Combining dose, size and image quality to identify outliers in CT examinations

February 10, 2020

An Dedulle1,2, Jurgen Jacobs1, Hilde Bosmans2,3, Niki Fitousi1
1 Qaelum NV, Gaston Geenslaan 9, 3001 Leuven, Belgium
2 Department of Imaging & Pathology, KU Leuven, Herestraat 49, 3000 Leuven, Belgium
3 Department of Radiology, University Hospitals of Leuven, Herestraat 49, 3000 Leuven, Belgium

Abstract Body

Based on accepted and presented research on ECR 2019 (Dedulle et al. Evaluating the use of a noise index to identify outliers in CT examinations, European Congress of Radiology, Vienna 2019).

Purpose

Advanced dose monitoring systems are able to automatically identify and inform the user for radiation doses above certain dose threshold levels (e.g. diagnostic reference levels). This function though may miss cases where a dose falls below the defined threshold, yet is not proper for this patient. Evaluating just the tube output is not enough to provide an answer for the correctness of the technique, if for example the patient is thin but received high dose or overweight and received low dose. The purpose of this study is to evaluate which outliers would be identified if multi-parameter alerting would be implemented, such as the combined use of patient size, CTDIvol and a noise index for clinical chest CT examinations.

Methods

Fifty chest CT examinations (Siemens Somatom Force, Germany) with the same study protocol were used. The examinations were analysed by the dose management system DOSE (Qaelum, Belgium), installed in the University Hospitals Leuven. The required examination and dosimetric data for the purpose of the study, such as the Water Equivalent Diameter (WED) as a metric for patient size, the CTDIvol of each acquisition and the vertical offset, were extracted by DOSE.

In parallel, the noise of the images was evaluated in terms of the Global Noise Index (GNL). From each examination, the CT slices from the expiration series with the lung (Br59d) kernel and slice thickness of 1mm were used. The GNL was calculated following the methodology of Christianson et al. (AJR 2015) and Ria et al. (Med Phys 2017). The algorithm was applied on ten equally spaced CT slices of the acquisition and the final GNL of each acquisition was the average of the ten slices.

1. First the slice was segmented into tissue types. For the analysis, the soft tissue with Hounsfield Units (HU) range from -100 to 300 was used.
2. For each soft tissue pixel, the standard deviation was calculated in a central kernel with a size of 6mm. This resulted in a noise map with the standard deviation (in HU) of the surrounding kernel for each soft tissue pixel (figure 1B).
3. A histogram of the soft tissue noise map (with the standard deviations) was generated. The GNL of that slice was determined as the mode of the histogram (figure 1C).

Combining dose size and image quality to identify outliers in CT examinations f1

Figure 1: Schematic overview of the noise detection algorithm (Christianson et al. AJR 2015, Ria et al. Med Phys 2017). A) Original Image, B) Noise map of soft tissue, C) Histogram of soft tissue noise map.

Next, the CTDIvol of the acquisition was correlated with the patient size in terms of WED. Additionally the GNL was added as a third parameter in the analysis. Outliers for the two- and three parameter analysis were examined.

Results

An exponential increase of the CTDIvol with increasing WED, was observed (figure 2A). The correlation of the WED and CTDIvol was significant (p < 0.01). The three dose outliers that were observed had a CTDIvol higher than expected for their patient size. Two of them (Nr. 2 & 3) could be explained by investigating the vertical positioning of the patient during the examination. Particularly, both of them were positioned closer to the tube during the localizer (> 3cm), causing magnification, for which literature demonstrates an increase of up to 30% for the CTDIvol of the spiral acquisition. Additionally, outlier number 2 had one arm down and in the beam path during the localizer and acquisition, which also results in an increase of CTDIvol. Similarly, outlier number 1 had both arms down and in the beam path, thus positioning was again the reason for the increase in CTDIvol.

The GNL was added as a third parameter to the CTDIvol vs WED analysis (figure 2B). A higher GNL (red) means more noisy images and thus a lower image quality. Outlier 1 and 2 had a relatively low GNL and thus a relatively high image quality. This was expected due to the higher CTDIvol of these cases. The third outlier had a GNL around the median of all examinations, since the increased CTDIvol was partially compensated by the increased attenuation of the arms in the beam path. Interestingly enough, a fourth outlier appeared, with higher GNL than the other examinations. In this case, the patient was positioned further from the tube (> 4cm), which caused minification on the localizer image. The patient resembled smaller and this caused a decrease of the CTDIvol during the spiral acquisition and thus an increase in GNL and a lower image quality. This outlier was missed when only two parameters were correlated (CTDIvol vs WED).

Combining dose size and image quality to identify outliers in CT examinations f22

Figure 2: A) The CTDIvol of the chest acquisition as a function of the water equivalent diameter (WED) of the patient, B) the global noise level (GNL) was added as a third parameter. The outliers are marked with a number.

Conclusion

Tube output versus patient size gives a first indication of potential outliers. However, it is not always enough in order to identify all outliers and hidden problems. Extra relevant outliers could be detected by including an automated image quality index. In the current study, most outliers were caused by the patient positioning.
Calculation and combination of relevant parameters by a dose management system can save time to the user by highlighting real outliers and potential underlying problems.

Keywords

Thorax, CT, Dosimetry, Physics, Quality Assurance

Disclosure statement

• An Dedulle has a PhD grant from the Flanders Innovation & Entrepreneurship agency [Grant No. HBC.2016.0233] and works as a PhD researcher for Qaelum NV in cooperation with the University of Leuven.
• Niki Fitousi is head of research in Qaelum NV.
• Jurgen Jacobs is CEO and co-founder of Qaelum NV.
• Hilde Bosmans is board member and co-founder of Qaelum NV.

Qaelum respects your data privacy. Why cookies? Simply because they are used to help the website function, to improve your browser experience. Click 'I understand' to accept cookies and go directly to the site or click 'More information'.

I understand