Ekert Kaspar, Hinterleitner Clemens, Baumgartner Karolin, Fritz Jan, Horger Marius
Department of Radiology, Diagnostic and Interventional Radiology, Eberhard Karls University, Hoppe-Seyler-Str. 3, 72076 Tübingen, Germany.
Department of Internal Medicine II, Eberhard-Karls-University, Otfried-Müller-Str. 10, 72076 Tübingen, Germany.
Cancers (Basel). 2020 Mar 24;12(3):761. doi: 10.3390/cancers12030761.
Identifying MRI-based radiomics features capable to assess response to systemic treatment in multiple myeloma (MM) patients. Retrospective analysis of whole-body MR-image data in 67 consecutive stage III MM patients (40 men; mean age, 60.4 years). Bone marrow involvement was evaluated using a standardized MR-imaging protocol consisting of T1w-, short-tau inversion recovery- (STIR-) and diffusion-weighted-imaging (DWI) sequences. Ninety-two radiomics features were evaluated, both in focally and diffusely involved bone marrow. Volumes of interest (VOI) were used. Response to treatment was classified according to International Myeloma Working Group (IMWG) criteria in complete response (CR), very-good and/or partial response (VGPR + PR), and non-response (stable disease (SD) and progressive disease (PD)). According to the IMWG-criteria, response categories were CR ( = 35), VGPR + PR ( = 19), and non-responders ( = 13). On apparent diffusion coefficient (ADC)-maps, gray-level small size matrix small area emphasis (Gray Level Size Zone (GLSZM) small area emphasis (SAE)) significantly correlated with CR ( < 0.001), whereas GLSZM non-uniformity normalized (NUN) significantly ( < 0.008) with VGPR/PR in focal medullary lesions (FL), whereas in diffuse involvement, 1st order root mean squared significantly ( < 0.001) correlated with CR, whereas for VGPR/PR Log (gray-level run-length matrix (GLRLM) Short Run High Gray Level Emphasis) proved significant ( < 0.003). On T1w, GLRLM NUN significantly ( < 0.002) correlated with CR in FL, whereas gray-level co-occurrence matric (GLCM) informational measure of correlation (Imc1) significantly ( < 0.04) correlated with VGPR/PR. For diffuse myeloma involvement, neighboring gray-tone difference matrix (NGTDM) contrast and 1st order skewness were significantly associated with CR and VGPR/PR ( < 0.001 for both). On STIR-images, CR correlated with gray-level co-occurrence matrix (GLCM) Informational Measure of Correlation (IMC) 1 ( < 0.001) in FL and 1st order mean absolute deviation in diffusely involved bone marrow ( < 0.001). VGPR/PR correlated at best in FL with GSZLM size zone NUN ( < 0.019) and in all other involved medullary areas with GLSZM large area low gray level emphasis ( < 0.001). GLSZM large area low gray level emphasis also significantly correlated with the degree of bone marrow infiltration assessed histologically ( = 0.006). GLCM IMC 1 proved significant throughout T1w/STIR sequences, whereas GLSZM NUN in STIR and ADC. MRI-based texture features proved significant to assess clinical and hematological response (CR, VPGR, and PR) in multiple myeloma patients undergoing systemic treatment.
识别基于MRI的放射组学特征,以评估多发性骨髓瘤(MM)患者对全身治疗的反应。对67例连续的III期MM患者(40例男性;平均年龄60.4岁)的全身MR图像数据进行回顾性分析。使用由T1加权、短tau反转恢复(STIR)和扩散加权成像(DWI)序列组成的标准化MR成像方案评估骨髓受累情况。评估了92个放射组学特征,包括局灶性和弥漫性受累骨髓。使用感兴趣体积(VOI)。根据国际骨髓瘤工作组(IMWG)标准将治疗反应分为完全缓解(CR)、非常好和/或部分缓解(VGPR + PR)以及无反应(稳定疾病(SD)和进展性疾病(PD))。根据IMWG标准,反应类别为CR(n = 35)、VGPR + PR(n = 19)和无反应者(n = 13)。在表观扩散系数(ADC)图上,灰度小尺寸矩阵小面积强调(灰度级大小区域(GLSZM)小面积强调(SAE))与CR显著相关(P < 0.001),而在局灶性髓质病变(FL)中,GLSZM归一化非均匀性(NUN)与VGPR/PR显著相关(P < 0.008),而在弥漫性受累中,一阶均方根与CR显著相关(P < 0.001),而对于VGPR/PR,Log(灰度级游程长度矩阵(GLRLM)短游程高灰度级强调)被证明具有显著性(P < 0.003)。在T1加权像上,GLRLM NUN在FL中与CR显著相关(P < 0.002),而灰度级共生矩阵(GLCM)相关信息测度(Imc1)与VGPR/PR显著相关(P < 0.04)。对于弥漫性骨髓瘤受累,相邻灰度色调差异矩阵(NGTDM)对比度和一阶偏度与CR和VGPR/PR均显著相关(两者P均 < 0.001)。在STIR图像上,CR在FL中与灰度级共生矩阵(GLCM)相关信息测度(IMC)1显著相关(P < 0.001),在弥漫性受累骨髓中与一阶平均绝对偏差显著相关(P < 0.001)。VGPR/PR在FL中与GSZLM大小区域NUN相关性最佳(P < 0.019),在所有其他受累髓质区域与GLSZM大面积低灰度级强调显著相关(P < 0.001)。GLSZM大面积低灰度级强调也与组织学评估的骨髓浸润程度显著相关(P = 0.006)。GLCM IMC 1在整个T1加权/STIR序列中均被证明具有显著性,而GLSZM NUN在STIR和ADC中具有显著性。基于MRI的纹理特征被证明对于评估接受全身治疗的多发性骨髓瘤患者的临床和血液学反应(CR、VPGR和PR)具有显著性。