放射组学在多发性骨髓瘤中的相关研究及进展
Related Research and Progress of Radiomics in Multiple Myeloma
DOI: 10.12677/acm.2025.1551438, PDF, HTML, XML,   
作者: 张楠楠, 胡 雪*:重庆医科大学附属第一医院输血科,重庆
关键词: 放射组学多发性骨髓瘤影像学生物标志物深度学习Radiomics Multiple Myeloma Imaging Biomarkers Deep Learning
摘要: 多发性骨髓瘤的精准诊疗长期依赖侵入性骨髓活检,但其全身代表性和重复性不足,限制了预后评估的可靠性。放射组学通过从多模态影像(X线、CT、PET/CT、MRI)中提取高通量定量特征,结合手工特征与深度学习算法(如卷积神经网络),为多发性骨髓瘤的非侵入性全景评估提供了新路径。研究表明,放射组学模型在鉴别MM与脊柱转移瘤、检测微小残留病及预测骨髓浆细胞浸润中显著优于传统影像学。然而,其临床应用受限于可解释性不足、数据标准化缺乏及多中心验证缺失。未来需通过跨学科整合、算法优化及大规模前瞻性研究,推动放射组学从科研向临床转化,最终实现MM个体化分层治疗。
Abstract: The precise diagnosis and treatment of multiple myeloma (MM) have long relied on invasive bone marrow biopsy. However, its insufficient systemic representativeness and repeatability have limited the reliability of prognostic evaluation. Radiomics, by extracting high-throughput quantitative features from multimodal imaging (X-ray, CT, PET/CT, MRI), and combining manual features with deep learning algorithms (such as convolutional neural networks), provides a new pathway for the non-invasive panoramic assessment of multiple myeloma. Studies have shown that radiomics models are significantly superior to traditional imaging techniques in differentiating MM from spinal metastases, detecting minimal residual disease, and predicting bone marrow plasma cell infiltration. Nevertheless, its clinical application is restricted by insufficient interpretability, lack of data standardization, and absence of multicenter validation. In the future, it is necessary to promote the transformation of radiomics from scientific research to clinical practice through interdisciplinary integration, algorithm optimization, and large-scale prospective studies, and ultimately achieve individualized stratified treatment for MM.
文章引用:张楠楠, 胡雪. 放射组学在多发性骨髓瘤中的相关研究及进展[J]. 临床医学进展, 2025, 15(5): 810-818. https://doi.org/10.12677/acm.2025.1551438

1. 引言

多发性骨髓瘤(Multiple Myeloma, MM)是一种克隆浆细胞异常增殖的恶性疾病,其特征是在全身骨髓多个部位存在MM浆细胞[1] [2]。MM是第二常见的血液疾病类型,虽然在40岁以下患者中很少见,但其在老年受试者中的发病率有所上升[3]。随着人口老龄化加剧,MM发病率在世界范围内持续升高,MM病例从1990年的65,940例增加到2019年的155,688例,增加了1.36倍[4]

MM是高度异质性的疾病,患者生存时间在数月至数年不等。通过建立完善的临床分期系统,可以精确判断患者预后,选择合适的治疗方案。影像学和骨髓活检结果在MM的诊断中具有决定性作用[5],从髂后上棘非引导骨髓穿刺性活检中获得的参数如浆细胞浸润(Plasma Cell Infiltration, PCI) [6]-[8]或细胞遗传学异常[9]-[12]也被证明具有生物标志物的价值,因此现在用于分期、风险分层和反应评估[11] [13]。侵入性骨髓活检的缺点是不能同时进行多病灶和频繁的重复检查[5]。传统骨髓活检的全身代表性和重复性不足,因此,我们迫切需要能捕获肿瘤表现的空间分布和局部特征信息的影像学手段,据报道[14],对于各种肿瘤实体,影像组学可以预测肿瘤组织特征。影像组学通过整合影像与深度学习技术,能够显著提升MM的非侵入性诊断和预后评估,本文围绕影像组学在多发性骨髓瘤中的相关研究进展进行综述。

2. 影像学在多发性骨髓瘤中的作用

2.1. 常规X线骨骼检查

X线(X-ray, XR)在医学中运用已经非常广泛,易于获取且相对便宜。近80%的多发性骨髓瘤患者在全身放射骨骼检查(Whole-Body X-ray Radiography of the Skeleton, WBXR)中会有骨骼受累的证据,通常影响椎骨(66%)、肋骨(45%)、颅骨(40%)、肩部(40%)、骨盆(30%)和长骨(25%) [15]。在X线上,骨髓瘤病变最常表现为边界清晰的溶解性病变,边缘不硬化,产生清晰的“穿孔”外观。目前常规X线存在的不可忽视的缺点即灵敏度相对较低,只有当30%或更多的骨小梁被破坏时,病变才可能在放射学上被看见[16]

2.2. CT

与常规XR相比,电子计算机断层扫(Computed Tomography, CT)是一种更灵敏的检测溶骨性疾病的成像方式。例如,在对212名MM和SMM患者的回顾性评估中,25% (54名患者)的WBXR呈阴性,而在全身低剂量CT (Whole-body low-dose computed tomography, WBLDCT)中呈现溶骨性病变(P < 0.001)  [17]。由于CT在检测MM溶骨性病变方面具有优异的敏感性,最新的国际骨髓瘤工作组(International Myeloma Working Group, IMWG)诊断指南指出,在CT或PET-CT的CT部分发现的一个或多个符合MM骨病的标准的溶骨性破坏部位(≥5 mm),无论在X线片上是否可见,这都应被视为符合破坏性骨病变[18]。在CT上,溶骨性骨髓瘤骨病的特点是溶解性、边界清楚的骨病变,边缘无硬化,产生“穿孔”外观。当在软组织内核上观察时,未经处理的溶解性病变表现出内部软组织衰减。多发性骨髓瘤也可表现为扩张性溶解性病变,或更不常见的是骨硬化性病变[19]。然而,CT在治疗反应评估中的作用相对有限,特别是与PET-CT和MRI相比。

2.3. PET-CT

正电子发射断层显像–计算机断层显像(Positron Emission Tomography-Computed Tomography, PET-CT)非常适合MM患者的诊断、预测和治疗反应评估[20]-[22]。PET-CT将肿瘤代谢活动的功能评估与低剂量CT提供的形态学信息相结合,因此成为评估基线和治疗后髓内和髓外疾病的有用工具(图1)。目前IMWG诊断标准强调了PET-CT对于诊断活动性MM的重要性[18]。虽然PET-CT在诊断和预测方面具有明确的价值,但其最重要的作用是评估治疗反应和确定疾病进展。一项包含10项研究、690名患者的大型荟萃分析得出结论,PET-CT有助于评估治疗反应和评估新的疾病部位[23]。美中不足的是,PET-CT的缺点包括成本高、可用性有限以及缺乏标准化成像标准。与WBLDCT相比,PET-CT还具有更高的辐射暴露和更长的检查时间。

Figure 1. Schematic diagram of lesions in PET-CT

1. PET-CT病灶示意图

2.4. MRI

磁共振成像(Magnetic Resonance Imaging, MRI)具有良好的软组织对比度,可以在CT看到骨质破坏之前检测出MM的骨髓信号改变[16]。多发性骨髓瘤有五种MRI模式(图2),它们分别是局灶性病变;均匀弥漫性受累;局灶性和弥漫性受累的混合型;盐和胡椒样改变,其特征是骨髓不均匀,并伴有脂肪岛,以及骨髓正常[24]-[26]。MRI是检测早期骨髓浸润最敏感的检查方法。由于MRI具有优越的软组织对比度,与PET-CT相比,MRI可以更好地观察脂肪变性,表明治疗有积极的反应[27] [28]。然而,MRI并非没有缺点。对于一些植入金属植入物或不兼容MRI的医疗设备的患者,MRI是禁忌的。MRI也需要大量时间,对于骨痛患者,可能会首选更方便的影像学检查。最后,与WBLDCT相比,MRI相对比较昂贵。

注:脊柱矢状T1加权图像显示:A,正常;B,弥漫均匀型;C,局灶型(箭头);D,弥漫性局灶(箭头描绘局灶性病变);E,弥漫性不均匀或“盐和胡椒”的骨髓受累模式。

Figure 2. MRI patterns of multiple myeloma (MM) in the thoracolumbar spine

2. MRI上胸腰椎内MM的各种模式

3. 放射组学

近年来,影像学技术飞速发展,为解决空间异质性难题带来了新的契机[29]。放射组学诞生于2012年,是一种新的图像分析方法,可以从标准医学图像中自动提取大量高通量的定量特征[30]。近年来,放射组学在计算机辅助诊断、预测、治疗等方面的发展的基础上得到了巨大的发展。放射组学代表了发现新型成像生物标志物的高通量数据挖掘方法[31]。从结构和功能图像中提取放射学特征可以反映病变的解剖和功能信息[32]。放射组学的建立是基于医学成像包含大多数肉眼看不见的生物信息,但可以通过使用高通量方法进行量化,与基因组数据中使用的方法相似。放射组学主要使用两种主要方法:手工特征和深度学习。

3.1. 手工提取特征的放射组学

手工提取特征的放射组学特征提供了图像中某个感兴趣区域的信息,可能是肿瘤区域或整个器官[33]。这些特征是多样的,但与其他特征相关(例如,临床、治疗和基因组学信息)。一个主要的挑战是以定量的方式收集和整合多模式数据源,以提供不同的临床特征,以便准确和可靠地估计结果。手工特征放射组学主要包括数据选择与整理、数据预处理、病灶分割、特征提取和模型建立五个阶段[34]

3.2. 基于深度学习的放射组学

与手工特征的放射组学相比,深度学习利用数据驱动的方法来创建模拟简化的脑神经元相互作用的模型。该算法采用深度神经网络来识别模式,通常由多个可训练的非线性操作组成,称为层,每个层将输入的数据转换为一个代表,以促进模式识别[35]。随着越来越多的层利用对输入数据的转换,数据越来越抽象为深度特征表示。最后,将得到的深度学习特征从网络的最终层转换为预期的结果。深度学习的优点是不需要对成像扫描进行预分割[36]

3.2.1. 数据采集和预处理

基于深度学习的放射组学研究需要大量的样本来训练神经网络。因此,需要收集尽可能多的样本。除了图像的标准化和重新采样之外,增强数据还代表了一种额外的数据处理过程,特别是用于深度学习辅助放射学分析[37]。常见的数据增强方法包括图像旋转和缩放、随机垂直、水平翻转等。

3.2.2. 卷积神经网络

在放射学中,大多数基于深度学习的模型利用卷积神经网络来基于成像特征得出预测结果。卷积神经网络作为一种专门为学习图像空间分布而设计的神经网络,因其在诊断方面的功效而受到广泛关注[38]。正如卷积神经网络能够学习显示恶性疾病的影像学特征一样,越来越多的研究表明,卷积神经网络能够根据与预后、风险和分子谱相关的肿瘤特征的细微差异对患者进行分层[39]。当使用患者的临床特征进行训练时,卷积神经网络的卷积层能够学习识别反映预后结果的新成像表型。卷积神经网络被用于二维和三维输入数据,经过许多输入数据的修改,以便从组合的图像类型中学习[40]

3.2.3. 训练基于深度学习的模型

为了训练基于深度学习的模型,神经网络基于训练数据集的子集迭代更新。神经网络首先可以利用每批成像数据生成结果预测[41]。随后,使用损失函数将这种预测与匹配的实际结果进行比较,损失函数测量网络输出的有效性。接下来,利用从损失函数中获得的值来更新通过网络层进行的操作,导致大多数由网络表现不佳的样本通知的更改[42]。第二个患者数据集称为调优数据集,可用于在训练和优化模型的配置和学习过程时监测效果,随后在一个或多个独立数据集中进行验证[43]。训练神经网络通常需要比开发放射组学模型更多的数据。机器学习模型是由一组参数定义的,这些参数指定了方法的每个潜在配置[44]

然而,深度学习的“黑箱”方法,即可解释性差一直是临床应用的主要缺陷[45]。此外,深度学习需要大量数据才能真正表达其性能,并且可能患者队列不足以有效利用深度学习架构[46]

4. 放射组学在MM中的应用

一些研究已经调查了影像组学在鉴别诊断多发性骨髓瘤方面的作用,以及解决更具有挑战性的任务,比如评估治疗反应,或者一些预后模型的建立,甚至是影像组学对于整个肿瘤异质性的预测。

4.1. 鉴别多发性骨髓瘤与脊柱转移瘤

转移瘤与多发性骨髓瘤的准确鉴别具有重要的临床意义,因为它们的治疗决策和预后因病变性质的不同而不同。脊柱转移瘤与多发性骨髓瘤具有许多重叠的影像学特征,尤其是多发性骨髓瘤的溶骨性特征,使得常规影像学手段难以鉴别诊断。在一项研究中,建立了用于区分多发性骨髓瘤转移的放射组学诺谟图,在训练集、验证集和外部测试集中,曲线下面积分别为0.856、0.853和0.762 [47]。该研究开发了一种基于MRI的放射组学列线图,可以预测骨病变是骨髓瘤病变还是骨转移,具有中等到良好的表现。建立了一种非侵入性和有价值的方法来区分脊髓转移和多发性骨髓瘤。

4.2. 预测多发性骨髓瘤的治疗反应

微小残留病(Minimal Residual Disease, MRD)是评估多发性骨髓瘤治疗反应的标准。Wu等人开发和验证了基于腰椎MRI的放射组学诺谟图,以检测MM治疗后的MRD。他们使用十六个特征来建立放射组学特征。并证明放射组学诺谟图在检测MRD状态方面表现出良好的性能(曲线下面积:训练集中为0.980,测试集中为0.903) [48]。徐文健博士团队通过结合放射组学特征和独立的临床风险因素构建诺谟图[49]。通过测定血清和尿中M蛋白水平、无血清轻链比率以及骨髓浆细胞百分比来评估治疗反应。该研究选择了十三个特征来构建放射组学特征,选择国际分期系统(ISS)分期作为独立的临床因素。证明与临床模型相比,放射组学特征的诺谟图显示出更好的校准和更高的辨别能力,有助于预测MM对化疗的反应,并可用于临床决策。

4.3. 预测局部骨髓活组织检查结果

多发性骨髓瘤及其前体阶段,浆细胞浸润(PCI)和细胞遗传学畸变对分期、风险分层和反应评估都很重要。Wennmann等人建立的自动图像分析框架允许非侵入性预测PCI的代理参数,该参数与实际的骨髓活检PCI显著相关[5]。该研究共纳入来自8个中心的512例患者的672例MRI和370例相应的骨髓活检。在所有内部和外部测试组中,最佳模型预测的PCI与活检的实际PCI显著相关(P ≤ 0.01)。

5. 常见的深度学习模型

深度学习技术主要分为三类:1) 用于监督或判别学习的深度网络;2) 用于无监督或生成学习的深度网络;3) 用于混合学习的深度网络,结合了监督、无监督及其他相关学习方法。常见的深度学习模型包括卷积神经网络(Convolutional Neural Network, CNN)、循环神经网络(Recurrent Neural Network, RNN)、自动编码器(Autoencoders, AE)以及深度置信网络(Deep Belief Networks, DBN)等。Qiu等人[50]提出了一种区域卷积神经网络(region-based convolutional neural network, R-CNN)的深度学习框架,用于检测和分割骨髓瘤细胞。该框架与所提出的特征选择金字塔网络相结合,利用一个掩膜聚合模块来优化高确定性的实例掩膜,将这些掩膜合并成一张单一的分割图,并借助额外的语义分割分支所得到的结果来提升分割性能。CNNs特别擅长捕捉图像的分层表示,从边缘和纹理等低级特征到对准确分割至关重要的高级语义特征。这种分层特征提取能力在医学成像中至关重要,其中纹理或形状的细微差异包含关键的诊断信息。CNN架构,如U-Net [51],DeepLab [52]系列和V-Net,已成为医学图像分割领域的突出研究焦点。U-Net结构利用了编码器和解码器组件,能够同时捕捉全局和局部特征,使其适用于各种各样的医学成像任务。

6. 结论与展望

在研究MM患者的各种成像方法之间的选择仍然存在很大的可变性,即使在三级中心,影像科医生之间的图像解释也存在很大的可变性,一致性不佳。因此,医学影像学对MM患者的潜力仍有待完全发掘。未来,人工智能和放射组学在不同骨病变中的应用,以及定量方法报告CT和MRI的广泛实施,将为医学成像研究MM提供新的见解。最终,医学成像数据可以与患者的结果相结合,目的是寻找放射生物标志物来预测疾病的预后和治疗反应。

目前MM的诊断和预后情况太过于依赖骨髓穿刺,然而骨髓穿刺全身代表性不足,已成为现行MM预后分层的技术瓶颈,影响预后预测精准性,迫切需要研究新的全景性评估手段。讨论基于人工智能的放射组学在该领域的局限性和未来的机遇。在个性化医疗时代,新兴的基于人工智能的放射组学方法的深入理解,将为MM带来创新的解决方案并实现临床应用。然而,基于人工智能的放射组学模型在其可解释性方面是有限的,并且在没有对大型代表性数据集进行训练的情况下往往会过度拟合。这方面,进一步的研究将是必要的,以达到必要的标准化水平和多中心验证,将放射组学工具带到床边。总之,放射组学为MM的精准诊疗提供了革命性工具,但其从科研向临床的跨越仍需跨学科协作、技术创新及大规模验证,最终实现从“一刀切”治疗到个体化医疗的范式转变。

NOTES

*通讯作者。

参考文献

[1] Rasche, L., Chavan, S.S., Stephens, O.W., Patel, P.H., Tytarenko, R., Ashby, C., et al. (2017) Spatial Genomic Heterogeneity in Multiple Myeloma Revealed by Multi-Region Sequencing. Nature Communications, 8, Article No. 268
https://doi.org/10.1038/s41467-017-00296-y
[2] Zeissig, M.N., Zannettino, A.C.W. and Vandyke, K. (2020) Tumour Dissemination in Multiple Myeloma Disease Progression and Relapse: A Potential Therapeutic Target in High-Risk Myeloma. Cancers, 12, Article 3643.
https://doi.org/10.3390/cancers12123643
[3] Gray, F.D. (1980) Internal Medicine. JAMA: The Journal of the American Medical Association, 243, 2190-2191.
https://doi.org/10.1001/jama.1980.03300470050029
[4] Zhou, L., Yu, Q., Wei, G., Wang, L., Huang, Y., Hu, K., et al. (2021) Measuring the Global, Regional, and National Burden of Multiple Myeloma from 1990 to 2019. BMC Cancer, 21, Article No. 606.
https://doi.org/10.1186/s12885-021-08280-y
[5] Wennmann, M., Ming, W., Bauer, F., Chmelik, J., Klein, A., Uhlenbrock, C., et al. (2023) Prediction of Bone Marrow Biopsy Results from MRI in Multiple Myeloma Patients Using Deep Learning and Radiomics. Investigative Radiology, 58, 754-765.
https://doi.org/10.1097/rli.0000000000000986
[6] Waxman, A.J., Mick, R., Garfall, A.L., et al. (2015) Classifying Ultra-High Risk Smoldering Myeloma. Leukemia, 29, 751-753.
[7] Rajkumar, S.V., Larson, D. and Kyle, R.A. (2011) Diagnosis of Smoldering Multiple Myeloma. New England Journal of Medicine, 365, 474-475.
https://doi.org/10.1056/nejmc1106428
[8] Kastritis, E., Terpos, E., Moulopoulos, L., Spyropoulou-Vlachou, M., Kanellias, N., Eleftherakis-Papaiakovou, E., et al. (2012) Extensive Bone Marrow Infiltration and Abnormal Free Light Chain Ratio Identifies Patients with Asymptomatic Myeloma at High Risk for Progression to Symptomatic Disease. Leukemia, 27, 947-953.
https://doi.org/10.1038/leu.2012.309
[9] Neben, K., Jauch, A., Hielscher, T., Hillengass, J., Lehners, N., Seckinger, A., et al. (2013) Progression in Smoldering Myeloma Is Independently Determined by the Chromosomal Abnormalities Del(17p), T(4;14), Gain 1q, Hyperdiploidy, and Tumor Load. Journal of Clinical Oncology, 31, 4325-4332.
https://doi.org/10.1200/jco.2012.48.4923
[10] San Miguel, J., Mateos, M., Gonzalez, V., Dimopoulos, M.A., Kastritis, E., Hajek, R., et al. (2019) Updated Risk Stratification Model for Smoldering Multiple Myeloma (SMM) Incorporating the Revised IMWG Diagnostic Criteria. Journal of Clinical Oncology, 37, 8000-8000.
https://doi.org/10.1200/jco.2019.37.15_suppl.8000
[11] Palumbo, A., Avet-Loiseau, H., Oliva, S., Lokhorst, H.M., Goldschmidt, H., Rosinol, L., et al. (2015) Revised International Staging System for Multiple Myeloma: A Report from International Myeloma Working Group. Journal of Clinical Oncology, 33, 2863-2869.
https://doi.org/10.1200/jco.2015.61.2267
[12] Weinhold, N., Salwender, H.J., Cairns, D.A., et al. (2021) Chromosome 1q21 Abnormalities Refine Outcome Prediction in Patients with Multiple Myeloma—A Meta-Analysis of 2,596 Trial Patients. Haematologica, 106, 2754-2758.
[13] Mateos, M.-V., Kumar, S., Dimopoulos, M.A., et al. (2020) International Myeloma Working Group Risk Stratification Model for Smoldering Multiple Myeloma (SMM). Blood Cancer Journal, 10, Article 102.
[14] Aerts, H.J.W.L. (2016) The Potential of Radiomic-Based Phenotyping in Precision Medicine. JAMA Oncology, 2, 1636-1642.
https://doi.org/10.1001/jamaoncol.2016.2631
[15] Kyle, R.A. and Rajkumar, S.V. (2004) Multiple Myeloma. New England Journal of Medicine, 351, 1860-1873.
https://doi.org/10.1056/nejmra041875
[16] Baffour, F.I., Glazebrook, K.N., Kumar, S.K. and Broski, S.M. (2020) Role of Imaging in Multiple Myeloma. American Journal of Hematology, 95, 966-977.
https://doi.org/10.1002/ajh.25846
[17] Hillengass, J., Moulopoulos, L.A., Delorme, S., Koutoulidis, V., Mosebach, J., Hielscher, T., et al. (2017) Whole-Body Computed Tomography versus Conventional Skeletal Survey in Patients with Multiple Myeloma: A Study of the International Myeloma Working Group. Blood Cancer Journal, 7, e599-e599.
https://doi.org/10.1038/bcj.2017.78
[18] Rajkumar, S.V., Dimopoulos, M.A., Palumbo, A., Blade, J., Merlini, G., Mateos, M., et al. (2014) International Myeloma Working Group Updated Criteria for the Diagnosis of Multiple Myeloma. The Lancet Oncology, 15, e538-e548.
https://doi.org/10.1016/s1470-2045(14)70442-5
[19] Mahnken, A.H., Wildberger, J.E., Gehbauer, G., Schmitz-Rode, T., Blaum, M., Fabry, U., et al. (2002) Multidetector CT of the Spine in Multiple Myeloma: Comparison with MR Imaging and Radiography. American Journal of Roentgenology, 178, 1429-1436.
https://doi.org/10.2214/ajr.178.6.1781429
[20] Bredella, M.A., Steinbach, L., Caputo, G., Segall, G. and Hawkins, R. (2005) Value of FDG PET in the Assessment of Patients with Multiple Myeloma. American Journal of Roentgenology, 184, 1199-1204.
https://doi.org/10.2214/ajr.184.4.01841199
[21] Cavo, M., Terpos, E., Nanni, C., Moreau, P., Lentzsch, S., Zweegman, S., et al. (2017) Role of 18F-FDG PET/CT in the Diagnosis and Management of Multiple Myeloma and Other Plasma Cell Disorders: A Consensus Statement by the International Myeloma Working Group. The Lancet Oncology, 18, e206-e217.
https://doi.org/10.1016/s1470-2045(17)30189-4
[22] Bartel, T.B., Haessler, J., Brown, T.L.Y., Shaughnessy, J.D., van Rhee, F., Anaissie, E., et al. (2009) F18-Fluorodeoxyglucose Positron Emission Tomography in the Context of Other Imaging Techniques and Prognostic Factors in Multiple Myeloma. Blood, 114, 2068-2076.
https://doi.org/10.1182/blood-2009-03-213280
[23] Caldarella, C., Treglia, G., Isgrò, M.A., Treglia, I. and Giordano, A. (2012) The Role of Fluorine-18-Fluorodeoxyglucose Positron Emission Tomography in Evaluating the Response to Treatment in Patients with Multiple Myeloma. International Journal of Molecular Imaging, 2012, Article 175803.
https://doi.org/10.1155/2012/175803
[24] Baur-Melnyk, A., Buhmann, S., Dürr, H.R. and Reiser, M. (2005) Role of MRI for the Diagnosis and Prognosis of Multiple Myeloma. European Journal of Radiology, 55, 56-63.
https://doi.org/10.1016/j.ejrad.2005.01.017
[25] Moulopoulos, L.A., Gika, D., Anagnostopoulos, A., Delasalle, K., Weber, D., Alexanian, R., et al. (2005) Prognostic Significance of Magnetic Resonance Imaging of Bone Marrow in Previously Untreated Patients with Multiple Myeloma. Annals of Oncology, 16, 1824-1828.
https://doi.org/10.1093/annonc/mdi362
[26] Dutoit, J.C. and Verstraete, K.L. (2016) MRI in Multiple Myeloma: A Pictorial Review of Diagnostic and Post-Treatment Findings. Insights into Imaging, 7, 553-569.
https://doi.org/10.1007/s13244-016-0492-7
[27] Broski, S.M., Goenka, A.H., Kemp, B.J. and Johnson, G.B. (2018) Clinical PET/MRI: 2018 Update. American Journal of Roentgenology, 211, 295-313.
https://doi.org/10.2214/ajr.18.20001
[28] Kogan, F., Broski, S.M., Yoon, D. and Gold, G.E. (2018) Applications of PET‐MRI in Musculoskeletal Disease. Journal of Magnetic Resonance Imaging, 48, 27-47.
https://doi.org/10.1002/jmri.26183
[29] 方兴宇. 全身弥散加权成像定量分析在多发性骨髓瘤中的应用[D]: [博士学位论文]. 北京: 中国医学科学院北京协和医学院, 2020.
[30] Wu, G., Jochems, A., Refaee, T., Ibrahim, A., Yan, C., Sanduleanu, S., et al. (2021) Structural and Functional Radiomics for Lung Cancer. European Journal of Nuclear Medicine and Molecular Imaging, 48, 3961-3974.
https://doi.org/10.1007/s00259-021-05242-1
[31] Bera, K., Braman, N., Gupta, A., Velcheti, V. and Madabhushi, A. (2021) Predicting Cancer Outcomes with Radiomics and Artificial Intelligence in Radiology. Nature Reviews Clinical Oncology, 19, 132-146.
https://doi.org/10.1038/s41571-021-00560-7
[32] Gidwani, M., Chang, K., Patel, J.B., Hoebel, K.V., Ahmed, S.R., Singh, P., et al. (2023) Inconsistent Partitioning and Unproductive Feature Associations Yield Idealized Radiomic Models. Radiology, 307, e220715.
https://doi.org/10.1148/radiol.220715
[33] Hatt, M., Krizsan, A.K., Rahmim, A., et al. (2023) Joint EANM/SNMMI Guideline on Radiomics in Nuclear Medicine: Jointly Supported by the EANM Physics Committee and the SNMMI Physics, Instrumentation and Data Sciences Council. European Journal of Nuclear Medicine and Molecular Imaging, 50, 352-375.
[34] Wei, J., Jiang, H., Zhou, Y., Tian, J., Furtado, F.S. and Catalano, O.A. (2023) Radiomics: A Radiological Evidence-Based Artificial Intelligence Technique to Facilitate Personalized Precision Medicine in Hepatocellular Carcinoma. Digestive and Liver Disease, 55, 833-847.
https://doi.org/10.1016/j.dld.2022.12.015
[35] Cui, Y., Zhang, J., Li, Z., Wei, K., Lei, Y., Ren, J., et al. (2022) A CT-Based Deep Learning Radiomics Nomogram for Predicting the Response to Neoadjuvant Chemotherapy in Patients with Locally Advanced Gastric Cancer: A Multicenter Cohort Study. eClinicalMedicine, 46, Article 101348.
https://doi.org/10.1016/j.eclinm.2022.101348
[36] Dai, J., Wang, H., Xu, Y., Chen, X. and Tian, R. (2023) Clinical Application of AI-Based PET Images in Oncological Patients. Seminars in Cancer Biology, 91, 124-142.
https://doi.org/10.1016/j.semcancer.2023.03.005
[37] Faber, J., Kügler, D., Bahrami, E., Heinz, L., Timmann, D., Ernst, T.M., et al. (2022) CerebNet: A Fast and Reliable Deep-Learning Pipeline for Detailed Cerebellum Sub-Segmentation. NeuroImage, 264, Article 119703.
https://doi.org/10.1016/j.neuroimage.2022.119703
[38] Wang, R., Dai, W., Gong, J., Huang, M., Hu, T., Li, H., et al. (2022) Development of a Novel Combined Nomogram Model Integrating Deep Learning-Pathomics, Radiomics and Immunoscore to Predict Postoperative Outcome of Colorectal Cancer Lung Metastasis Patients. Journal of Hematology & Oncology, 15, Article No. 11.
https://doi.org/10.1186/s13045-022-01225-3
[39] Geras, K.J., Mann, R.M. and Moy, L. (2019) Artificial Intelligence for Mammography and Digital Breast Tomosynthesis: Current Concepts and Future Perspectives. Radiology, 293, 246-259.
https://doi.org/10.1148/radiol.2019182627
[40] Ziegelmayer, S., Reischl, S., Harder, F., Makowski, M., Braren, R. and Gawlitza, J. (2021) Feature Robustness and Diagnostic Capabilities of Convolutional Neural Networks against Radiomics Features in Computed Tomography Imaging. Investigative Radiology, 57, 171-177.
https://doi.org/10.1097/rli.0000000000000827
[41] Han, S., Oh, J.S., Kim, Y., Seo, S.Y., Lee, G.D., Park, M., et al. (2022) Fully Automatic Quantitative Measurement of 18F-FDG PET/CT in Thymic Epithelial Tumors Using a Convolutional Neural Network. Clinical Nuclear Medicine, 47, 590-598.
https://doi.org/10.1097/rlu.0000000000004146
[42] van der Voort, S.R., Incekara, F., Wijnenga, M.M.J., Kapsas, G., Gahrmann, R., Schouten, J.W., et al. (2023) Combined Molecular Subtyping, Grading, and Segmentation of Glioma Using Multi-Task Deep Learning. Neuro-Oncology, 25, 279-289.
https://doi.org/10.1093/neuonc/noac166
[43] Pang, S., Field, M., Dowling, J., Vinod, S., Holloway, L. and Sowmya, A. (2022) Training Radiomics-Based CNNs for Clinical Outcome Prediction: Challenges, Strategies and Findings. Artificial Intelligence in Medicine, 123, Article 102230.
https://doi.org/10.1016/j.artmed.2021.102230
[44] Sexauer, R., Yang, S., Weikert, T., Poletti, J., Bremerich, J., Roth, J.A., et al. (2022) Automated Detection, Segmentation, and Classification of Pleural Effusion from Computed Tomography Scans Using Machine Learning. Investigative Radiology, 57, 552-559.
https://doi.org/10.1097/rli.0000000000000869
[45] Guiot, J., Vaidyanathan, A., Deprez, L., et al. (2021) A Review in Radiomics: Making Personalized Medicine a Reality via Routine Imaging. Medicinal Research Reviews, 42, 426-440.
[46] Lee, S.B., Cho, Y.J., Hong, Y., Jeong, D., Lee, J., Kim, S., et al. (2021) Deep Learning-Based Image Conversion Improves the Reproducibility of Computed Tomography Radiomics Features. A Phantom Study. Investigative Radiology, 57, 308-317.
https://doi.org/10.1097/rli.0000000000000839
[47] Zhang, S., Liu, M., Li, S., Cui, J., Zhang, G. and Wang, X. (2023) An MRI-Based Radiomics Nomogram for Differentiating Spinal Metastases from Multiple Myeloma. Cancer Imaging, 23, Article No. 72.
https://doi.org/10.1186/s40644-023-00585-4
[48] Wu, Z., Wang, H., Zheng, Y., Fei, H., Dong, C., Wang, Z., et al. (2023) Lumbar MR-Based Radiomics Nomogram for Detecting Minimal Residual Disease in Patients with Multiple Myeloma. European Radiology, 33, 5594-5605.
https://doi.org/10.1007/s00330-023-09540-0
[49] Wu, Z., Bian, T., Dong, C., Duan, S., Fei, H., Hao, D., et al. (2022) Spinal MRI-Based Radiomics Analysis to Predict Treatment Response in Multiple Myeloma. Journal of Computer Assisted Tomography, 46, 447-454.
https://doi.org/10.1097/rct.0000000000001298
[50] Qiu, X., Lei, H., Xie, H. and Lei, B. (2022) Segmentation of Multiple Myeloma Cells Using Feature Selection Pyramid Network and Semantic Cascade Mask RCNN. 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI), Kolkata, 28-31 March 2022, 1-4.
https://doi.org/10.1109/isbi52829.2022.9761460
[51] Ronneberger, O., Fischer, P. and Brox, T. (2015) U-Net: Convolutional Networks for Biomedical Image Segmentation. Proceedings of the Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015: 18th International Conference, Munich, 5-9 October 2015, 234-241.
[52] Chen, L., Papandreou, G., Kokkinos, I., Murphy, K. and Yuille, A.L. (2018) Deeplab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs. IEEE Transactions on Pattern Analysis and Machine Intelligence, 40, 834-848.
https://doi.org/10.1109/tpami.2017.2699184