Comparing respiratory-triggered T2WI MRI with an artificial intelligence-assisted technique and motion-suppressed respiratory-triggered T2WI in abdominal imaging
Original Article

Comparing respiratory-triggered T2WI MRI with an artificial intelligence-assisted technique and motion-suppressed respiratory-triggered T2WI in abdominal imaging

Nan Wang1, Yuhui Liu2, Jiangnan Ran1, Qi An1, Lihua Chen1, Ying Zhao1, Dan Yu3, Ailian Liu1, Lina Zhuang1, Qingwei Song1

1Department of Radiology, The First Affiliated Hospital of Dalian Medical University, Dalian, China; 2School of Medical Imaging, Dalian Medical University, Dalian, China; 3United Imaging Research Institute of Intelligent Imaging, United Imaging Healthcare Technology, Beijing, China

Contributions: (I) Conception and design: N Wang, Q Song; (II) Administrative support: A Liu, L Zhuang; (III) Provision of study materials or patients: Y Liu, J Ran; (IV) Collection and assembly of data: L Chen, Y Zhao; (V) Data analysis and interpretation: Q An, D Yu; (VI) Manuscript writing: All authors; (VII) Final approval of manuscript: All authors.

Correspondence to: Qingwei Song, MD; Lina Zhuang, MD. Professor of Radiological Medicine, Department of Radiology, The First Affiliated Hospital of Dalian Medical University, Xigang District, Zhongshan Road, No. 222, Dalian 116011, China. Email: songqw1964@163.com.

Background: Magnetic resonance imaging (MRI) plays a crucial role in the diagnosis of abdominal conditions. A comprehensive assessment, especially of the liver, requires multi-planar T2-weighted sequences. To mitigate the effect of respiratory motion on image quality, the combination of acquisition and reconstruction with motion suppression (ARMS) and respiratory triggering (RT) is commonly employed. While this method maintains image quality, it does so at the expense of longer acquisition times. We evaluated the effectiveness of free-breathing, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging (ACS-RT T2WI) compared to conventional acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging (ARMS-RT T2WI) in abdominal MRI, assessing both qualitative and quantitative measures of image quality and lesion detection.

Methods: In this retrospective study, 334 patients with upper abdominal discomfort were examined on a 3.0T MRI system. Each patient underwent both ARMS-RT T2WI and ACS-RT T2WI. Image quality was analyzed by two independent readers using a five-point Likert scale. The quantitative measurements included the signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), peak signal-to-noise ratio (PSNR), and sharpness. Lesion detection rates and contrast ratios (CRs) were also evaluated for liver, biliary system, and pancreatic lesions.

Results: There ACS-RT T2WI protocol had a significantly reduced median scanning time compared to the ARMS-RT T2WI protocol (148.22±38.37 vs. 13.86±1.72 seconds). However, ARMS-RT T2WI had a higher PSNR than ACS-RT T2WI (39.87±2.72 vs. 38.69±3.00, P<0.05). Of the 201 liver lesions, ARMS-RT T2WI detected 193 (96.0%) and ACS-RT T2WI detected 192 (95.5%) (P=0.787). Of the 97 biliary system lesions, ARMS-RT T2WI detected 92 (94.8%) and ACS-RT T2WI detected 94 (96.9%) (P=0.721). Of the 110 pancreatic lesions, ARMS-RT T2WI detected 102 (92.7%) and ACS-RT T2WI detected 104 (94.5%) (P=0.784). The CR analysis showed the superior performance of ACS-RT T2WI in certain lesion types (hemangioma, 0.58±0.11 vs. 0.55±0.12; biliary tumor, 0.47±0.09 vs. 0.38±0.09; pancreatic cystic lesions, 0.59±0.12 vs. 0.48±0.14; pancreatic cancer, 0.48±0.18 vs. 0.43±0.17), but no significant difference was found in others like focal nodular hyperplasia (FNH), hepatapostema, hepatocellular carcinoma (HCC), cholangiocarcinoma, metastatic tumors, and biliary calculus.

Conclusions: ACS-RT T2WI ensures clinical reliability with a substantial scan time reduction (>80%). Despite minor losses in detail and SNR reduction, ACS-RT T2WI does not impair lesion detection, marking its efficacy in abdominal imaging.

Keywords: Artificial intelligence; compressed sensing; magnetic resonance imaging (MRI); lesion assessment

Submitted Jan 08, 2025. Accepted for publication Jun 10, 2025. Published online Aug 19, 2025.

doi: 10.21037/qims-2025-71

Introduction

Abdominal diseases significantly affect human health, and their effective treatment hinges on accurate lesion detection. Magnetic resonance imaging (MRI) plays a crucial role in the diagnosis of abdominal conditions, employing a standard protocol that includes T2-weighted, diffusion-weighted, and dynamic contrast-enhanced sequences (1). However, a comprehensive assessment, especially of the liver, requires multi-planar T2-weighted sequences to detect and characterize focal liver lesions (2).

The scanning difficulty of abdominal MRI arises from respiratory motion. To mitigate the effect of respiratory motion on image quality, the combination of acquisition and reconstruction with motion suppression (ARMS) and respiratory triggering (RT) is commonly employed (3). This technology uses a radial k-space filling method that is insensitive to motion, employing a rotational signal acquisition approach. Each acquisition captures the central data of k-space. Multiple collections are conducted on the same plane, using overlapping common information to eliminate the effects of motion during each acquisition. The ARMS technique can be applied to various body parts, allowing for imaging in multiple orientations and weights, significantly reducing both susceptibility artifacts and motion artifacts in the images. While this method maintains image quality, it does so at the expense of longer acquisition times.

In addition, respiratory motion artifacts can be reduced by the accelerating method. Recent advancements have seen the adoption of the half Fourier (HF), parallel acquisition technique (PAT), and compressed-sensing (CS) methods in MRI to expedite the scanning process (4). The integration of artificial intelligence (AI), notably including deep neural networks, in reconstructing under-sampled k-space data has been a breakthrough, allowing for faster MRI data acquisition without sacrificing diagnostic image quality (5). Artificial intelligence-assisted compressed-sensing (ACS) technology, a novel approach in k-space data sampling, is increasingly being applied across various anatomical regions and sequences (6-8). Meanwhile, practical strategies like a single-breath-hold (SBH) or multi-breath-hold (MBH) can shorten scan times but may be challenging for some patients (9). Despite these advancements, certain patient groups face challenges with breath-hold scanning techniques, including those with reduced vital capacity post lung surgery, hearing-impaired individuals who cannot follow breath-holding instructions, and patients too weak to hold their breath due to illness.

This study aimed to qualitatively and quantitatively evaluate the efficacy of artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging (ACS-RT T2WI) in abdominal MRI in comparison to the conventional acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging (ARMS-RT T2WI) to address the needs of these patient groups. We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-71/rc).

Methods

Study sample

This was an exploratory, single-center, retrospective study. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Board of The First Affiliated Hospital of Dalian Medical University (approval number: YJ-KS-KY-2024-631), and the requirement of individual consent for this retrospective analysis was waived.

The imaging data of patients presenting with upper abdominal discomfort between June 20 and November 30, 2023 were collected. Informed consent was obtained in writing from all patients before the MRI procedures. The study included patients without general MRI contraindications or severe claustrophobia. Patients were excluded from the study if they met any of the following exclusion criteria: were unable to complete the examination, had chronic liver disease without occupying lesions, and/or had undergone upper abdominal surgery. In total, 334 patients were enrolled in the study, resulting in a dataset of 408 lesions, of which 294 lesions measured over 1 cm in diameter. The patient selection process is illustrated in Figure 1.

Figure 1 Study flowchart. MRI, magnetic resonance imaging; SNR, signal-to-noise ratio; CNR, contrast-to-noise ratio; PSNR, peak signal-to-noise ratio; CR, contrast ratio.

MRI protocol

The MRI scans were conducted on the 3.0 T uMR Omega (United Imaging Healthcare, Shanghai, China) with an anterior phased-array coil. The patients were positioned supine with their head first and arms at their sides for enhanced comfort. The protocol included the following two T2-weighted sequences: (I) ARMS-RT T2WI (repetition time (TR): 3,500 ms; echo time (TE): 84.16 ms; field of view (FOV): 38 cm × 38 cm; echo train length: 31; flip angle: 130°; acquisition voxel size: 1.25 mm; reconstruction voxel size: 0.83 mm; sense 3; scan time: 2 min 38 s); and (II) ACS-RT T2WI (TR: 4,190 ms; TE: 81.37 ms; FOV: 38 cm × 30 cm; echo train length: 76; flip angle: 110°; acquisition voxel size: 1.25 mm; reconstruction voxel size: 0.83 mm; ACS 3; scan time: 33.5 s). Both sequences acquired 24 slices per volume (slice thickness 6 mm + gap 1.2 mm), covering a total craniocaudal length of 171.6 mm.

ACS

ACS uses convolutional neural networks (CNNs) to enhance image acquisition speed. Traditional CNN applications in clinical settings face challenges due to their “black-box” unpredictability. ACS addresses this by integrating AI outputs as supplementary constraints in the CS framework, incorporating a regularization term to reconcile differences between traditionally reconstructed images and those generated by AI.

The ACS neural networks are trained on an extensive dataset of two million fully sampled images, predominantly from human volunteers (98%) and phantoms (2%). This training enables the networks to refine the image reconstruction process, incorporating k-space architectural design and multiscale sparsification. This approach synergizes the CS, HF, and PAT, ensuring that ACS output closely matches the gold standard of fully-sampled images as validated by simulations. For details on the architecture of the deep learning (DL)-based magnetic resonance reconstruction framework used in this study, see (10).

Image analysis

The image analysis was conducted using uWS MR version R005 software. To maintain objectivity, the radiologists reviewed the images without prior knowledge of the imaging parameters or patient information. Scanning durations for ARMS-RT T2WI and ACS-RT T2WI were recorded to highlight the efficiency gains from ACS.

Qualitative analysis

Two experienced radiologists, with eight and four years of experience in abdominal imaging, respectively, conducted image quality assessments to minimize recall bias and ensure methodological blindness. All the images were randomized and displayed simultaneously as images A and B, with adjustments made to window widths, contrasts, and levels for each sequence to prevent recall bias and uphold methodological blindness. Five critical aspects (i.e., respiratory motion artifacts, image interlacing, liver fine structure, gastrointestinal tract visibility, and diagnostic confidence) were evaluated using a five-point Likert scale on which 1 represented non-diagnostic; 2 represented poor; 3 represented acceptable; 4 represented good; and 5 represented excellent. The average score from both radiologists was used unless discrepancies arose; in which case, a senior radiologist was consulted. The total subjective evaluation score for image quality was then calculated.

Quantitative analysis (normal group)

Elliptical regions of interest (ROIs) ranging from 90 to 150 mm2 were precisely positioned within the largest layers of the four liver lobes (left medial, left lateral, right anterior, and right posterior). Circular ROIs, varying between 4 and 12 mm2, targeted the proximal, middle, and distal ends of the common bile duct, while ROIs of 15 to 30 mm2 were allocated to the head, body, and tail of the pancreas. The average signal intensity (SI) values from these ROIs were meticulously recorded for the liver, bile duct, and pancreas. Additionally, circular ROIs placed on the right erector spinae at the first hepatic portal level facilitated the measurement of the average SI and standard deviation (SD), enabling the calculation of the SNR and contrast-to-noise ratio (CNR) for each organ using the following equations:

SNRliver=SIliverSDerector spinae

SNRcommon bile duct=SIcommon bile ductSDerector spinae

SNRpancreas=SIpancreasSDerector spinae

CNRliver=SIliver−SIerector spinaeSDerector spinae

CNRcommon bile duct=SIcommon bile duct−SIerector spinaeSDerector spinae

CNRpancreas=SIpancreas−SIerector spinaeSDerector spinae

In this study, the adaptive peak signal-to-noise ratio (PSNR) was introduced as a sophisticated metric designed to evaluate the intrinsic quality of the images. This metric extends beyond the conventional signal-to-noise ratio (SNR) by incorporating a nuanced approach to noise estimation and signal assessment. Unlike the SNR, which primarily measures the ratio of overall signal power to noise power, the adaptive PSNR focuses on the specific attributes of noise and its relationship to the signal in an image. The noise estimation process, critical for calculating the adaptive PSNR, leverages methods such as the one described by Immerkär (11), which uses a Laplacian operator to isolate noise components by emphasizing high-frequency areas indifferent to the image’s structural content. The adaptive PSNR is computed using the following formula: 20log10(MAX1Noise level), where MAX1 represents the maximum pixel intensity of the image, and Noise level signifies the estimated noise within the image. This formula is similar to the traditional PSNR calculation but is adapted to assess the noise in a single image rather than comparing it to a reference image.

Image sharpness was quantified in the Fourier domain by measuring the relative amount of high-frequency content. After computing the centered two-dimensional Fourier transform of each image, we identified all the frequency components whose magnitude exceeded 0.05% of the maximum spectral amplitude; the sharpness score was defined as the fraction of these “high-frequency” coefficients relative to the total number of coefficients. In this way, the images with more pronounced edges and fine detail (which manifested as greater energy in the high-frequency bands) received higher sharpness values, enabling objective comparison across reconstruction methods.

Lesion detection

MRI diagnostic reports (dual-echo T1 quick3D, T2-weighted imaging [T2WI], diffusion weighted imaging, and T1 quick3D water fat imaging contrast enhancements) were used as the standard reference for detecting lesions in this study. Two independent radiologists, with eight and four years of experience in abdominal imaging, respectively, reviewed the images. They had access to both the T2-weighted images and additional sequences as required (dual-echo T1 quick3d and diffusion-weighted imaging). Each radiologist documented the lesions identified without prior exposure to the patients’ clinical data or knowledge of any contrast enhancements used during scanning. Discrepancies between the radiologists were resolved through a consensus discussion, or if necessary, by consulting a third, more senior radiologist.

Quantitative analysis (lesion diameter ≥1 cm)

The ROIs were accurately positioned at the most significant cross-section of each lesion and its adjacent areas. Three circular ROIs (20–260 mm2) were placed within the focal lesions, ensuring the exclusion of large vessels, areas of necrosis, or hemorrhage. Similarly, for the biliary system and pancreas, the ROIs were carefully drawn to encompass the entirety of the lesions while avoiding non-representative regions. Additionally, background ROIs were set at comparable anatomical levels within the same segments of the liver (90–150 mm2), biliary system (4–12 mm2), and pancreas (15–30 mm2) to normalize the SI measurements against the background tissue noise. The average SI for each set of ROIs was calculated. The contrast ratio (CR) was then determined using the following equations:

CRliver=SIlesion−SIliverSIlesion+SIliver

CRbiliary system=SIbiliary duct−SIlesionSIbiliary duct+SIlesion

CRpancreas=SIlesion−SIpancreasSIlesion+SIpancreas

Statistical analysis

The statistical analysis was conducted using SPSS software (version 22.0, IBM). The intraclass correlation coefficient (ICC) was used to assess the consistency in the measurements and subjective evaluations, with an ICC greater than 0.75 indicating strong agreement. The Kolmogorov-Smirnov test was used to determine data distribution normality. To compare image quality differences between the ARMS-RT T2WI and ACS-RT T2WI protocols, both the paired t-test (for normally distributed data) and the Wilcoxon signed-rank test (for non-normally distributed data) were applied. The McNemar’s test was used to evaluate variations in the lesion detection rates. The significance level was set at P<0.05 for all tests, marking statistical significance.

Results

Patient characteristics

This study included 334 patients, with an age range of 19 to 89 years (mean ± SD: 57.3±13.8 years), in whom, a total of 408 lesions were identified. The lesions were classified based on final image diagnoses (according to the clinical routine diagnosis sequence), and included 201 liver focal lesions, highlighting various conditions such as hemangiomas, liver abscess, and hepatocellular carcinoma (HCC). Additionally, 97 biliary system and 110 pancreatic focal lesions were documented, each with specific case counts for conditions like calculus, cancer, and cystic focuses. These findings are detailed in Table 1 and illustrated in the patient flowchart (Figure 1).

Table 1

Patient and lesion characteristics

Characteristics Value
Total number of patients 334
Gender
   Male 159
   Female 175
Age (years) 57.3±13.8 [19–89]
Lesions 408
Liver
   Hemangioma 90
   Abscess 15
   Focal nodular hyperplasia 9
   Hepatocellular carcinoma 31
   Metastatic tumor 48
   Cholangiocarcinoma 8
Biliary system
   Calculus 79
   Cancer 18
Pancreas
   Cystic focus 78
   Cancer 32

Data are presented as the number, mean ± standard deviation [range].

Scanning time

ACS-RT T2WI showed a marked reduction in the scanning time (13.86±1.72 seconds) compared to ARMS-RT T2WI (148.22±38.37 seconds), translating to an 82.4% to 96.6% improvement in efficiency.

Comparison of image quality

Inter-reader agreement on image quality was strong (ICC >0.75) (Table S1), leading to the selection of data from the more experienced reader for the detailed analysis. The quantitative analysis of 42 patients’ images showed no significant difference in the SNR and CNR between the two protocols (P>0.05, respectively). However, the PSNR was notably higher in ARMS-RT T2WI, while the ACS-RT T2WI excelled in sharpness. Subjective evaluations across all patients indicated superior liver fine structure visualization with ARMS-RT T2WI, but improved gastrointestinal tract conspicuity with ACS-RT T2WI. These results are set out in Table 2 and visualized in Figure 2. The examples corresponding to each Likert Scale are visualized in Figure S1.

Table 2

Comparison of image quality between ARMS-RT T2WI and ACS-RT T2WI

Variable ARMS-RT T2WI (n=42) ACS-RT T2WI (n=42) T P
SNRliver 25.31±10.14 23.40±9.38 –0.640* 0.522
SNRcommon bile duct 134.64±57.44 134.84±63.45 –0.097* 0.923
SNRpancreas 36.38±15.65 34.58±12.47 –0.012* 0.990
CNRliver –2.64±6.36 –4.13±7.85 –1.666* 0.096
CNRcommon bile duct 106.68 ±50.02 107.31±59.56 –0.266* 0.791
CNRpancreas 8.43±9.17 9.05±9.57 –0.700* 0.484
PSNR 39.87±2.72 38.69±3.00 3.687 0.001
Sharpness 0.27±0.02 0.36±0.03 –22.966 < 0.001
Respiratory motion artifacts 4.26±0.67 4.21±0.66 –1.526* 0.127
Image interlacing 4.13±0.70 4.16±0.73 –0.526* 0.599
Fine structure of liver 4.40±0.59 4.10±0.76 –7.127* < 0.001
Gastrointestinal tract 3.71±0.77 4.14±0.68 –7.805* < 0.001
Diagnostic confidence 4.18±0.67 4.13±0.71 –1.597* 0.110
Total score 20.69±1.53 20.75±1.65 –0.719* 0.472

Data are presented as the mean ± standard deviation. *, the data with a non-normal distribution. ARMS-RT T2WI, acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging; ACS-RT T2WI, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging; CNR, contrast-to-noise ratio; PSNR, peak signal-to-noise ratio; SNR, signal-to-noise ratio.

Figure 2 Stacked bar charts showing the qualitative image assessment stratified by sequence type. The liver fine structure scores of ARMS-RT T2WI were significantly higher than those of ACS-RT T2WI. The conspicuity of the gastrointestinal tract was significantly higher on ACS-RT T2WI than ARMS-RT T2WI. Most of the images were rated as “excellent” and “good”. ACS-RT T2WI, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging; ARMS-RT T2WI, acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging.

Comparison of lesion detection

The lesion detection rates were calculated according to the method detailed in the literature (12,13). The comparison of the lesion detection rates and CRs between the ARMS-RT T2WI and ACS-RT T2WI protocols across the 291 patients revealed no significant differences in the detection rates of liver, biliary system, and pancreatic lesions. Specifically, the detection rates for liver and biliary system lesions were over 95% for both protocols, while the pancreatic lesion detection rates were slightly lower but still closely matched (Figure 3).

Figure 3 Dot plots showing the quantitative evaluation metrics for each participant stratified by sequence type. In the analysis of hemangioma, biliary tumor, pancreatic cystic lesions, and pancreatic cancer, the CR values of ACS-RT T2WI were higher than those of ARMS-RT T2WI (P<0.05). ACS-RT T2WI, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging; ARMS-RT T2WI, acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging; CR, contrast ratio; FNH, focal nodular hyperplasia; HCC, hepatocellular carcinoma.

In the analysis of the 291 patients, the lesion detection rates for liver, biliary system, and pancreatic lesions were closely matched between ARMS-RT T2WI and ACS-RT T2WI, and no statistically significant difference between the two protocols was found (P>0.05). Specifically, ARMS-RT T2WI detected 193 liver lesions (96.0%) and ACS-RT T2WI detected 192 liver lesions (95.5%) (P=0.787). ARMS-RT T2WI detected a 4-mm micro metastasis, but ACS-RT T2WI did not (Figure 4). Of the 97 biliary system lesions, ARMS-RT T2WI detected 92 (94.8%) and ACS-RT T2WI detected 94 (96.9%) (P=0.721). The missed lesions were mall calculus in biliary system (lesion diameter <5 mm). Of the 110 pancreatic lesions, ARMS-RT T2WI detected 102 (92.7%) and ACS-RT T2WI detected 104 (94.5%) (P=0.784). The missed lesions were pancreatic cystic lesions (lesion diameter <5 mm). The disagreement in the two methods related to the layer deviation of the small lesions. The CR analysis of the 294 lesions revealed that ACS-RT T2WI had higher values for hemangioma, biliary tumor, pancreatic cystic lesions, and pancreatic cancer (P<0.05). However, no statistical difference in the CR values was found for focal nodular hyperplasia (FNH), hepatapostema, HCC, cholangiocarcinoma, metastatic tumor, and biliary calculus between the two protocols (P>0.05). These results are detailed in Table 3 and illustrated in Figures 3-7.

Figure 4 T2WI in a 76-year-old man with a metastatic liver tumor. ARMS-RT T2WI (A-C) and ACS-RT T2WI (D-F) had equally high image quality (total score: 23 vs. 21). ARMS-RT T2WI (B) showed a 4-mm micro metastasis (arrowheads), which ACS-RT T2WI (E) did not. ARMS-RT T2WI (C) and ACS-RT T2WI (F) showed a 8-mm metastasis (short arrows). ACS-RT T2WI had a 89.3% lower scanning time than ARMS-RT T2WI (14 vs. 130.9 s). ACS-RT T2WI, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging; ARMS-RT T2WI, acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging.

Table 3

Comparison of the CR values of the two sequences between different diseases

Variable ARMS-RT T2WI ACS-RT T2WI T P
CRhemangioma (n=90) 0.55±0.12 0.58±0.11 –2.957* 0.003
CRFNH (n=9) 0.36±0.16 0.39±0.20 –0.821 0.436
CRliver abscess (n=15) 0.34±0.16 0.32±0.14 0.391 0.707
CRHCC (n=31) 0.30±0.10 0.32±0.12 –1.563 0.130
CRcholangiocarcinoma (n=8) 0.46±0.07 0.39±0.10 2.226 0.061
CRmetastatic tumor (n=48) 0.40±0.14 0.40±0.15 –0.150 0.882
CRbiliary calculus (n=79) 0.67±0.16 0.70±0.13 –1.232 0.228
CRbiliary tumor (n=18) 0.38±0.09 0.47±0.09 –2.526 0.045
CRpancreatic cystic lesions (n=78) 0.48±0.14 0.59±0.12 –9.126 <0.001
CRpancreatic cancer (n=32) 0.43±0.17 0.48±0.18 –3.501* <0.001

Data are presented as the mean ± standard deviation. *, the data with a non-normal distribution; ARMS-RT T2WI, acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging; ACS-RT T2WI, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging; CR, contrast ratio; FNH, focal nodular hyperplasia; HCC, hepatocellular carcinoma.

Figure 5 T2WI in a 65-year-old woman with pancreatic cystic lesions. The patient breathed regularly and cooperated well. ARMS-RT T2WI (A) revealed more branches of the intrahepatic bile ducts (arrowheads) than ACS-RT T2WI (D). ACS-RT T2WI (E,F) increased the boundary sharpness of the lesion (B,E; short arrows) and the fine structure of lesion (C,F; long arrows) compared to ARMS-RT T2WI (B,C). ACS-RT T2WI had a 90.1% lower scanning time than ARMS-RT T2WI (13.2 vs. 132.8 s). ACS-RT T2WI, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging; ARMS-RT T2WI, acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging.
Figure 6 T2WI in a 65-year-old man with a biliary tumor, biliary obstruction, and liver metastatic tumor. The patient underwent biliary drainage. ARMS-RT T2WI (A-C) was a little coarse. ACS-RT T2WI (D-F) reduced the motion artifacts and increased the boundary sharpness of the lesion. ACS-RT T2WI (D) visualized the biliary tumor (arrowheads) better than ARMS-RT T2WI (A). ACS-RT T2WI (E) also visualized the biliary obstruction (short arrows) better than ARMS-RT T2WI (B). ACS-RT T2WI (F) and ARMS-RT T2WI (C) had equal value in the lesion detection of the liver metastatic tumor (long arrows; CR, 0.39 vs. 0.38). ACS-RT T2WI had a 92.5% lower scanning time than ARMS-RT T2WI (15.4 vs. 204.5 s). ACS-RT T2WI, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging; ARMS-RT T2WI, acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging; CR, contrast ratio.
Figure 7 T2WI in a 58-year-old woman with a right posterior lobe hepatapostema. ARMS-RT T2WI (A-C) and ACS-RT T2WI (D-F) had equally high image quality (total score: 22 vs. 22; contrast ratio, 0.31 vs. 0.32). The abscess wall, necrosis, and the inflammatory changes of the adjacent liver parenchyma in the two sequences were consistent. ACS-RT T2WI had a 90% lower scanning time than ARMS-RT T2WI (14 vs. 140 s). ACS-RT T2WI, artificial intelligence-assisted compressed-sensing respiratory-triggered T2-weighted imaging; ARMS-RT T2WI, acquisition and reconstruction with motion-suppression respiratory-triggered T2-weighted imaging.

Discussion

MRI has high sensitivity in detecting abdominal lesions of the liver, biliary system, and pancreas. T2WI is particularly adept at identifying necrosis, inflammation, liquid, and solid tumor components (14). T2WI also offers invaluable insights into the morphology and characteristics of lesions, beyond simple metrics like diameter and shape, including the potential invasion of adjacent structures by malignant lesions (15). To counter respiratory motion, RT and ARMS are used. However, these strategies often require prolonged acquisition times to satisfy the Nyquist criterion (16). As the volume of clinical examinations continues to increase, so too does the demand for accelerated imaging techniques that can deliver ultra-fast scans without sacrificing image quality, underscoring the urgency for innovation in this area.

The clinical application of traditional acceleration methods like HF, PAT, and CS faces limitations due to their restricted acceleration capabilities. The past decade has seen a significant surge in interest in AI, including machine learning and DL, due to its potential in healthcare (17). CNN-based ACS is emerging as a promising solution for accelerated imaging (18). Despite its gradual adoption into clinical practice, ACS requires comprehensive validation across various settings to ensure its effectiveness and applicability, similar to the validation process for CS. This involves extensive testing of DL models to fully understand their generalizability and limitations.

A study on the use of ACS in abdominal T2WI sequences underscored its clinical promise (19). Sheng et al.’s comparison of SBH-T2WI and MBH-T2WI revealed the superior image quality, lesion detectability, and contrast of SBH-T2WI, suggesting its potential to standardize liver MRI protocols (20). Similarly, Li et al.’s research indicated that SBH-T2WI using ACS significantly improved image quality and lesion detection while reducing scan times compared to conventional RT-T2WI (13). These findings collectively support the efficacy of ACS in diagnostic imaging, given its ability to both reduce acquisition time and maintain image quality.

While previous research has often emphasized breath-holding techniques in MRI scanning (21), many patients face challenges with such methods for various reasons. Embracing a free-breathing protocol could significantly enhance patient comfort and the success rate of scans (22). Our comparison of ACS-RT T2WI and ARMS-RT T2WI showed that both methods offer comparable image quality, but differ in specific technical metrics such as the PSNR and sharpness. For patients with regular breathing (Figure 5), ARMS combined with a breathing trigger can ensure enough K-space data and display a more complete structure. ACS-RT T2WI exhibited enhanced sharpness and potentially clearer imaging in areas prone to motion artifacts, while sustaining the SNR levels sufficiently for diagnostic evaluation.

Detailed quantitative analyses, including ROI-based calculations of the SNR and CNR, revealed no significant overall difference between the two techniques. However, ARMS-RT T2WI had a higher PSNR, indicating more uniform image clarity, while ACS-RT T2WI performed better in terms of sharpness, which can aid in delineating edges and may reduce certain types of motion artifacts. In some special cases (Figure 6), the low-signal “shade” on ACS-RT represents a flow-void artifact, which is attenuated in ARMS-RT by motion-weighted sampling. The observed strengths and weaknesses of each sequence thus highlight the nuanced balance between resolution, speed, and artifact suppression.

Moreover, the superior subjective scores for gastrointestinal tract visualization with ACS-RT T2WI appear to stem from its advanced k-space sampling strategy (23). ACS-RT T2WI employs Cartesian filling, which, combined with rapid acquisition, significantly mitigates motion artifacts. When the lesion size is close to the slice thickness, different respiratory conditions may lead to a slight difference in slice thickness that cannot be displayed. The structure of ACS-RT T2WI and ARMS-RT T2WI are not completely consistent. However, ACS-RT T2WI’s focus on broader structural clarity sometimes comes at the expense of the liver’s fine structure—a trade-off that still did not compromise overall lesion detection.

Individual missed lesions were small lesions (lesion diameter <5 mm). The reason for missed lesions is that observers mainly rely on T2WI (slice thickness: 6 mm) for lesion detection. However, this did not affect the analysis and comparison of the two sequences in this study. It will also not limit the clinical application of the two sequences, as the final diagnosis in the clinic needs to be based on the patient’s medical history, enhanced images, magnetic resonance cholangiopancreatography, and even thin-layer high-resolution T2WI. However, the existence of missed lesions suggests the necessity of multi-modal MRI examination. Notably, for liver, biliary system, and pancreatic focal lesions larger than 1 cm, ACS-RT T2WI provided lesion detection rates on par with ARMS-RT T2WI, and in some cases had comparable or even higher CRs.

This study had several limitations. First, the single-center design restricts the generalizability of our findings, calling for multi-center validation. Second, pathology-based disease classification was not feasible for most patients, so imaging diagnoses were used instead, limiting the assessment of lesion subtypes. Finally, no direct comparison with breath-hold sequences was performed, leaving open questions about how ACS-RT T2WI might fare against optimized breath-hold approaches.

Conclusions

The study confirmed the clinical reliability of ACS-RT T2WI, despite potential shortcomings in fine structure depiction. While ACS-RT T2WI offers improved image quality and a reduced scanning time compared to ARMS-RT T2WI, further validation and multi-center studies are needed to fully realize its potential benefits, especially in patients with a compromised breath-holding ability or severe illness.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-71/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-71/dss

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-71/coif). D.Y. is an employee of United Imaging Healthcare Technology. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Board of the First Affiliated Hospital of Dalian Medical University (approval number: YJ-KS-KY-2024-631), and the requirement of individual consent for this retrospective analysis was waived.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

  1. Yang T, Li Y, Ye Z, Yao S, Li Q, Yuan Y, Song B. Diffusion Weighted Imaging of the Abdomen and Pelvis: Recent Technical Advances and Clinical Applications. Acad Radiol 2023;30:470-82. [Crossref] [PubMed]
  2. Gatos I, Tsantis S, Karamesini M, Spiliopoulos S, Karnabatidis D, Hazle JD, Kagadis GC. Focal liver lesions segmentation and classification in nonenhanced T2-weighted MRI. Med Phys 2017;44:3695-705. [Crossref] [PubMed]
  3. Kang KA, Kim YK, Kim E, Jeong WK, Choi D, Lee WJ, Jung SH, Baek SY. T2-Weighted Liver MRI Using the MultiVane Technique at 3T: Comparison with Conventional T2-Weighted MRI. Korean J Radiol 2015;16:1038-46. [Crossref] [PubMed]
  4. Aoike T, Fujima N, Yoneyama M, Fujiwara T, Takamori S, Aoike S, Ishizaka K, Kudo K. Development of three-dimensional MR neurography using an optimized combination of compressed sensing and parallel imaging. Magn Reson Imaging 2022;87:32-7. [Crossref] [PubMed]
  5. Xie Y, Tao H, Li X, Hu Y, Liu C, Zhou B, Cai J, Nickel D, Fu C, Xiong B, Chen S. Prospective Comparison of Standard and Deep Learning-reconstructed Turbo Spin-Echo MRI of the Shoulder. Radiology 2024;310:e231405. [Crossref] [PubMed]
  6. Liu K, Xi B, Sun H, Wang J, Chen C, Wen X, Zhang Y, Zeng M. The clinical feasibility of artificial intelligence-assisted compressed sensing single-shot fluid-attenuated inversion recovery (ACS-SS-FLAIR) for evaluation of uncooperative patients with brain diseases: comparison with the conventional T2-FLAIR with parallel imaging. Acta Radiol 2023;64:1943-9. [Crossref] [PubMed]
  7. Yan X, Luo Y, Chen X, Chen EZ, Liu Q, Zou L, Bao Y, Huang L, Xia L. From Compressed-Sensing to Deep Learning MR: Comparative Biventricular Cardiac Function Analysis in a Patient Cohort. J Magn Reson Imaging 2024;59:1231-41. [Crossref] [PubMed]
  8. Foreman SC, Neumann J, Han J, Harrasser N, Weiss K, Peeters JM, Karampinos DC, Makowski MR, Gersing AS, Woertler K. Deep learning-based acceleration of Compressed Sense MR imaging of the ankle. Eur Radiol 2022;32:8376-85. [Crossref] [PubMed]
  9. Zhao Y, Peng C, Wang S, Liang X, Meng X. The feasibility investigation of AI -assisted compressed sensing in kidney MR imaging: an ultra-fast T2WI imaging technology. BMC Med Imaging 2022;22:119. [Crossref] [PubMed]
  10. Cheng T, Li F, Jiang X, Yu D, Wei J, Yuan Y, Xu H. Comparison of different acceleration factors of artificial intelligence-compressed sensing for brachial plexus MRI imaging: scanning time and image quality. BMC Med Imaging 2024;24:309. [Crossref] [PubMed]
  11. Immerkär J. Fast Noise Variance Estimation. Computer Vision and Image Understanding 1996;64:300-2.
  12. 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;47:133-42. [Crossref] [PubMed]
  13. Li H, Hu C, Yang Y, Zhao Y, Lin C, Li Z, Liu Q. Single-breath-hold T2WI MRI with artificial intelligence-assisted technique in liver imaging: As compared with conventional respiratory-triggered T2WI. Magn Reson Imaging 2022;93:175-80. [Crossref] [PubMed]
  14. Yokoo T, Masaki N, Parikh ND, Lane BF, Feng Z, Mendiratta-Lala M, Lee CH, Khatri G, Marsh TL, Shetty K, Dunn CT, Al-Jarrah T, Aslam A, Davenport MS, Gopal P, Rich NE, Lok AS, Singal AG. Multicenter Validation of Abbreviated MRI for Detecting Early-Stage Hepatocellular Carcinoma. Radiology 2023;307:e220917. [Crossref] [PubMed]
  15. Keerthivasan MB, Galons JP, Johnson K, Umapathy L, Martin DR, Bilgin A, Altbach MI. Abdominal T2-Weighted Imaging and T2 Mapping Using a Variable Flip Angle Radial Turbo Spin-Echo Technique. J Magn Reson Imaging 2022;55:289-300. [Crossref] [PubMed]
  16. Shanbhogue K, Tong A, Smereka P, Nickel D, Arberet S, Anthopolos R, Chandarana H. Accelerated single-shot T2-weighted fat-suppressed (FS) MRI of the liver with deep learning-based image reconstruction: qualitative and quantitative comparison of image quality with conventional T2-weighted FS sequence. Eur Radiol 2021;31:8447-57. [Crossref] [PubMed]
  17. Mervak BM, Fried JG, Wasnik AP. A Review of the Clinical Applications of Artificial Intelligence in Abdominal Imaging. Diagnostics (Basel) 2023.
  18. Wang G, Ye JC, Mueller K, Fessler JA. Image Reconstruction is a New Frontier of Machine Learning. IEEE Trans Med Imaging 2018;37:1289-96. [Crossref] [PubMed]
  19. Chen F, Cheng JY, Taviani V, Sheth VR, Brunsing RL, Pauly JM, Vasanawala SS. Data-driven self-calibration and reconstruction for non-cartesian wave-encoded single-shot fast spin echo using deep learning. J Magn Reson Imaging 2020;51:841-53. [Crossref] [PubMed]
  20. Sheng RF, Zheng LY, Jin KP, Sun W, Liao S, Zeng MS, Dai YM. Single-breath-hold T2WI liver MRI with deep learning-based reconstruction: A clinical feasibility study in comparison to conventional multi-breath-hold T2WI liver MRI. Magn Reson Imaging 2021;81:75-81. [Crossref] [PubMed]
  21. Chen F, Taviani V, Malkiel I, Cheng JY, Tamir JI, Shaikh J, Chang ST, Hardy CJ, Pauly JM, Vasanawala SS. Variable-Density Single-Shot Fast Spin-Echo MRI with Deep Learning Reconstruction by Using Variational Networks. Radiology 2018;289:366-73. [Crossref] [PubMed]
  22. Choi ES, Kim JS, Nickel MD, Sung JK, Lee JK. Free-breathing contrast-enhanced multiphase MRI of the liver in patients with a high risk of breath-holding failure: comparison of compressed sensing-accelerated radial and Cartesian acquisition techniques. Acta Radiol 2022;63:1453-62. [Crossref] [PubMed]
  23. Han S, Lee JM, Kim SW, Park S, Nickel MD, Yoon JH. Evaluation of HASTE T2 weighted image with reduced echo time for detecting focal liver lesions in patients at risk of developing hepatocellular carcinoma. Eur J Radiol 2022;157:110588. [Crossref] [PubMed]
Cite this article as: Wang N, Liu Y, Ran J, An Q, Chen L, Zhao Y, Yu D, Liu A, Zhuang L, Song Q. Comparing respiratory-triggered T2WI MRI with an artificial intelligence-assisted technique and motion-suppressed respiratory-triggered T2WI in abdominal imaging. Quant Imaging Med Surg 2025;15(9):7761-7773. doi: 10.21037/qims-2025-71

Article Options

Download Citation

Share