Hiroyuki Nagata1, Yoshiharu Ohno1,2, Takeshi Yoshikawa2,3, Kaori Yamamoto4, Masato Ikedo4, Masao Yui4, Maiko Shinohara4, Akiyoshi Iwase5, Takahiro Matsuyama2, Takahiro Ueda2, Hirotaka Ikeda2, Yoshiyuki Ozawa2, and Hiroshi Toyama2
1Joint Research Laboratory of Advanced Medical Imaging, Fujita Health University School of Medicine, Toyoake, Japan, 2Radiology, Fujita Health University School of Medicine, Toyoake, Japan, 3Diagnostic Radiology, Hyogo Cancer Center, Akashi, Japan, 4Canon Medical Systems Corporation, Otawara, Japan, 5Fujita Health University Hospital, Toyoake, Japan
Synopsis
Keywords: Liver, Cancer
We hypothesize that CS with DLR can improve spatial resolution, image
quality and tumor detection on Gd-EOB-enhanced 3D T1WI in suspected liver tumor
patients. The purpose of this study was
to determine the utility of CS and DLR for image quality and liver tumor
detection improvements on high-resolution 3D CE-T1WI (HR-CE-T1WI) as compared
with conventional 3D CE-T1WI with PI (conventional CE-T1WI) in patients with
suspected liver tumors.
Introduction
Contrast-enhanced (CE-) MRI with gadolinium ethoxybenzyl
diethlenetriamine pentaacetic acid (Gd-EOB-DTPA) is currently one of the
essential sequences and applied for liver tumor detection and evaluation in
routine clinical practice. For
Gd-EOB-DTPA enhanced T1-weighted imaging (T1WI) has been frequently obtained by
means of 3D T1WI with conventional parallel imaging with or without breath
holding. Since 2000s, conventional
parallel imaging (PI) such as sensitivity encoding, etc. has been widely
applied for improving temporal and spatial resolutions in routine MR imaging. In the last several years, compressed sensing
(CS) with or without PI has been introduced by major MR vendors and applied in
routine clinical practice (1-3). Moreover,
deep learning reconstruction (DLR) has also been tested or applied to not only
MRI, but also CT and suggested as useful for improving image quality for
various clinical aims (4-7). However, CS
has been mainly applied for improving temporal resolution for dynamic CE-MRI or
reducing examination time in various organs.
Moreover, DLR is applied for image noise reduction on MRI or CT, and
little influence to diagnostic performance on MRI, except prostate DWI (5). Furthermore, no one has been assessed the
utility of CS with DLR for high-resolution 3D CE-T1WI with Gd-EOB-DTPA to
enhance liver tumor detection capability in patients with suspected liver
tumors. We hypothesize that CS with DLR
can improve spatial resolution, image quality and tumor detection on
Gd-EOB-enhanced 3D T1WI in suspected liver tumor patients. The purpose of this study was to determine
the utility of CS and DLR for image quality and liver tumor detection
improvements on high-resolution 3D CE-T1WI (HR-CE-T1WI) as compared with
conventional 3D CE-T1WI with PI (conventional CE-T1WI) in patients with suspected
liver tumors.Materials and Methods
Seventy-seven
patients (55 men, 22 women, mean age 63 years) with suspected liver tumor
underwent dynamic CE-CT, 3D Gd-EOB-DTPA-enhanced T1WI at 3T MR systems (Vantage
Centurian, Canon Medical Systems Corporation, Otawara, Japan) by conventional
3D CE-T1WI and HR-CE-T1WI with CS and reconstructed with DLR, surgical or
interventional treatments, pathological examinations or more than 2 years
follow-up examinations. In each patient,
HR-CE-T1WI was obtained by static 3D T1-weighted fast gradient echo (FFE)
sequence with CS and reconstructed with 1.5mm contiguous slice thickness by DLR
method. On the other hand, conventional
CE-T1WI was obtained by same sequence with conventional parallel imaging and
reconstructed as 3mm contiguous slice thickness without DLR. The breath-holding time of each CE-T1WI was
28sec. Then, signal-to-noise ratio (SNR)
of liver and contrast-to-noise ratio (CNR) between liver and tumor or cyst on
each 3D CE-T1WI were assessed by ROI measurements. The probability of tumor on hepatobiliary
phase was assessed on each CE-T1WI with 5-point scoring system by a
board-certified abdominal radiologist with 28 years experiences and a
board-certified chest radiologist with more than 28 years experiences of liver
MRI. To determine the utility of CS with
PI and DLR for improving image quality on each 3D CE-T1WI, SNRs and CNRs were
compared between two methods by paired t-test.
On comparison of tumor detection capability between two methods,
Jackknife free-response receiver operating characteristic (JAFROC) analysis was
performed to compare malignant tumor detection capabilities between two
methods. Finally, figure of merit (FOM) values, sensitivity (SE) and
false-positive rate/data set (FPR) for each radiologist and consensus
assessment were also compared between two methods by using McNemar’s test or
the signed rank test.Results
Representative cases are shown in Figures 1. Comparisons of SNR and CNRs between
conventional CE-T1WI and HR-CE-T1WI are shown in Figure 2. SNR of HR-CE-T1WI was significantly higher
than that of conventional CE-T1WI (p=0.002), although there were no significant
differences of CNRs between tumor or cyst and liver (p>0.05). Comparisons of JAFROC analysis results and
detection capability between conventional CE-T1WI and HR-CE-T1WI are shown in
Figure 3. SEs or FPRs of HR-T1WI by
consensus (SE: 0.90, FPR: 0.27/data set), reader 1 (SE: 0.92, FPR: 0.27/data
set) and reader 2 (SE: 0.94) were significantly better than those of conventional
CE-T1WI by consensus (SE: 0.84, p=0.004; FPR: 0.34, p=0.04), reader 1 (SE:
0.87, p=0.008; FPR: 0.34/data set, p=0.04) and reader 2 (SE: 0.89, p=0.02),
although there were no significant difference of FOM between two methods at
consensus (p=0.07), reader 1 (p=0.14) and reader 2 (p=0.10). Conclusion
Compressed sensing and deep learning reconstruction has a potential to
improve image quality and liver tumor detection improvements on HR-CE-T1WI as
compared with conventional 3D CE-T1WI in patients with suspected liver tumors.Acknowledgements
This study was technically and financially supported by Canon Medical Systems Corporation. References
- Feng L, Benkert T, Block KT,
Sodickson DK, Otazo R, Chandarana H. Compressed sensing for body MRI. J Magn
Reson Imaging. 2017;45(4):966-987.
- Ikeda H, Ohno Y, Murayama K, et
al. Compressed sensing and parallel imaging accelerated T2 FSE sequence for
head and neck MR imaging: Comparison of its utility in routine clinical
practice. Eur J Radiol. 2021;135:109501.
- Choi MH, Kim B, Han D, Lee YJ.
Compressed sensing for breath-hold high-resolution hepatobiliary phase imaging:
image noise, artifact, biliary anatomy evaluation, and focal lesion detection
in comparison with parallel imaging. Abdom Radiol (NY). 2022;47(1):133-142.
- Ueda T, Ohno Y, Yamamoto K, et
al. Compressed sensing and deep learning reconstruction for women's pelvic MRI
denoising: Utility for improving image quality and examination time in routine
clinical practice. Eur J Radiol. 2021;134:109430.
- Ueda T, Ohno Y, Yamamoto K, et
al. Deep Learning Reconstruction of Diffusion-weighted MRI Improves Image
Quality for Prostatic Imaging. Radiology. 2022;303(2):373-381.
- Obama Y, Ohno Y, Yamamoto K, et
al. MR imaging for shoulder diseases: Effect of compressed sensing and deep
learning reconstruction on examination time and imaging quality compared with
that of parallel imaging. Magn Reson Imaging. 2022;94:56-63.
- Matsuyama T, Ohno Y, Yamamoto
K, et al. Comparison of utility of deep learning reconstruction on 3D MRCPs
obtained with three different k-space data acquisitions in patients with IPMN.
Eur Radiol. 2022;32(10):6658-6667.