Key words abdomen - image manipulation/reconstruction - CT - ureter - urinary tract - urolithiasis
Introduction
Multidetector computed tomography (MDCT) has already proven its high diagnostic value
by achieving a high sensitivity (98 %) and specificity (96 %) in the detection of
urinary stone disease [1 ]
. Over the last decade, the number of MDCT scans performed for the detection of urinary
stone disease has continuously increased [2 ].
With a lifetime-estimated recurrence rate of approximately over 50 % [3 ], patients suffering from urolithiasis are often repeatedly referred for cross-sectional
imaging. The portion of young patients affected is above-average [4 ].
In combination with the growing accessibility and usage of MDCT in these cases, the
scans have evolved to become an increasing part of population radiation exposure.
As a result, development of low-dose protocols was needed. After broad implementation,
these have proven to be efficient and further accepted as a standard procedure for
the diagnosis of urolithiasis [5 ]
[6 ]
[7 ]. Early ethical concerns [8 ]
[9 ] have been invalidated [10 ].
However, with this development growing image noise proved to be the trade-off for
lowering radiation dose levels. Iterative reconstruction (IR) tools have been shown
to be successful in compensating for this loss in image quality at the existing dose
reduction levels in the past [11 ]
[12 ]
[13 ]
. With ongoing improvement of IR tools and its ability to decrease even higher image
noise ratios, the radiation dose reduction potential changes and needs to be re-evaluated
each time.
Most recently, a new generation of IR tools called Iterative Model Reconstruction
(IMR® ; Philips Healthcare, Best, the Netherlands) was introduced. With the underlying target
being to once again lower radiation dose exposure “as low as reasonably achievable”
(ALARA), the new potential capability of IMR to allow for further lowering has to
be evaluated.
Thus, the purpose of this study is to assess the impact of IMR on reader confidence
with respect to stone detection and on image quality in comparison to FBP and iDose4
in abdominal MDCT with radiation doses below 2 mSv.
Materials and Methods
Patient collective
In this retrospective study the data from 32 consecutive patients (21 men; mean age:
53.5 years; range: 20 – 82 years) with suspected urinary stone disease were analyzed.
The study was reviewed and approved by the local ethics committee. Due to the retrospective
nature of the study protocol, the ethical review board waived the need for written
informed consent.
MDCT image acquisition
MDCT examinations were performed using a 256-slice MDCT scanner (Brilliance iCT, Philips
Healthcare, Best, The Netherlands). All patients underwent non-contrast MDCT of the
upper urinary tract performed in craniocaudal direction from the top of the kidneys
to the bottom of the urinary bladder. As only a prototype version of IMR was available
in our study, which was neither FDA- nor CE-cleared, regular diagnosis based on images
reconstructed with IMR was not allowed. Hence, the dose settings that were appropriate
for iDose, which was used for regular diagnosis, had to be applied, using the following
parameters: tube voltage of 120 KV; reference tube current time product of 40 mAs,
corresponding to a CTDIvol of 2.7 mGy for a standard-sized patient; gantry rotation
time of 0.4s; pitch of 0.922; collimation of 64 × 1.25 mm. Automatic exposure control
system (ACS) in combination with z-axis dose modulation (Z-DOM) was activated.
MDCT image reconstruction:
The raw data of all MDCTs were reconstructed with FBP, iDose4 (level 4) and three different IMR levels (IMR level 1 (IMR L1), level 2 (IMR L2)
and level 3 (IMR L3)) allowing for an intraindividual comparison between all reconstruction
methods.
IMR reconstruction was performed with the “Body Routine” setting. The different levels
of IMR refer to the degree of noise reduction provided by the IMR software. In the
case of FBP and iDose, the standard body kernel (B) was applied. A slice thickness
of 3 mm with 50 % overlap was used for all data sets. The recommended fixed setting
“BodyRoutine” was used for IMR.
Subjective image evaluation
Multiplanar reconstructions were created for every examination. A commercially available
PACS system (Centricity PACS-IW, GE Healthcare, Chicago, Illinois, United States)
was used for image interpretation. Two radiologists with 5 and 10 years of abdominal
MDCT experience independently evaluated all data sets in a blinded fashion.
All reconstructions were randomized and six separate review sessions were performed
in order to achieve a maximum level of blinding.
The subjective image quality was rated based on a 5-point Likert scale defined as
follows:
poor image quality, major noise, with poorly defined anatomic details.
reduced image quality, substantial noise, with limitations in anatomic details.
acceptable image quality, moderate noise.
good image quality, minor noise, with clear anatomic details.
excellent image quality, no noise, with distinct anatomic details.
Ratings were separately performed for the renal pelvis, the urinary bladder, and the
proximal, middle and distal third of the ureteral course. The rating of anatomical
structures was mainly influenced by the ability to clearly outline the displayed anatomical
borders with particular regard to the urinary and renal tract.
For the determination of image appearance and the known blurring effect of IR techniques,
the 5-point Likert scale targeted the “blotchiness” of the entire MDCT data set influencing
the detailed depiction of anatomical details only:
major blotchy appearance.
increased blotchy appearance.
moderate blotchy appearance.
minor blotchy appearance.
no blotchy appearance.
Reader confidence with respect to diagnosing urolithiasis was graded dividing the
data into the sections identical to those performed for the analysis of the subjective
image quality (renal pelvis, proximal, middle and distal ureteral third and the urinary
bladder, respectively). Ratings were performed for both the left and the right side
separately, applying the following definition of calculus score:
definitely no calculus
probably no calculus
indeterminate
probably calculus
definitely calculus
Finally, consensus readout was performed and the reference standard was defined. To
determine the diagnostic value, definite stone detection or exclusion groups were
determined. A relation to possible findings was set and a ‘confidence’ index was generated,
which is defined as
Thus, reader confidence with respect to diagnosis was evaluated. A subsequent correlation
of the radiologist’s diagnosis and clinical findings of urolithiasis or the lack thereof
was not carried out. Thus, a corresponding reference standard was not established.
Objective image evaluation
Radiation dose and noise levels
For every MDCT scan a dose report was generated. All relevant dose parameters including
CT dose index volume (CTDIvol ), dose-length product (DLP) and tube voltage (KV) were documented.
For the quantitative analysis the Hounsfield Units (HU) and the standard deviations
(SD) were determined by placing circular regions of interest at the following locations:
the subcutaneous fat at the dorsolateral back; the right subcutaneous gluteal fat;
the aortic lumen; the internal obturator muscle and the urinary bladder. These measurements
were performed on three adjacent slices and mean values were obtained.
Statistical analysis
Continuous variables were expressed as mean ± SD. Categorical variables are presented
as mean + 95 % confidence interval. Intergroup comparisons between ≥ 3 related groups
were assessed applying the Friedman-test. Post-hoc analysis between 2 related groups
was performed using the paired Student’s t-test for normally distributed data and
the Wilcoxon signed-rank test for skewed data samples. Interobserver agreement between
the two reviewers was assessed by weighted kappa. Statistical significance was defined
for 2-tailed p- values of less than 0.05. Data collection and statistical calculations were performed
using SPSS Statistics 22 software (IBM Inc. SPSS Statistics, Chicago, IL).
Results
Subjective image quality
The subjective image quality results for each part of the urinary tract are displayed
in [Fig. 1 ]. First, a mean score for the image quality of the upper urinary tract was generated
for all segments. The scores substantially increased from FBP to iDose4 as well as from iDose4 to IMR. FBP scored an average of 2.0 points on the Likert scale. iDose4 accomplished an average score of 2.9. IMR level 1 was rated to perform with a substantial
increase in the subjective image quality (4.2) and a decrease of image noise. With
further noise reduction by the IMR software at the higher levels 2 and 3, the mean
score for image quality slightly decreased to an average of 4.0 and 3.9, respectively.
Fig. 1 Impact of FBP, iDose4 and IMR levels 1 – 3 on mean subjective image quality for each
section of the urinary tract.
Objective image quality
The CT numbers remained constant for each anatomical structure. No substantial differences
between iteration tools or levels were found. No pattern of deviation could be related.
The observed extent of variation ranged from 0.5 % to 2.0 %. The mean SD determined
for all anatomical landmarks showed a considerable decrease when applying iDose4 and IMR in comparison to FBP. Using iDose4 with level 4, image noise was reduced by 45 % on average. IMR L1 allowed for an even
greater reduction of 70 %. With IMR L3 the highest noise reduction of 79 % was achieved.
The constancy of the mean CT numbers as well as the reduction in mean SD were calculated
for each anatomical structure as shown in [Fig. 2 ].
Fig. 2 Continuous decrease in noise levels from FBP to iDose4 and ongoing from IMR levels
1 to 3 in the urinary bladder as an example. No substantial alteration in CT numbers
(HU mean) can be observed.
Reader confidence with respect to stone detection
The results show an increase in reader confidence with respect to definite stone detection
and exclusion from FBP to iDose4 and to IMR. FBP achieved a confidence index of 66 %, iDose4 of 81 % and IMR of 96 % independent of the level ([Fig. 3 ]). Both radiologists reported a high influence of the extent of distinction of anatomical
structures on their confidence regarding stone detection. This particularly affected
evaluation in the area of the lower ureter and bladder.
Fig. 3 Rising reader confidence in stone detection from FBP to iDose4 and to IMR levels
1 – 3 (L1 – 3).
A drop of only 2 % with IMR level 1 and 3 % with levels 2 and 3 from the upper to
the lower urinary tract was observed. Subsequently, a nearly consistent high level
of reliability with IMR level 1 (96 %) and level 2 (97 %) was shown. With FBP and
iDose4 , however, a drop of 26 % (FBP) and 16 % (iDose4 ) was shown.
Image appearance and noise
No blotchy appearance was found on the images reconstructed using FBP (mean value
of 5). Alterations in the image appearance were noted only with the introduction of
iDose4 or IMR. In iDose4 the blotchiness was assessed to be between minor and no blotchy appearance (mean
value of 4.6). With further influence of the IMR reconstruction technique, increasing
blotchiness was observed (mean values of 3.7 (IMR L1), 3.2 (IMR L2) and 2.8 (IMR L3)),
resulting in a moderate blotchy image appearance.
Radiation dose
The mean effective tube current time product was 41 ± 12 mAs (range 19 – 80 mAs) with
a fixed tube voltage of 120 KV. The mean CTDIvol was 2.7 ± 0.8 mGy (range 1.29 – 5.44 mGy). On average, the DLP amounted to 126 ± 38 mSv × cm
(range from 49 – 253 mSv × cm). By applying the conversion factor of k = 0.015 mSv/mGy × cm
from AAPM report 96, this resulted in an overall mean effective dose of 1.9 ± 0.6 mSv.
Discussion:
MDCT has been shown to be highly reliable in the diagnosis of urinary stone disease
[14 ]. However, the challenge to diminish radiation dose exposure remains. With the initial
introduction of IR techniques, a continuous reduction of MDCT dose settings with a
simultaneous positive influence on image quality was achieved in the past.
Several studies have evaluated the impact of IR on image quality parameters as well
as on the reader’s diagnostic confidence with regard to pinpointing pathological findings.
Not only the detection of urolithiasis but also various other types of CT examinations
and findings were investigated [15 ]
[16 ]
[17 ]. With the very recent introduction of the next generation IR tool IMR, this study
aimed to evaluate the changes of these effects in comparison to its predecessors FBP
and iDose4 .
Our results show that IMR reduces image noise and increases reader confidence with
respect to stone detection to a large extent. This is reflected in substantially reduced
noise levels (reduction of 70 % to 80 % in comparison to FBP and of 50 % to 60 % in
comparison to iDose4 (level 4)) and improved reader confidence with regard to stone detection (from 66 %
with FBP to 81 % with iDose4 and 96 % with each IMR level ([Fig. 3 ])). Thus, an increase of up to 30 % from FBP to IMR and of 15 % from iDose4 to IMR was observed, which represents a significant improvement.
However, with growing influence of the IMR algorithm in higher IMR levels, especially
IMR level 3, a noticeable altering of image appearance can be observed. This alteration,
in other studies already described as “blotchiness” or a “waxy” appearance, refers
to blurring with a loss of sharpness of anatomical contours [14 ]
[18 ]. This visual phenomenon accounts for the reduction of subjective image quality as
well as the image appearance score despite an ongoing noise reduction from IMR level
1 to level 3 ([Fig. 4 ]).
Fig. 4 Observed continuous decrease in image appearance with further impact of iteration
techniques and levels due to greater “blotchiness” and softening of anatomical contours.
The decrease in image appearance from IMR L1 to L3, however, had no negative impact
on reader confidence regarding stone detection. This is due to the fact that the high
density of urinary calculi allows easy identification in non-enhanced abdominal MDCT
independent of the reconstruction technique ([Fig. 5 ]). Especially in IMR L3, discrete alterations of anatomical details had no significant
impact on reader confidence. Still it has to be taken into account that no clinical
reference standard was defined. Instead, the problem reported by both readers was
not the identification but in some cases the reliable assignment of possible calculi
to part of the urinary tract. This problem was mainly caused by higher noise levels
arising partly from lower noise reduction potential, especially by FBP, and partly
from anatomical regions often related to higher noise levels and artifacts. Referring
to the urinary tract, these regions are the middle and lower third of the ureter as
well as the urinary bladder and its ostia ([Fig. 1 ], [5 ]). In many cases this problem is essential for differentiation between urolithiasis
and nonspecific calcifications such as phleboliths or calcified lymph nodes.
Fig. 5 Urinary stone detection in the lower left ureter. A: FBP; B: iDose4; C: IMR L1; IMR
L2; IMR L3. The example demonstrates the ongoing decrease in image appearance with
higher iteration levels. Due to increasing delineation of the ureter, the assignability
of the concrement improves.
In this study reader confidence regarding stone detection varied greatly from the
upper urinary tract to the lower. This change in reader confidence, however, directly
depends on the IR tool being used. With FBP the confidence index dropped by 23 % (from
69 % to 43 %) and with iDose4 by 16 % (from 84 % to 68 %) from the upper to the lower urinary tract, whereas a
drop of only a few percent was observed with IMR (L1 from 96 % to 94 %, L2 and L3
from 97 % to 94 %). This leads to the conclusion that even though higher IMR levels
alter image appearance and even very fine anatomical structures can potentially be
lost [19 ], the diagnostic value for this specific task substantially improves. Nevertheless,
the discussion of missing a possible differential diagnosis with altering image appearance
as stated by Veldhoen et al. [14 ] remains.
However, CT examinations for urolithiasis especially concern young patients who often
undergo multiple CT scans over the years [14 ]. The obligation to consider radiation dose-related cancer risks remains unchanged
[4 ]
[20 ]
[21 ]. Based on our results, as well as similar results [15 ], the newly introduced radiation dose reduction potential offered by IMR has to be
discussed.
Our study achieved an average effective dose of 1.9 mSv without altering the already
implemented standard dose settings. In consideration of the significant improvements
in reader confidence, a further dose reduction should be possible with IMR. This was
shown by Park et al. as well as other studies [22 ]
[23 ]
[24 ]
[25 ], which achieved average effective dose levels as low as 0.68 mSv. Future studies
need to evaluate the full dose reduction potential, taking the loss of a possible
differential diagnosis and the alteration of image appearance into account. The trade-off
needs to be evaluated and optimal dose settings have to be identified.
This study had the following limitations:
As already mentioned, the prototype version of IMR used in this study had no FDA or
CE clearance. To ensure regular diagnosis without IMR, the impact of reductions below
the level that is appropriate with iDose could not be investigated.
Based on the fact that a clinical correlation of the radiologist’s interpretations
did not take place, the possible radiological diagnoses of urolithiasis could not
be confirmed or disproved. Thus, a clinical reference standard is not defined.
Even though the radiologists were blinded, an ideal blinding could not be assured
in all cases due to the visible differences between each iterative reconstruction
technique (FBP, iDose4 and IMR), such as noise levels and sharpness of anatomical structures.
A further limitation was that possible differential diagnoses were not investigated.
IMR proved to have a positive effect on reader confidence regarding the diagnosis
or exclusion of urolithiasis. On the other hand, the loss of a possible differential
diagnosis due to an alteration in image appearance by the IMR algorithm, especially
in IMR L3, remains uncertain.
Furthermore, the size of possible stones and a final confirmation of determined stones
were not established in this study.
In conclusion, IMR substantially exceeds the capability of its predecessors iDose4 and FBP to improve image noise levels as well as subjective image quality for non-enhanced
low-dose MDCT of the urinary tract. Thus, on the basis of missing clinical correlation,
considerably better subjective reader confidence was achieved.
The results of the presented study allude to a further potential to reduce radiation
dose exceeding the dose settings applied. However, the extent of this potential needs
to be clarified in future studies.
