The variation in sensitivity and noise was low and similar for both voltages. The variations in uniformity (center and edge) and low-contrast resolution was lower for the higher voltage of 70 kV compared to that for 60 kV. We can see from Table 4.1 that the CV for the sensitivity parameter was less than 3%, that of the noise parameter was less than 1%, that of uniformity at the center was less than 10% and that of uniformity at the edge was less than 5%. The CV for the low-contrast resolution parameter was less than 13%.

(29)

4.3. Image quality

Figure 4.8: Repeatability evaluation for sensitivity, noise, uniformity, and low- contrast resolution at the exposure time of 0.04 s and for tube voltages of 60 and 70 kV.

Table 4.1: Coefficient of variation (%) for sensitivity, noise, uniformity, and low- contrast resolution for four different tube voltage and exposure time combination settings.

Exposure setting Sensitivity Noise Uniformity (center)

Uniformity (edge)

Low-contrast resolution

60 kV, 0.032 s 2.09 0.95 4.24 2.66 11.93

60 kV, 0.040 s 1.30 0.79 9.39 4.23 12.10

70 kV, 0.040 s 1.27 0.65 3.48 2.17 8.60

60 kV, 0.050 s 1.08 0.42 3.97 3.16 6.84

(30)

4.3. Image quality

Figure 4.9 shows the variation in the spatial resolution parameters (spatial frequency at 10 and 50% of the MTF) at a fixed tube voltage of 60 kV and for exposure times 0.04 and 0.08 s. The variations in both spatial resolution parameters were small.

The variation in the spatial frequency at 10% of the MTF decreased, while that at 50% increased with exposure time. We can see from Table 4.2 that the CV for both spatial resolution parameters was less than 5%.

Figure 4.9: Variation in spatial resolution across exposure times 0.04 s and 0.08 s for a fixed tube voltage of 60 kV

Table 4.2: Coefficient of variation (%) for MTF at 10% and 50% for voltage of 60 kV and exposure times of 0.04 and 0.08 s.

Exposure time MTF-10% MTF-50%

0.040 s 3.40 2.66

0.080 s 1.47 4.95

(31)

4.4. Collimator alignment

4.4 Collimator alignment

Figure 4.10 shows a set of images obtained with the alignment test object for a tube voltage and exposure time of 60 kV and 0.125 s, respectively. The opaque (dark) areas at the edges of the images are due to unexposed parts of the receptor. A subjective analysis was performed by looking at the images with the naked eye. The size of the opaque areas and the shape of the exposed areas were features looked at closely. The shape of the exposed central portions in both images confirmed the use of a rectangular PID. It is seen from Figure 4.10(a) that the the entire surface of the receptor was not covered by the field. Even after rotating the PID through 180°, the second image obtained Figure 4.10(b) showed similar characteristics, thus indicating that the alignment of the PID with the primary collimator is correct.

Figure 4.10: Images of collimator alignment for test. (a) Image obtained at the 0°

position of the PID, (b) Image obtained at the 180° position of the PID.

(32)

5. Discussion

The purpose of a QA program is to facilitate operators of a dental clinic in obtaining images of diagnostic quality with minimum exposure of patients and staff to ionizing radiation. My aim with this thesis was to develop ways to improve the QA of IO X-rays systems by implementing quantitative methods of image quality assessment that could track changes in the performance of the system more accurately without adding a large workload for the clinical staff. A single image receptor and an X- ray machine were used in this study. I implemented and evaluated tests for the image receptor’s sensitivity, uniformity, noise, spatial resolution, and low-contrast resolution for the combination of the Schick 33 image receptor and a Focus IO X-ray tube from GE Healthcare. A semi-quantitative test for the correct alignment of the PID with the primary collimator was also developed.

5.1 Receptor response

The relationship between the recorded incident dose at the receptor and the mean pixel value measured from an ROI at the centre of the image was used to determine the signal transfer curve of the receptor. Fourier methods in image analysis such as the approach used to determine the MTF from an image with a sharp edge are based on the assumption that the system can be approximated to a linear system [32].

The signal transfer curve was found to be linear for exposures below the saturation exposure, as seen in Figure 4.1(b). Beyond the saturation exposure, the mean pixel values did not change with exposure. This implies that images obtained above the saturation exposure will all have an identical appearance. Thus, the exposures below the saturation exposure point of the receptor was deemed useful or practical for QA purposes.

Separation of the variance in pixel values into components allows the identification of the dominant noise source in the useful exposure range. The noise model in equation 3.1 is comprised of two components: quantum and structured noise. From Figure 4.2, it is seen that quantum noise is the principal source of noise, as desired and constitutes a fraction greater than 70% of the total noise in the useful exposure range. Nonetheless, the presence of a structured noise component indicates that the variance in pixel values is not entirely due to the Poisson nature of X-ray production and detection. Also, it is observed that the fraction of the total noise due to structured noise increases with dose, while that due quantum noise decreases with dose at the detector.

(33)

5.2. Image quality

5.2 Image quality

Sensitivity was calculated as the mean pixel value from an ROI placed at the center of a flat-field image. As seen in Figure 4.3(a), sensitivity increases with exposure. The variation in sensitivity from repeated measurements was small with the CV less than 3% at all the tested combinations of tube voltage and exposure time. Uniformity was expressed as a percentage. The lower the value the better the uniformity of the receptor. The CV of uniformity measured at the center was below 5% for all the exposure settings investigated except at a tube voltage of 60 kV and exposure time of 0.04 s, where it was about 9%. A similar observation was made for the uniformity measured at the edges, where the CV was less than 3% for all the exposure settings investigated except again at a tube voltage of 60 kV and exposure time of 0.04 s, where it was about 4% (see Table 4.1). I do not have an explanation for this deviation. Routine measurement of uniformity can be helpful to identify changes in sensitivity over the entire receptor’s surface which could not be accounted for by mere random variations in pixel values.

The noise parameter was measured as the CV of the pixel value expressed as a percentage. A more appropriate way to characterize noise in a radiographic image is by computing the noise power spectrum (NPS) [13]. However, the method used in this thesis is sufficient for routine QC due to its simplicity. Deviation in noise beyond an acceptable limit may signify changes in the performance of the receptor or the X-ray tube that could not be accounted for by simply randomness in the production and detection of X-rays photons. Repeatability evaluation showed that the variation in noise was small. The CV was below 1% at all the tested combinations of tube voltage and exposure time. The area under the C-D curve decreased slightly with increased in exposure time as seen in Figures 4.5. The smaller the area, the better the low-contrast resolution of the receptor. The slight improvement in low-contrast resolution with increase in exposure is perhaps due to a reduction in noise. The variation in the low-contrast resolution was moderate with the CV between 6 - 13%.

(see Table 4.1).

The presampling MTF was determined from a single slanted edge image. Spatial resolution was measured as the spatial frequencies at 10% and 50% of the MTF.

As seen in Figure 4.6, the measured presampling MTF did not change much with exposure settings. However, the effects of other factors in the calculation such as the size of the ROI for the determination of the ESF and the method used in smoothing the ESF and the LSF needs to be investigated further. The CV of the spatial resolution parameters was less than 5% (see Table 4.2). The technique used in this work offers a simple practical approach for determining the MTF of receptors for IO X-ray imaging. The edge device is easy to fabricate and the method is less prone to physical imperfections in the fabrication of the device.

5.3 Collimator alignment test

Visual inspection of the PID and primary collimator arrangement with the naked

(34)

5.4. Application to routine QC

and prone to observer variability. The method described in this thesis uses a simple radiographic test object designed to produce images containing information related to the alignment of the collimators. A semi-quantitative evaluation of the images obtained using the collimator alignment test objects was straightforward. The large difference in contrast between an exposed and an unexposed part of the receptor’s surface in the images makes it easy to determine how much of the receptor’s surface was outside of the beam. In the situation of a perfect alignment of the collimators, the X-ray field is not supposed to cover the entire surface of the receptor. An objective method would be more consistent and suitable. My greatest disappoint- ment was that I did not figure out a method to automate the testing of the images.

Further studies should consider how this can be done.

5.4 Application to routine QC

Quantitative assessment of image quality is vital to detect changes in the performance of the system that may result in degradation in diagnostic image quality. The initial step to establish baseline performance values for the image quality parameters used might be time-consuming, but once this is done, subsequent QC testing will be straight forward. The baseline performance will be used to detect any changes in the system during QC testing. The testing frequency should be such as to ensure reasonable confidence that the system is functioning as expected between tests. The sensitivity and the noise parameters are sensitive to changes, and therefore need to be monitored more frequently. Meanwhile, uniformity, low-contrast resolution, and the spatial resolution parameters are less sensitive to changes, and thus can be tested less frequently. The recommended testing frequency for these tests is given in Table 5.1.

Table 5.1: Recommended frequency for routine QC testing.

Test Frequency

Sensitivity Weekly or daily.

Noise Weekly or daily.

Uniformity Monthly.

Spatial resolution During acceptance testing of a new sensor or whenever damage of the sensor is suspected.

Low-contrast resolution During acceptance testing of a new sensor or whenever damage of the sensor is suspected.

Collimator alignment Annually and after maintenance of the system.

The QC process can be optimized through the creation of software for automated analysis of the acquired images. In this case, the sensitivity, noise, uniformity, and the low-contrast resolution metrics could all be monitored at the same frequency from a single flat-field image.

(35)

6. Conclusion

The image quality tests implemented in this work are objective and repeatable.

Only two images: a flat-field image and an image of a sharp edge are required during each QC of the receptor. Sensitivity, uniformity, noise, and low-contrast resolution are all determined from the same flat-field image, while spatial resolution is determined from the edge image. Repeatability of the image quality parameters assessed was found to be acceptable for an automated workflow for QA testing of IO images. The setup for the assessment of the alignment of the collimators described in this work is very practical and suitable as part of a routine QC that can limit repeat examinations. The tests described here were implemented for a specific sensor and X-ray tube combination, but the methods could easily be adapted for different systems by simply adjusting certain parameters. Moreover, further work needs to be done for the present methods to be implemented into a fully automated QA program. Future studies should consider the development of a method to automate the testing of the collimator alignment test images and reproducibility measurements to ensure consistency of the methods across different sensors and X-ray systems.

Also, the long-term consistency of the methods needs to be investigated by acquiring measurements over an extended period following a planned timetable.

(36)

Bibliography

[1] P. van der Stelt, “Better imaging: The advantages of digital radiography,” The Journal of the American Dental Association, vol. 139, pp. S7 – S13, 2008.

[2] United Nations. Scientific Committee on the Effects of Atomic Radiation,

“Sources and effects of ionizing radiation: Sources,” 2000.

[3] J. Valentin, “The 2007 recommendations of the international commission on radiological protection. icrp publication 103.,” 2007.

[4] Sveriges Riksdag, “Strålskyddslag (2018:396),” 2018. [Online; accessed 20- January-2020].

[5] Strålsäkerhetsmyndigheten (SSM), “Strålsäkerhetsmyndighetens föreskrifter om anmälningspliktiga verksamheter ssmfs 2018:2,” 2018.

[6] D. Mondou, E. Bonnet, J.-L. Coudert, M. Jourlin, R. Molteni, and V. Pachod,

“Criteria for the assessment of intrinsic performances of digital radiographic intraoral sensors,” Academic Radiology, vol. 3, no. 9, pp. 751 – 757, 1996.

[7] A. G. Farman and T. T. Farman, “A comparison of 18 different x-ray detectors currently used in dentistry,” Oral Surgery, Oral Medicine, Oral Pathology, Oral Radiology, and Endodontology, vol. 99, no. 4, pp. 485 – 489, 2005.

[8] H. Udupa, P. Mah, S. B. Dove, and W. D. McDavid, “Evaluation of image quality parameters of representative intraoral digital radiographic systems,”

Oral Surgery, Oral Medicine, Oral Pathology and Oral Radiology, vol. 116, no. 6, pp. 774 – 783, 2013.

[9] K. Hellén-Halme, C. Johansson, and M. Nilsson, “Comparison of the performance of intraoral x-ray sensors using objective image quality assessment,” Oral Surgery, Oral Medicine, Oral Pathology and Oral Radiology, vol. 122, no. 6, pp. 784 – 785, 2016.

[10] J. M. Iannucci, Dental radiography : principles and techniques. St. Louis, Mo.:

Elsevier Saunders, 4th ed.. ed., 2012.

[11] A. Shetty, F. T. Almeida, S. Ganatra, A. Senior, and C. Pacheco-Pereira, “Ev- idence on radiation dose reduction using rectangular collimation: a systematic review,” International Dental Journal, vol. 69, pp. 84–97, 4 2019.

[12] J. Hubar, Fundamentals of Oral and Maxillofacial Radiology. Fundamentals (Dentistry), Wiley, 2017.

Quality Assurance of Intra-oral X-ray Images