第一部分 基于深度学习重建技术的快速膀胱MRI的前瞻性研究

目的：膀胱影像报告和数据系统（vesical imaging reporting and data system, VI-RADS）推荐的磁共振成像（magnetic resonance imaging, MRI）检查方案包括多平面的T2加权成像（T2-weighted imaging, T2WI）、扩散加权成像（diffusion-weighted imaging, DWI）及动态对比增强成像。然而，多序列的扫描方案不仅耗时较长，增加了患者的身心负担，还可能因运动伪影导致图像质量下降，影响诊断的准确性。深度学习重建（deep learning reconstruction, DLR）技术可在缩短扫描时间的同时提升图像质量，但其在膀胱MRI扫描方案中的应用尚不广泛。因此，本研究旨在通过前瞻性临床研究评估DLR在快速膀胱MRI扫描中的应用对检查时间、图像质量及基于VI-RADS诊断膀胱癌肌层浸润的影响。

材料与方法：这项前瞻性研究纳入2022年8月至2023年2月期间70例膀胱癌患者，其中15例为肌层浸润性膀胱癌（15/70，21%）。扫描方案包括轴位、冠状位的标准 T2WI（standard T2WI, T2WI_S）、标准DWI（standard DWI, DWI_S）、动态对比增强成像、应用DLR的快速T2WI（Fast T2WI with DLR, T2WI_DL）和应用DLR的快速 DWI（Fast DWI with DLR, DWI_DL）。3名分别具有21年、9年和4年腹部放射学经验的医师对图像质量进行评估。通过测量信噪比（signal-to-noise ratio, SNR）、对比噪声比（contrast-to-noise ratio, CNR）来定量比较标准序列和应用DLR的快速序列的图像质量。通过评估各序列整体图像质量、图像清晰度、病变显著性和伪影情况进行定性图像分析。此外，在DWI_S和DWI_DL上测量膀胱癌的表观扩散系数（apparent diffusion coefficient, ADC），并对病灶进行VI-RADS 评分。采用配对t检验或配对Wilcoxon符号秩检验比较标准序列和应用DLR快速序列的图像质量评分、SNR、CNR和ADC。使用受试者工作特征曲线下面积（area under the receiver operating characteristic curve, AUC）评估VI-RADS评分的诊断性能。组内相关系数（intraclass correlation coefficient, ICC）用于评估SNR、CNR和ADC的观察者间一致性；使用Fleiss kappa值评估定性图像质量评分和VI-RADS评分的观察者间一致性。

结果：与轴位、冠状位T2WI_S和DWI_S相比，轴位、冠状位T2WI_DL和DWI_DL的采集时间由5分57秒缩短至3分13秒，定性图像质量评分包括整体图像质量、图像清晰度、病变显著性均显著提高。轴位T2WI_DL的SNR显著高于轴位T2WI_S（医师1：17.0 ± 7.3 vs 30.7 ± 14.3，p < 0.001），轴位T2WI_DL的CNR也显著高于轴位T2WI_S（医师1：5.6 ± 1.9 vs 9.3 ± 3.4，p < 0.001），而冠状位T2WI_DL与DWI_DL的SNR和CNR也显著高于对应的冠状位T2WI_S与DWI_S的SNR和CNR（全部 p < 0.001）。DWI_S与DWI_DL的ADC无显著差异（医师1：1.24 ± 0.25 vs. 1.22 ± 0.23）。评估膀胱癌肌层浸润方面，标准序列和应用DLR快速序列在VI-RADS评分的AUC之间无显著差异（医师1：0.96 vs 0.97, p > 0.05）。三位医师测得的ADC、SNR、CNR的ICC表现出较低至很好的一致性（ICC：0.53-0.97）；VI-RADS评分具有很好一致性（Fleiss kappa值：0.84-0.85）；定性图像质量的评估表现出中等至较好的一致性（Fleiss kappa值：0.60-0.82）。      

结论：DLR应用于T2WI和DWI能够有效缩短扫描时间，显著提高图像质量，且不会显著影响病灶的ADC及评估膀胱癌肌层浸润的诊断准确性。

第二部分 应用深度学习重建技术的小视野DWI在评估膀胱癌肌层浸润中的价值

目的：膀胱癌是泌尿系统中常见的恶性肿瘤，准确评估其肌层浸润情况对临床决策至关重要。多参数磁共振成像（magnetic resonance imaging, MRI）是评估膀胱癌局部分期的重要工具，其中弥散加权成像（diffusion-weighted imaging, DWI）是关键序列。然而，常规全视野（full field-of-view, fFOV）DWI面临磁敏感伪影、图像畸变及空间分辨率受限等问题。小视野（reduced field-of-view, rFOV）DWI旨在解决这些问题，但存在扫描时间长等局限性。近年来，深度学习重建（deep learning reconstruction, DLR）技术在医学影像领域展现出巨大潜力。本研究旨在探索应用DLR技术的rFOV DWI在改善评估膀胱癌肌层浸润准确性方面的价值。

材料与方法：2022年8月至2023年3月期间，该前瞻性研究连续纳入86名进行膀胱MRI检查的膀胱癌患者。检查方案包括三个不同类型的DWI序列：常规fFOV DWI、标准小视野（standard rFOV, rFOV_STA）DWI以及应用DLR的快速小视野（rFOV DWI with DLR, rFOV_DLR）DWI。图像分析在AW 4.7工作站上进行，由两位具有不同腹部MRI经验的放射科医师独立评估整体图像质量、伪影、膀胱壁清晰度，并采用4分Likert量表进行评分。同时，根据VI-RADS评分标准独立评估肌层浸润情况，并测量病灶的表观扩散系数（apparent diffusion coefficient, ADC）。此外，我们还测量了不同DWI序列的信噪比（signal-to-noise ratio, SNR）和对比噪声比（contrast-to-noise ratio, CNR），以评估图像质量的客观指标。定量数据和Likert评分采用Friedman检验分析，并通过Dunn’s配对后续检验进行比较。多重比较采用Bonferroni校正的P值。使用受试者工作特征曲线分析VI-RADS在预测膀胱癌肌层浸润中的准确性，并计算敏感度、特异度、阳性预测值、阴性预测值、准确率和受试者工作特征曲线下面积（area under the receiver operating characteristic curve, AUC）。

结果：在图像质量评估方面，rFOV_DLR DWI的整体图像质量、伪影和膀胱壁清晰度评分均优于rFOV_STA DWI，且与fFOV DWI相当。定量评估结果显示，rFOV_DLR DWI的SNR高于rFOV_STA DWI（医师1：87.2 ± 39.2 vs 55.6 ± 21.7，p < 0.001），且与fFOV DWI无显著差异（医师1：102.7 ± 53.1 vs 87.2 ± 39.2，p = 0.28）。肿瘤与髂腰肌之间的CNR在rFOV_DLR DWI中明显高于rFOV_STA DWI（医师1：7.5 ± 2.3 vs 6.7 ± 1.9，p < 0.001）；肿瘤与髂腰肌之间的CNR在rFOV_DLR DWI中也明显高于fFOV DWI（医师1：7.5 ± 2.3 vs 6.3 ± 1.8，p < 0.001）。在应用VI-RADS评估膀胱癌肌层浸润情况方面，rFOV_STA DWI和rFOV_DLR DWI的AUC显著高于fFOV DWI （所有p < 0.05）。fFOV DWI、rFOV_STA DWI和rFOV_DLR DWI三者之间的ADC无显著差异。rFOV_DLR DWI在图像质量评估和VI-RADS评分方面表现出较好或很好的一致性。

结论：与fFOV DWI相比，rFOV_DLR DWI可提高评估膀胱癌肌层浸润的诊断准确性。应用DLR技术不仅缩短了采集时间，还提高了整体图像质量，同时对ADC和诊断性能无显著影响。这些结果有助于rFOV_DLR DWI在临床中的广泛应用，为膀胱癌的精准诊断与治疗提供更加高效、可靠的影像学支持。

第三部分 前瞻性比较二维与三维T2加权成像的图像质量及评估膀胱癌肌层浸润的诊断表现

目的：明确膀胱癌是否存在肌层浸润对治疗决策至关重要。T2加权成像（T2-weighted imaging, T2WI）是评估膀胱癌局部分期的基本序列。本研究旨在比较三维（three-dimensional, 3D）与二维（two-dimensional, 2D）T2WI的图像质量和应用膀胱成像报告和数据系统（vesical imaging reporting and data system, VI-RADS）评估膀胱癌肌层浸润情况的诊断性能。

材料与方法：本前瞻性研究于2022年8月至2023年5月期间纳入101名膀胱癌患者。所有患者均进行了多参数MRI扫描，包括2D T2WI、3D T2WI、扩散加权成像和动态对比增强成像。两名放射科医师进行图像分析。使用4分Likert量表对图像质量进行评分，主要评估整体图像质量、信噪比主观印象、膀胱壁边界清晰度、靶病灶显著性以及运动伪影。根据VI-RADS评分标准，对靶病灶进行评分，并将评分结果与病理结果进行比较。最终，通过受试者工作特征曲线（area under the receiver operating characteristic curve, AUC）分析，评估2D T2WI、3D T2WI、2D方案（包括2D T2WI、扩散加权成像和动态对比增强成像）以及3D方案（包括3D T2WI、扩散加权成像和动态对比增强成像）的VI-RADS评分在评估膀胱癌肌层浸润中的表现，并进行McNemar检验分析其准确性。

结果：图像质量评估显示，3D T2WI在整体图像质量、膀胱壁边界清晰度及靶病灶显著性方面均显著优于2D T2WI（p < 0.001）。然而，3D T2WI的信噪比主观印象评分显著低于2D T2WI（p < 0.001）。在应用VI-RADS评估膀胱癌肌层浸润的诊断性能方面，两位医师的3D T2WI评分的AUC显著高于2D T2WI评分（医师1: 0.94 vs. 0.91, p = 0.02；医师2: 0.92 vs. 0.88, p = 0.04）。对于医生1，3D T2WI评分在预测膀胱癌肌层浸润方面的准确率（0.83 vs. 0.79, p = 0.04）、敏感度（0.85 vs. 0.81, p = 0.04）和特异度（0.83 vs. 0.79, p = 0.04）均高于2D T2WI评分。而对于医生2，尽管3D T2WI评分在敏感度上优于2D T2WI评分（0.85 vs0.77, p = 0.04），但在准确率和特异度上两者差异无统计学意义。3D方案的VI-RADS评分在准确率上也优于2D方案（医师1: 0.93 vs. 0.92, p = 0.02；医师2: 0.93 vs. 0.91, p = 0.02）。然而2D方案和3D方案的整体VI-RADS评分在评估膀胱癌肌层浸润的AUC相近，两者间差异无统计学意义（p > 0.05）。

结论：本研究表明，3D T2WI的图像质量和评估膀胱癌肌层浸润的诊断性能显著优于常规2D T2WI。3D T2WI能够更清晰地显示膀胱壁和病灶边界，从而提高了3D T2WI评分诊断膀胱癌肌层浸润情况的准确性。然而，3D T2WI对整体VI-RADS评分的诊断性能提升有限。

Abstract

Part I: A Prospective Study on Fast Bladder MRI with Deep Learning Reconstruction Techniques

Objective: The vesical imaging reporting and data system (VI-RADS) recommends a magnetic resonance imaging (MRI) protocol that includes multiplanar T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced imaging. However, this multi-sequence protocol is time-consuming, increasing both the physical and psychological burden on patients, and may lead to decreased image quality due to patient movement, thereby affecting diagnostic accuracy. Deep learning reconstruction (DLR) technology can shorten scan time while improving image quality, yet its application in bladder MRI protocols remains limited. Therefore, this prospective clinical study aims to evaluate the impact of DLR on scan time, image quality, and VI-RADS-based diagnosis of muscle-invasive bladder cancer (MIBC) in fast bladder MRI.

Materials and Methods: This prospective study enrolled 70 patients with bladder cancer from August 2022 to February 2023, including 15 cases of muscle-invasive bladder cancer (15/70, 21%). The MRI protocol included axial and coronal standard T2WI (T2WI_S), standard DWI (DWI_S), fast T2WI with DLR (T2WI_DL), and fast DWI with DLR (DWI_DL). Three radiologists with 21, 9, and 4 years of experience in abdominal radiology assessed image quality. Quantitative analysis of image quality was performed by measuring the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) between standard and DLR-accelerated sequences. Qualitative image analysis involved evaluating overall image quality, sharpness, lesion conspicuity, and artifacts. Apparent diffusion coefficient (ADC) of bladder cancer lesions were measured on DWI_S and DWI_DL, and VI-RADS scoring was performed. Paired t-tests or Wilcoxon signed-rank tests were used to compare image quality scores, SNR, CNR, and ADC values between standard and DLR-accelerated sequences. The diagnostic performance of VI-RADS was assessed using the area under the receiver operating characteristic curve (AUC). Interobserver consistency for SNR, CNR, and ADC was evaluated using the intraclass correlation coefficient (ICC), and Fleiss kappa values were used to assess interobserver agreement for qualitative image quality scores and VI-RADS scoring.

Results: Compared to axial and coronal T2WI_S and DWI_S, the scan time for axial and coronal T2WI_DL and DWI_DL was reduced from 5 minutes 57 seconds to 3 minutes 13 seconds. Qualitative image quality scores, including overall image quality, sharpness, and lesion conspicuity, were significantly improved. The axial T2WI_DL showed a significantly higher SNR than the axial T2WI_S (Radiologist 1: 17.0 ± 7.3 vs. 30.7 ± 14.3, p < 0.001), and the axial T2WI_DL had a significantly higher CNR than the axial T2WI_S (Radiologist 1: 5.6 ± 1.9 vs. 9.3 ± 3.4, p < 0.001). The coronal T2WI_DL and DWI_DL also exhibited significantly higher SNR and CNR than their respective standard sequences (all p < 0.001). There was no significant difference in ADC values between DWI_S and DWI_DL (Radiologist 1: 1.24 ± 0.25 vs. 1.22 ± 0.23, p > 0.05). For diagnosing muscle-invasive bladder cancer, no statistically significant difference was observed in the AUCs of VI-RADS scores between standard and DLR-accelerated sequences (Radiologist 1: 0.96 vs. 0.97, p > 0.05). The ICC for ADC, SNR, and CNR measurements showed moderate to excellent agreement among the three radiologists (ICC: 0.53–0.97); VI-RADS scoring demonstrated excellent interobserver agreement (Fleiss kappa: 0.84–0.85); and qualitative image quality assessments showed moderate to good agreement (Fleiss kappa: 0.60–0.82).

Conclusion: The application of DLR to T2WI and DWI can effectively reduce scan time and significantly improve image quality, without significantly affecting the ADC of lesions or the diagnostic accuracy for assessing muscle-invasive bladder cancer.

Part II: The Value of Reduced Field-of-View DWI with Deep Learning Reconstruction Techniques in Assessing Muscle Invasion of Bladder Cancer

Objective: Bladder cancer is a common malignant tumor of the urinary system, and accurately assessing its muscle invasion is crucial for clinical decision-making. Multiparametric magnetic resonance imaging (MRI) is an important tool for evaluating the local staging of bladder cancer, with diffusion-weighted imaging (DWI) being a key sequence. However, conventional full field-of-view (fFOV) DWI faces challenges such as magnetic susceptibility artifacts, image distortion, and limited spatial resolution. Reduced field-of-view (rFOV) DWI can address these issues but is limited by longer scan times. In recent years, deep learning reconstruction (DLR) technology has shown great potential in the field of MRI. This study aims to explore the value of rFOV DWI with DLR in improving the accuracy of assessing muscle invasion in bladder cancer.

Materials and Methods: This prospective study consecutively enrolled 86 bladder cancer patients who underwent bladder MRI between August 2022 and March 2023. The examination protocol included three different types of DWI: conventional fFOV DWI, standard rFOV (rFOV_STA) DWI, and fast rFOV with DLR (rFOV_DLR) DWI. Imaging analysis was performed on an AW 4.7 workstation by two radiologists with different levels of experience in abdominal MRI, who independently assessed overall image quality, artifacts, and bladder wall clarity using a 4-point Likert scale. Additionally, muscular invasion was independently evaluated according to the VI-RADS scoring system, and the apparent diffusion coefficient (ADC) of the lesions were measured. Furthermore, the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of different DWI sequences were measured to assess objective image quality. Quantitative data and Likert scores were compared using the Friedman test, followed by Dunn’s post hoc test. Multiple comparisons were adjusted for using the Bonferroni correction of p-values. The accuracy of VI-RADS in predicting muscular invasion in bladder cancer was analyzed using receiver operating characteristic curves, calculating sensitivity, specificity, positive predictive value, negative predictive value, accuracy, and the area under the receiver operating characteristic curve (AUC).

Results: In image quality assessment, the overall image quality, artifacts, and bladder wall clarity scores of rFOV_DLR DWI were superior to those of rFOV_STA DWI and were comparable to fFOV DWI. Quantitative evaluation showed that rFOV_DLR DWI had a higher SNR than rFOV_STA DWI (Radiologist 1: 87.2±39.2 vs 55.6±21.7, p < 0.001), with no significant difference compared to fFOV DWI (Radiologist 1: 102.7±53.1 vs 87.2±39.2, p = 0.28). The CNR between the tumor and the iliopsoas muscle was significantly higher in rFOV_DLR DWI compared to rFOV_STA DWI (Radiologist 1: 7.5±2.3 vs 6.7±1.9, p < 0.001) and also higher than fFOV DWI (Radiologist 1: 7.5±2.3 vs 6.3±1.8, p < 0.001). In the evaluation of muscular invasion using VI-RADS, both rFOV_STA DWI and rFOV_DLR DWI showed significantly higher AUCs compared to fFOV DWI (all p < 0.05). There were no significant differences in the ADC between fFOV DWI, rFOV_STA DWI, and rFOV_DLR DWI. rFOV_DLR DWI demonstrated good or excellent consistency in image quality assessment and VI-RADS scoring.

Conclusion: Compared to fFOV DWI, rFOV DWI with DLR improves the diagnostic accuracy in assessing muscle invasion of bladder cancer. The DLR technology not only shortens acquisition time but also enhances overall image quality, without exhibiting significant impacts on ADC and diagnostic performance. These findings support the broader clinical application of rFOV DWI, providing more efficient and reliable imaging support for the precise diagnosis and treatment of bladder cancer.

Part III: Prospective Comparison of 2D vs. 3D T2-Weighted Imaging for Assessing Muscle Invasion in Bladder Cancer: A Study on Image Quality and Diagnostic Accuracy

Objective: Determining the presence of muscle invasion in bladder cancer is crucial for treatment decision-making. T2-weighted imaging (T2WI) is a fundamental sequence for assessing bladder cancer staging. This study aims to compare the image quality and diagnostic performance of three-dimensional (3D) and two-dimensional (2D) T2WI using the Vesical Imaging Reporting and Data System (VI-RADS) in diagnosing muscle invasion of bladder cancer.

Materials and Methods: This prospective study enrolled 101 bladder cancer patients from August 2022 to May 2023. All patients underwent multiparametric MRI scans, including 2D T2WI, 3D T2WI, diffusion-weighted imaging, and dynamic contrast-enhanced imaging. Two radiologists analyzed the images. Image quality was assessed using a 4-point Likert scale, focusing on overall image quality, signal-to-noise ratio (SNR) impression, clarity of bladder wall boundaries, lesion conspicuity, and motion artifacts. Target lesions were scored according to the VI-RADS criteria, and the results were compared with pathological findings. The diagnostic performance of the VI-RADS score of 2D T2WI, 3D T2WI, 2D protocol (including 2D T2WI, diffusion-weighted imaging, and dynamic contrast-enhanced imaging), and 3D protocol (including 3D T2WI, diffusion-weighted imaging, and dynamic contrast-enhanced imaging) in assessing muscle invasion of bladder cancer was evaluated using the area under the receiver operating characteristic curve (AUC). McNemar's test was used to analyze accuracy.

Results: In assessing muscle invasion of bladder cancer based on VI-RADS scores, the AUC values of 3D T2WI scores were significantly higher than those of 2D T2WI scores for both radiologists (Radiologist 1: 0.94 vs. 0.91, p = 0.02; Radiologist 2: 0.92 vs. 0.88, p = 0.04). For Radiologist 1, the accuracy (0.83 vs. 0.79, p = 0.04), sensitivity (0.85 vs. 0.81, p = 0.04), and specificity (0.83 vs. 0.79, p = 0.04) of 3D T2WI scores were higher than those of 2D T2WI scores. For Radiologist 2, although the sensitivity of 3D T2WI scores was superior to that of 2D T2WI scores (0.85 vs. 0.77, p = 0.04), there were no significant differences in accuracy and specificity. The overall VI-RADS scores of the 3D protocol were also more accurate than those of the 2D protocol (Radiologist 1: 0.93 vs. 0.92, p = 0.02; Radiologist 2: 0.93 vs. 0.91, p = 0.02). However, the AUC of the 2D and 3D protocols in assessing muscle invasion of bladder cancer were similar, with no significant difference (p > 0.05). Image quality assessment showed that 3D T2WI significantly outperformed 2D T2WI in overall image quality, clarity of bladder wall boundaries, and lesion conspicuity (p < 0.001). However, the SNR impression of 3D T2WI was significantly lower than that of 2D T2WI (p < 0.001).

Conclusion: This study demonstrates that the image quality and diagnostic performance of 3D T2WI in assessing muscle invasion of bladder cancer are significantly superior to those of conventional 2D T2WI. 3D T2WI provides clearer visualization of bladder walls and lesion boundaries, thereby improving the accuracy of 3D T2WI scores in diagnosing muscle invasion of bladder cancer. However, the improvement in diagnostic performance of overall VI-RADS scores by 3D T2WI is limited.