Blocker-based techniques

In the first study of scatter in mammography, Barnes and Brezovich [105] measured experimentally SP R values for different parameters. They used the conventional tech- nique employed for scatter measurements in diagnostic radiography: the beam stop technique. Later on, this methodology was employed in X-ray mammography for a large number of authors to estimate both SP R and SF values [25, 104, 114, 115, 116]. The beam stop technique consists of imaging an object made of a high density material, typically lead (Pb), which absorbs the majority of the primary X-ray photons impinging from above. Normally, the beam stops are imaged with a breast tissue equivalent material, usually PMMA, to generate realistically scattered radiation within the image receptor.

Figure 3.4 illustrates the typical geometry of the beam stop where a lead disc, i.e. beam stop, is placed above the scattering material.

Figure 3.4: Typical diagram of a beam stop geometry.

As the beam stop is made of a high absorption material, most of primary radiation does not penetrate it, only the scatter signal S being observed beneath the beam stop.

46 Chapter 3. Scatter and its effects on mammograpghy

In contrast, the energy recorded due to primary and scatter components (P + S) are found elsewhere. As a results, both S and (P + S) signals are combined to calculate the primary only image P and eventually, the SP R and SF for the studied geometry can be estimated as described in Equations 3.7 and 3.8 respectively.

Using this principle, the SP R and SF values are estimated using a series of lead discs of similar thickness but decreasing diameter. An example of this geometry used for scatter beam stop assessment is illustrated in Figure 3.5(a). SP R, or SF , under each disc is derived from the scatter signal (S) and primary and scatter (P + S) measurements as explained above. Typical regions of interest (ROI) used to measure the signal inside (S) and outside (P + S) the lead disc are shown in Figure 3.5(b). Note that to account for non-uniformities in the image, such as the heel effect2, four ROIs are used to measure the signal in the background in different positions with respect to the disc.

To illustrate this, a 5cm thick PMMA slab was experimentally imaged in a Hologic Selenia X-ray system without an anti-scatter grid using an energy spectrum of 29kVp Mo/Rh. The scatter magnitude was calculated using the aforementioned beam stop methodology (See Figure 3.5(a)). The SP Rs were measured for each disc, then, the SP R value of the system was finally calculated after extrapolating the individual SP R values to a zero diameter disc as illustrated in Figure 3.5(c). In this case shown, SP R values of 1.00 and 1.18 were obtained when extrapolating to zero diameter using linear and logarithmic methods respectively.

Boone and Cooper [19] found that linear extrapolation might underestimate (small discs) or overestimate (large discs) the SP R whereas logarithmic extrapolation over- estimates the SP R in all the cases. So it is observed that the chosen method for extrapolation influences in the final SP R value, which represents a limitation of this method.

The above measurements using the beam stop method provide a single value to repre- sent the scattered radiation for a given geometry. However, as discussed in the litera- ture [19, 45, 46, 47] and in Section 3.4.6, the scatter magnitude depends on the spatial location where it is measured. In order to study the scatter distribution across the image, a phantom consisting of an 2D array of lead discs of uniform thickness equally distributed has been described in the literature [20, 117], and illustrated in Figure 3.6. The use of this phantom requires measurement of the SP R or SF values for each lead disc as previously described. Intermediate values can be interpolated assuming that the scatter is a slowly varying function.

The scattering due to light within the image receptor, i.e. veiling glare, is neglected in the MC simulations described in this work. However, as the beam stop images are acquired experimentally, the glare is intrinsic to the scatter measurements. It is worth pointing out that some authors have estimated the glare within the image receptor using the beam stop methodology [116, 118]. In this case, the scattering material is removed from the geometry and the lead disc is placed directly on the surface of the detector. Thus scatter produced within the detector is only recorded under the disc. Despite the beam stop being is used very often to experimentally estimate scattered radiation, it suffers from some drawbacks. For example, it requires of a large number

2

3.3. Techniques for scatter estimation 47

(a) (b)

(c)

Figure 3.5: (a) shows a common beam stop geometry consisting of lead discs of equal thickness and decreasing diameter. Common ROIs used to measure the signals beneath and elsewhere the discs are illustrated in (b). An example of extrapolation to 0 diameter using linear and logarithmic approximation is presented in (c). For this case, a 5cm thick PMMA slab has been imaged in a Hologic Selenia mammographic system without anti- scatter grid using an energy spectrum of 29kVp Mo/Rh.

of exposures for the image acquisition process [106]. This is necessary to quantify the statistical fluctuations associated to each value measured. Moreover, if the lead disc is too small, this can fall across pixels and suffer from partial volume effects, so several measurements with the beam stop phantom shifted are needed to reduce this effect. Furthermore, as mentioned above, the method of extrapolation used to calculate the scatter at zero diameter disc may lead to inaccuracies in the results as different approximations can provide different results. This has been a controversial point widely discussed in the literature [106].

For this reason, Cooper et al. [106] and Nykanen and Siltanen [108] described a method, previously proposed for Chan and Doi [119] in diagnostic radiology, which not only estimates the scatter magnitude but also its spatial distribution by measuring the edge spread function (ESF).

48 Chapter 3. Scatter and its effects on mammograpghy

Figure 3.6: Typical beam stop array used to estimate the SP R and SF across the entire image receptor experimentally. The black dots represent lead discs of uniform thickness.

They showed that two images of an edge spread device (ESD), which is made of lead, are needed to estimate the scattered radiation. Thus, their proposed methodology reduces the large number of acquisitions needed for the beam stop method. In the first image, the geometry is setup as shown in Figure 3.7(a). For the second, the ESD is re-orientated 180o, so a mirror image is generated.

(a) (b)

Figure 3.7: A diagram of the setup described by [106] to calculate the ESF is shown in (a). A representative intensity profile perpendicular to the edge of the ESD is illustrated in (b).

The ESF is calculated by taking a profile perpendicular to the edge of the ESD. A typical profile is shown in Figure 3.7(b). In this profile, the (P + S) signal recorded outside the ESD is reduced as the ESD is penetrated, where the toe of the sigmoid corresponds to the scatter component. As mentioned before, two mirror images are generated, so

3.3. Techniques for scatter estimation 49

by subtracting the different parts of the profile, scatter and primary signals can be separated. Therefore, the scatter line spread function (LSF) for a particular direction is calculated as the derivative of the scatter ESF.

The scatter PSF is eventually calculated by deriving the scatter LSF. However, as the PSF provides spatial information in all directions, and the LSF in just one direction, the scatter PSF is assumed radially symmetric. Otherwise, the LSF in more directions are needed, which would increase the number of image acquisitions.

It has been seen that the number of acquisitions can be greatly reduced compared to the pencil beam experiment. However, this method has also some disadvantages. In order to reduce the number of acquired images, the scatter response was assumed symmetric, so it might fail in non-symmetric regions, such as close to the edge of an object. However, this can be taken into account by measuring the LSFs in more directions, which would increase the number of images required as previously described. Furthermore, the image acquisition of two geometries must be acquired under the same conditions. In other words, the ESD used to calculate the ESF have to be carefully placed in order to obtain a mirror profile.