Three-dimensional CT imaging in extensor tendons using deep learning reconstruction: optimal reconstruction parameters and the influence of dose

The purpose of this study was to assess the optimal reconstruction parameters and the influence of tube current in extensor tendons three-dimensional computed tomography (3D CT) using deep learning reconstruction, using iterative reconstruction as a reference. In the phantom study, a cylindrical phantom with a 3 mm rod simulated an extensor tendon was used. The phantom images were acquired at tube current of 50, 100, 150, 200, and 250 mA. In the clinical study, CT scans of hand tendons were performed on nine hands from eight patients. All images were reconstructed using advanced intelligent clear-IQ engine (AiCE) parameters (body, body sharp, brain CTA, and brain LCD) and adaptive iterative dose reduction three dimensional (AIDR 3D). The objective image quality for tendon detectability was evaluated by calculating the low-contrast object specific contrast-to-noise ratio (CNRLO) in the phantom study and CNR and coefficient of variation (CV) in the clinical study. In the phantom study, CNRLO (at 200 mA) of AiCE parameters (body, body sharp, brain CTA, and brain LCD) and AIDR 3D were 5.2, 5.3, 5.3, 5.8, and 5.0, respectively. In the clinical study, AiCE brain CTA was higher CNR and lower CV values compared to other reconstruction parameters. AiCE without dose reduction may be an effective strategy for further improving the image quality of extensor tendons 3D CT. Our study suggests that the AiCE brain CTA is more suitable for extensor tendons 3D CT compared to other AiCE parameters.


Introduction
Extensor and flexor tendons can be injured in hand trauma, rheumatoid arthritis, and osteoarthritis, and they are commonly examined using echography, magnetic resonance imaging (MRI), and computed tomography (CT) [1][2][3][4][5][6].Among these examinations, CT can identify the extensor and flexor tendons as a three-dimensional (3D) image with volume rendering [4][5][6].It has been reported that 3D CT is valuable for confirming the clinical diagnosis of these tendons [5,6].Accurate 3D images that can determine the condition and location of tendons may preclude the need for surgery and assist with surgical planning.However, there are some limitations to 3D imaging of extensor tendons, such as the difficulty in assessing extensor tendons distal from the metacarpophalangeal joint and the uncertain identification of the extensor digitorum communis tendon (little finger) [6].Creating 3D images of extensor tendons is often challenging due to their thin structure, proximity to the skin and bone, and the smaller CT value difference between structures such as tendons and muscles compared to flexor tendons.Therefore, reducing image noise and improving contrast between tendons and muscles are important for acquiring high-quality 3D images of extensor tendons.The 3D CT scan of tendons using a high tube current setting is recommended to reduce image noise [7].
Recently, a deep learning-based reconstruction (DLR) technique using a deep convolutional neural network has been introduced.A commercially available DLR for CT by Canon, advanced intelligent clear-IQ engine (AiCE, Canon Medical Systems, Otawara, Japan), is trained to differentiate signal from noise and can suppress noise.Consequently, the AiCE reconstruction technique has improved image noise and allows for a reduction of radiation dose while maintaining image quality when compared to CT images using conventional iterative reconstruction (IR).It has been reported that AiCE is beneficial in various regions, including the chest, cardiac, abdomen, CT angiography, and orthopedics [8][9][10][11][12][13][14].However, it is not clear whether extensor tendons CT images reconstructed using AiCE enable a decrease in image noise and a reduction of radiation dose while enhancing the identification of extensor tendons on 3D images.In addition, while the manufacturer-recommended AiCE parameter for soft tissue is AiCE body sharp, the optimal AiCE reconstruction parameter for 3D CT of extensor tendons remains unclear.As a result, we conducted an investigation to determine the optimal AiCE reconstruction parameters and the influence of radiation dose when using AiCE for 3D CT of extensor tendons.The purpose of this study was to evaluate the optimal reconstruction parameters and the influence of tube current in extensor tendons 3D CT using AiCE, using IR as a reference.

Phantom design
In this study, we used a cylindrical phantom with a 3 mm acrylic cylindrical rod to simulate an extensor tendon (Fig. 1).The diameter of the phantom was 50 mm.The phantom was filled with diluted contrast agent.The CT value difference between the cylindrical rod and the background was set to Δ20 Hounsfield units (HU) to correspond to the contrast between extensor tendons and muscles in CT of the hand tendons.

Patient population
A total of eight patients were included in this study.All patients underwent CT of the hand tendons from September 2021 to September 2022 at our hospital in Gifu, Japan.We evaluated CT images of nine hands in these patients (only right hands, three patients; only left hands, four patients; both hands, one patient).The gender distribution was 50% men.The average age was 61 ± 9 years.

CT image acquisition
All CT images were acquired using a 80-detector row CT scanner (Aquilion PRIME SP; Canon Medical Systems, Otawara, Japan).The CT acquisition parameters were as follows: acquisition mode, helical; gantry rotation time, 0.75 s/rotation; calibration field-of-view (FOV), 320 mm; tube voltage, 120 kV.
In the phantom study, CT images were acquired with tube currents of 50, 100, 150, 200, and 250 mA.The volume CT dose indexes (CTDI VOL ) for each tube current were 2.6, 5.2, 7.8, 10.5, and 13.1 mGy.The scans were repeated 50 times.Reconstruction conditions were set to a slice thickness of 1.0 mm and a display-FOV of 100 mm.
In the clinical study, CT images were acquired with tube current adjusted by automatic exposure control with a reference noise index of 2.0-2.5.Reconstruction conditions were set to a slice thickness of 1.0 mm and a display-FOV of 100-130 mm.
Tube currents for both the phantom and clinical studies were determined based on the standard protocol for finger tendons set by the Japanese Society of Radiological Technology [7].
All CT images were reconstructed with adaptive iterative dose reduction three dimensional (AIDR 3D, Canon Medical Systems, Otawara, Japan) and four different parameters of AiCE (body, body sharp, brain CTA, and brain LCD).The reconstruction kernel with AIDR 3D was FC05 (soft-tissue kernel).

Phantom study
The low-contrast object specific contrast-to-noise ratio (CNR LO ) was calculated from CT images of the phantom to evaluate the objective image quality for the detectability of tendons.CNR LO is a quantitative evaluation index that reflects the spatial frequency characteristics of image noise and the measurement object frequency components and contrast [15].CNR LO was obtained using the following ROI T and ROI B were analyzed with the free software package ImageJ (National Institutes of Health, Bethesda, MD, USA) [16].Circular ROIs were placed on the extensor tendon-simulating rod and background to measure the CT value of ROI T and ROI B , respectively (Fig. 2).NPS was analyzed with a radial frequency method using CTmeasure Ver.0.98f (Japanese Society of CT Technology, Hiroshima, Japan) [17].Two square ROIs were placed on the background to measure the NPS, and an average NPS was calculated for each tube current and reconstruction parameter (Fig. 2).
The spatial frequency (ū) was calculated using the following formulas: ū indicates the most contributing spatial frequency for detectability corresponding to the diameter of the extensor tendon-simulating rod, J 1 () is a first-order Bessel function of the first order, d is the extensor tendon-simulating rod diameter, and u indicates the frequency.Using Eq. 2 and Eq. 3, the value of ū was 0.16 in this study.

Clinical study
Quantitative analysis Objective image quality for the detectability of tendons in clinical study was assessed using the contrast-to-noise ratio (CNR).The CNR was calculated using the following formula: ROI tendon and ROI muscle indicate the CT values measured in the extensor tendon and muscle ROI, respectively, and SD muscle indicates the standard deviation (SD) value at ROImuscle (Fig. 3).Furthermore, the coefficient of variation (CV) was calculated as the SD of CNR divided by the mean of the CNR in all patients.
All clinical images were analyzed using the free software package ImageJ (National Institutes of Health, Bethesda, MD, USA) for the comparison of reconstruction parameters [16].
Qualitative analysis Three radiology technologists (14 ± 4.9 years of experience, they were qualified as Japanese certifying organization of X-ray CT technologists for radiological technologists) were included in the subjective analysis.All 3D images were acquired using a 3D workstation (ziosta-tion2; Ziosoft Inc., Tokyo, Japan) and processed only with the same color presets on the 3D workstation.The extensor tendon 3D images were assessed for image noise and contrast between extensor tendons and muscles using a five- Statistical analysis Calculated values are presented as mean ± SD.The Friedman test, followed by Bonferroniadjusted Wilcoxon signed-rank test, was used to compare the tube currents (50, 100, 150, 200, and 250 mA) and the reconstruction methods (AIDR 3D, AiCE body, AiCE body sharp, AiCE brain CTA, and AiCE brain LCD).All statistical analysis were performed using the software package easy R (EZR) [18].A P value of < 0.05 was considered statistically significant.

Phantom study
Figure 4 shows the results of CNR LO for AIDR 3D and AiCE (body, body sharp, brain CTA, and brain LCD) at different tube currents.The CNR LO values for all reconstruction parameters improved as the tube current increased.The CNR LO value of AiCE brain CTA was higher than other reconstruction parameters at tube currents above 150 mA.Table 1 shows the results of NPS(ū) and the difference value between ROI T and ROI B used in the CNR LO analysis.The NPS(ū) values varied with the tube current for all reconstruction parameters.Among them, NPS(ū) of AiCE body sharp showed a slight change and a smaller value compared to other parameters.Furthermore, AiCE body sharp, at low tube currents (≤ 150 mA), showed lower differences in CT values between ROI T and ROI B compared to other reconstruction parameters.

Clinical study
Figures 5 and 6 show the results of CNR and CV for AIDR 3D and AiCE (body, body sharp, brain CTA, and brain LCD).The CNR value of AiCE brain CTA was higher than   other reconstruction parameters (P < 0.05).The CV value of AiCE brain CTA was lower than other reconstruction parameters.Table 2 shows the results of SD muscle and the difference value between ROI tendon and ROI muscle used in the CNR analysis.Both AiCE body and brain CTA showed low SD muscle values.However, in terms of difference in the CT value between ROI tendon and ROI muscle , AiCE body showed a lower value, whereas AiCE brain CTA showed a higher value.
Table 3 shows the results of qualitative analysis.The mean score of image noise for AIDR 3D and AiCE (body, body sharp, brain CTA, and brain LCD) were 2.0, 3.2, 2.3, 3.4, and 3.1, respectively.The mean score of contrast between the extensor tendons and muscles was significantly higher in AiCE brain CTA compared to other reconstruction parameters.Figure 7 shows an example of image noise and contrast between the extensor tendons and muscles using AIDR 3D and AiCE (body, body sharp, brain CTA, and brain LCD).Differences in the depiction of the extensor tendons were observed in 3D images of each reconstruction parameter.In particular, the 3D image of AiCE brain CTA allowed for clear identification of the extensor tendons and the raptured tendon stump.On the other hand, the 3D image of AiCE brain LCD depicted not only the extensor tendons but also the muscles and soft tissues.As a result, on the 3D image of AiCE brain LCD, the depicted soft tissues prevented the identification of the raptured tendon stump.

Discussion
This study evaluated the optimal reconstruction parameters and the influence of tube current on extensor tendons 3D CT using AiCE reconstruction.In the phantom study, AiCE brain CTA showed higher CNR LO values compared to other reconstruction parameters.Additionally, the CNR LO values for all reconstruction parameters decreased as the tube current decreased.We observed that the reduction in CNR LO due to low radiation doses could not be completely mitigated even with AiCE reconstruction.These findings indicate that extensor tendons 3D CT using AiCE without dose reduction may effectively enhance the visualization of extensor tendons.In the clinical study, the CNR value of AiCE brain CTA was higher than that of other reconstruction parameters.The findings from both the phantom and clinical studies demonstrate that AiCE brain CTA is particularly suitable for the depiction of extensor tendons in 3D CT.
In our phantom study, CNR LO of all AiCE parameters showed higher values than those of AIDR 3D at all radiation doses.Various authors have indicated that AiCE can improve image quality in chest, cardiac, and abdomen examinations compared to AIDR 3D [9][10][11].Similarly, we considered that AiCE was also effective in improving image quality in tendon examinations.
Focusing on the radiation dose, both AiCE brain CTA at a CTDI VOL of 7.8 mGy and AIDR 3D at a CTDI VOL of 10.5 mGy showed similar CNR LO values.These findings are consistent with the study by Tamura et al. on DLR at low-dose, which supports the results of our study [12].However, it is essential to note that while AiCE at lowdose was able to attain CNR LO values comparable to conventional reconstruction methods, the depiction of extensor tendons with similar image quality as conventional reconstruction methods remained often challenging.On the contrary, when AiCE was used without dose reduction, it demonstrated the potential to improve CNR LO compared to AIDR 3D.This indicates the possibility of enhancing the 3D depiction of extensor tendons with the radiation dose-maintained AiCE.Furthermore, we propose that    AiCE without dose reduction may lead to improved extensor tendon identification and offer time-saving benefits in 3D image creation.Focusing on the reconstruction parameters of AiCE, CNR LO showed the most improvement with AiCE body sharp at lower radiation doses (≤ 100 mA) and with AiCE brain CTA at higher radiation doses (≥ 150 mA).AiCE body sharp demonstrated effective noise reduction at low-dose due to its low NPS(ū) values even at low radiation doses.However, AiCE body sharp showed smaller differences in CT values between tendons and muscles at radiation dose of 150 mA or lower compared to other reconstruction parameters.Therefore, when using AiCE body sharp, which is a soft tissue parameter recommended by the CT manufacturer, it is essential to carefully observe the contrast between tendons and muscles in the extensor tendons 3D CT images.On the other hand, AiCE brain CTA showed higher CNR LO values, even at 150 mA or lower, compared to other reconstruction parameters, except for AiCE body sharp.Additionally, AiCE brain CTA demonstrated superior NPS(ū) values and differences in CT values between tendons and muscles under various radiation dose settings.These results suggest that the choice of AiCE parameters plays a critical role in acquiring optimal image quality under various radiation dose settings.
In our clinical study, we compared CNR, CV of CNR, and 3D image scores between different reconstruction parameters.The SD muscle value and the difference in CT value between tendons and muscles using AiCE body sharp were both small, which aligned with the findings from our phantom study.Thus, AiCE body sharp, a soft tissue parameter recommended by the CT manufacturer, may not be suitable for extensor tendons 3D CT due to the degraded contrast between tendons and muscles.In the 3D image scores of the contrast between tendons and muscles, AiCE brain LCD demonstrated the smallest score.Consistently, the extensor tendon 3D CT images reconstructed with AiCE brain LCD in the example case showed an increase in CT value of the extensor tendons, as well as surrounding muscles and soft tissues.As a consequence, the visualization of extensor tendons degraded due to the overall increase in CT values of both the extensor tendons and the surrounding soft tissues.AiCE brain CTA demonstrated superior CNR, the difference in CT value between tendons and muscles, and CV of CNR values compared to other reconstruction parameters.Particularly, the result of the CV values is crucial for extensor tendon 3D CT, which is sensitive to slight differences in CT values between tendons and muscles, and it helps minimize the impact of CT scanning variations and individual differences among patients.Therefore, we considered that AiCE brain CTA can provide a stable 3D image quality of extensor tendons.Furthermore, the 3D image scores significantly favored AiCE brain CTA and were consistent with the findings from quantitative analyses.Based on the findings from our phantom and clinical studies, AiCE brain CTA is suitable for extensor tendons 3D CT.
This study has limitations.First, we utilized a single CT scanner at one institution.Comparing different vendorspecific reconstruction methods could be of interest, as there is a possibility of variations in the reconstruction process across different CT vendors.Additionally, the DLR technology is subject to regular updates.Therefore, further investigations are required to evaluate the impact of updated DLR algorithms.Second, the study population was small, and the true status and location of tendons in our retrospective study could not be definitively determined.Therefore, our study lacked a verification standard, such as echography and magnetic resonance imaging, to compare the accuracy of the reconstruction parameters.

Conclusion
In conclusion, the AiCE technique is useful for extensor tendons 3D CT.In the 3D images of the extensor tendons using AiCE, reducing the dose leads to degradation of image quality.Therefore, a strategy of improving image quality without dose reduction using AiCE may be beneficial and could potentially enhance the visualization of the extensor tendons.Additionally, our study suggests that the AiCE brain CTA may be more suitable for extensor tendons 3D CT compared to other AiCE parameters.

Fig. 1
Fig. 1 Illustration of the phantom with simulated extensor tendon (3 mm).The CT value difference between the simulated extensor tendon and the background was 20 Hounsfield units (HU)

Fig. 2 Fig. 3
Fig. 2 The experimental setup diagram of ROIs in phantom study.ROI T and ROI B indicate CT values at the extensor tendon-simulating rod and background, respectively.Square ROIs were used to measure NPS

Fig. 4
Fig. 4 Graph of CNR LO with AiCE (body, body sharp, brain LCD, and brain CTA) and AIDR 3D

Fig. 6
Fig. 6 Graph of CV with AiCE (body, body sharp, brain CTA, and brain LCD) and AIDR 3D

Fig. 7
Fig. 7 The 3D images (A) and axial images (B) of extensor tendons using (a) AIDR 3D, (b) AiCE body, (c) AiCE body sharp, (d) AiCE brain CTA, and (e) AiCE brain LCD.This case was a 49-year-old man with an extensor pollicis longus (EPL) tendon rupture.In AiCE brain CTA, the contour of the extensor tendons could be identified

Table 1 NPS
(ū) values and difference in the CT values between ROI T and ROI B used in the CNR LO analysis CTDI VOL volume computed tomography dose indexes, NPS noise power spectrum, ROI regions of interest, AIDR 3D adaptive iterative dose reduction three dimensional CTDI VOL (mGy) NPS(ū) Difference in CT values between ROI T and ROI B AIDR 3D Body Body Sharp Brain CTA Brain LCD AIDR 3D Body Body Sharp Brain CTA Brain LCD

Table 2
SD muscle values and difference in the CT values between ROI tendon and ROI muscle used in the CNR analysis SD standard deviation, ROI regions of interest, AIDR 3D adaptive iterative dose reduction three dimensional

Table 3
Qualitative analysis in clinical study