Method of determining geometric patient size surrogates using localizer images in CT

Abstract Purpose Size‐specific dose estimates (SSDE) requires accurate estimates of patient size surrogates. AAPM Report 204 shows that the SSDE is the product of CTDIvol and a scaling factor, the normalized dose coefficient (NDC) which depends on patient size surrogates for CT axial images. However, SSDE can be determined from CT localizer prior to CT scanning. AAPM Report 220 charges that a magnification correction is needed for geometric patient size‐surrogates. In this study, we demonstrate a novel “model‐based” magnification correction on patient data. Methods 573 patient scans obtained from a clinical CT system including 229 adult abdomen, 284 adult chest, 48 pediatric abdomen, and 12 pediatric chest exams. LAT and AP dimensions were extracted from CT localizers using a threshold extraction method (the ACR DIR). The model‐based magnification correction was applied to the AP and LAT dimensions extracted using the ACR DIR. NDC was calculated using the effective diameter for the ACR DIR only, the model‐based localizer‐based and axial‐based approaches. The LAT and AP dimensions were extracted from the “gold” standard CT axial scans. Outliers are defined as points outside the 95% confidence intervals and were analyzed. Results NDC estimates for the localizer‐based model‐based approach had an excellent correlation (R2 = 0.92) with the gold standard approach. The effective diameter for ACR DIR and model‐based approaches are 8.0% and 1.0% greater than the gold standard respectively. Outliers were determined to be primarily patient truncation, with arms down or with devices. ACR DIR size extraction method fails for bariatric patients where the threshold is too high and some of their anatomy was included in the CT couch, and small patients due to the CT couch being included in the size measurement. Conclusion The model‐based magnification method gives an accurate estimate of patient size surrogates extracted from CT localizers that are needed for calculating NDC to achieve accurate SSDE.


| INTRODUCTION
Dose from medical use has increased from~15%-50% from the 1980s to 2006 with CT now representing 50% of this dose. 1 Keeping within the As Low As Reasonably Achievable (ALARA) principles is still a challenge for clinical staff including radiologists and medical physicists. 2 Quantifying absorbed dose to the patient is necessary.
The CTDIvol only represents the radiation output of a system for specific sets of conditions. [3][4][5][6][7][8] A method that scales CTDIvol with a scaling factor that depends on patient size exists. The American Association of Physicists in Medicine (AAPM) Report 204 8  can be achieved using CT localizer images. AAPM 220 charges that four sources of error be taken into account when extracting attenuation-based size surrogates from the CT localizer. However, three of four of these sources only need to be taken into account for patient size surrogate WED because it depends on patient attenuation. For this study we focus solely on the geometric size surrogates and therefore only require a magnification correction to the AP and LAT dimensions.
In a previous study conducted in our laboratory, we demonstrated a magnification/minification approach that takes into account how the edges of the anatomy are actually projected onto the image plane for both the LAT and AP dimensions. 9 These assumptions were different from other known methods [10][11][12][13][14] and the typical vendor's method which all use similar triangles to calculate the LAT and AP dimensions. The vendor's method performs a SID/SOD correction. The previous methods extend the vendor's approach by including a table offset where they assume that the x-ray intersecting the patient is at their widest extent, which is incorrect ( Table 1). The model-based magnification method approach assumes that the patient is an ellipse, and the first point of intersection between the patient and x-ray is taken into account. 9 This is because the patient's widest points as shown on the image are actually in-line with the xray projected at a point of contact on the patient that is not necessarily the widest point, as shown in Fig. 1. The approach was validated using elliptical phantoms placed at different table heights while centered in the x-direction. Table II of Burton et al. 9 demonstrated that the model-based method provides consistent accurate results, less than 1.8% of maximum error for absolute size for all measurement conditions relative to 30.9% and 7.5% for the vendor and Christensen/Raupach/Li approaches respectively. Using the model-based magnification correction approach, the patient size surrogates yield the best estimate of the actual dimension.
In this article, we evaluate our model-base magnification/minification correction of AP and LAT for NDC calculations on patient data.

2.A | Data collection
The following data were collected under a protocol that was IRB   The parameters displayed are the kilovoltage peak (kV), the Noise Index (NI) which refers to a vendor specific automatic exposure control setting, the pitch (table distance traveled in one 360 gantry rotation divided by beam collimation), the slice thickness (mm), the slice interval. Not shown is the kernel which uses "STANDARD'' (vendor specific name that refers to a soft tissue reconstruction kernel), the Reconstruction Option was set to PLUS, and the ASiR Level is 40% for all of the data shown in the

2.C | Magnification correction methods
The model-based method Burton et al. 9 9 . The CT image will give the true dimension of the patient which is why it will be used here as the "gold" standard for comparing LAT and AP dimensions, and NDC calculations which use these dimensions.
The LAT and AP were plotted for all patient data with and without the model-based approach as a function of the "gold" for all data points combined. A tight 95% confidence interval means that the data will show that many of the data points will be clus-  confidence interval range of~49 mm) meaning that AP will generally give an excellent estimate of the patient's AP provided that the ACR DIR method thresholds away the couch.   For all clinical data shown in Fig. 3, using the model-based magnification correction shows excellent agreement with the CT axial "gold" standard for lateral and AP dimensions. In Fig. 3(a), on average the LAT M for ACR DIR and LAT for model-based are 6.0% greater and 0.14% less than the "gold" standard respectively. In Fig. 3(b), on average AP M for ACR DIR and AP for the model-based method are 11.0% and 2.0% greater than the gold standard respectively. Figure 3 shows that the UW linear fits of D E as a function of (AP + LAT)/2, LAT, and AP compare well to the fits in the AAPM Report 204. ICRU92 were elliptical. The LAT dimensions over 400 mm from the AAPM Report 204 are the only data outside of our 95% confidence interval. These data agree well with Burton and Szczykutowicz. 17 which demonstrate a similar result using CT axial scans. In Fig. 4 there is excellent correlation of model-based magnification correction with the "gold" standard NDC. On average, the NDC for ACR DIR and model-based method are 10% greater and 0.8% greater than the "gold" standard respectively.

3.C | NDC comparisons
We explored the outlier points that fell outside of the 95% confidence intervals in Figs. 2(a) and 2(b). The most prevalent outliers are those where the ACR DIR either underestimates Fig. 3(a) or overestimates Fig. 3(b) the patient size for bariatric and pediatric cases, respectively. Figure 3(a) shows the bariatric patients with LAT dimensions roughly between 450-500 mm do not continue the linear trend and fall outside of the 95% confidence interval. This is because the ACR DIR method failed for bariatric patients due to underestimation that some of their anatomy was classified as belonging to the CT couch and thresholded away.

CONF LICTS OF INTEREST
The author has no conflicts of interest to disclose.
T A B L E 3 The mean error of effective diameter and SSDE from the ACR DIR method with and without magnification correction to the CT axial-based measurements. The regression (R 2 ) value is shown for both methods.