1381
Usefulness of 3-dimensional MR Texture Analysis (3D-MRTA) for Distinguishing Well-differentiated Liposarcoma fom Lipoma1Department of Radiology, Kyung Hee University Hospital at Gangdong, Seoul, Korea, Republic of, 2Department of Radiology, Kyung Hee University Hospital, Seoul, Korea, Republic of
Synopsis
Although magnetic resonance (MR) imaging
has been used as standard imaging tool, because of overlapping imaging features,
the discrimination between well-differentiated liposarcoma and lipoma is not
always clear. Also, no study has evaluated the diagnostic usefulness of 3-dimensional
MR texture analysis (3D-MRTA) for distinguishing well-differentiated
liposarcoma from lipoma. Therefore, the purpose of this study is to evaluate
the usefulness of 3D-MRTA for differentiation between well-differentiated
liposarcoma and lipoma, diagnosed on the basis of histopathological features.
Regarding mean, MPP, kurtosis, and entropy, values on all sequences were
significantly higher in well-differentiated liposarcoma than those in lipoma (p<0.001). The two best performing 3D-MRTA
parameters were kurtosis on T2WI (AUC, 0.86; 95 % CI, 0.77-0.95) and entropy on
FS-T1CE (AUC, 0.83; 95 % CI, 0.74- 0.93). There were good or excellent
interobserver agreements for all measurements (ICC, 0.750–0.885).
Introduction
For distinguishing well-differentiated liposarcoma from lipoma, magnetic resonance (MR) imaging has been used as standard imaging tool (1, 2). However, because of overlapping imaging features, the discrimination between well-differentiated liposarcoma and lipoma is not always clear (3). Although image-guided biopsy can be considered, it is vulnerable to sampling errors and potentially facilitate local tumor spread (4, 5). Therefore, a needs exists the non-invasive, advanced, less subjective imaging method that could help distinguish well-differentiated liposarcoma from lipoma. Texture analysis (TA) is a advanced image analysis technique that can detect and quantify heterogeneity of tissue characteristics which cannot be detected by the human eye (6, 7). For this reason is that complex microscopic tumor heterogeneity resulting in structures of different sizes and variation may be reflected indirectly by the distribution of grey-scale levels and/or pixel intensity on diagnostic images. With this background, we hypothesized that 3-dimensional MR texture analysis (3D-MRTA) may be useful for distinguishing well-differentiated liposarcoma from lipoma. To our best knowledge, no study has evaluated the diagnostic usefulness of 3D-MRTA for distinguishing well-differentiated liposarcoma from lipoma. Therefore, the purpose of this study is to evaluate the usefulness of 3D-MRTA for differentiation between well-differentiated liposarcoma and lipoma, diagnosed on the basis of histopathological features.Methods
The inclusion criteria were: (1) patients with histopathologically confirmed well-differentiated liposarcoma or lipoma via entire surgical resection and (2) patients with pre-operative MRI with contrast enhancement. From December 2006 to September 2018, a total of 68 consecutive patients with well-differentiated liposarcoma and lipoma were initially included. Among them, seven patients were excluded because of the metallic artifact (n=4), severe motion artifact (n=2), and recurrent lipomatous tumors (n=1). MRI was obtained using a 3-T system or 1.5-T system. The MRI protocols included a variety of sequences in axial, coronal, and sagittal planes using T2-weighted fast spin-echo (T2WI), T2-weighted fast spin-echo with fat suppression image (T2FS), T1-weighted image (T1WI), and fat-suppressed T1-contrast-enhanced (FS-T1CE) images. Axial T2WI, axial T2FS, axial T1WI, and axial FS-T1CE images were uploaded into proprietary TexRAD research software. Region of interests (ROIs) were drawn independently by two reviewers with a 1-week interval. Reviewers were blinded to any patients’ information including histological analysis. For ROI placement, “seedpoint” mode was used which was automatically drawing ROI along the margin of the mass by click the mass. If the automatically drawn ROI was located in outside of the mass, it was permitted that the reviewer modified the ROI freehand with “polygonal” mode. For 3D-MRTA, the ROIs were automatically drawn in all axial images in which the mass was included and summed the information from the each ROIs using “batch” reconstruction. Regarding the order of the MR sequence, all ROIs were initially drawn on the axial T2WI. And then, ROIs was copied and placed for axial T2FS, axial T1WI, and axial FS-T1CE. Quantification of histograms was based on mean, skewness, mean of positive pixels (MPP), kurtosis, and entropy. Statstically, the independent t-test, receiver operating characteristic (ROC) curve, and intraclass correlation coefficient (ICC) were performed.Results
A total of 61 patients (mean age 50.4±15.5 years; range), including 27 male (mean age 44.7±15.2 years; range 20-72) and 34 female (mean age 54.1±15.7 years; range 28-67), were ultimately enrolled in this study. Patients underwent surgical resection within 1 month of MR (mean 14 days, range 1-30 days). Based on histopathological feature, 21 patients with well-differenitated liposarcoma and 40 patients with lipoma were diagnosed. Regarding mean, MPP, kurtosis, and entropy, values on all sequences were significantly higher in well-differentiated liposarcoma than those in lipoma (p<0.001). However, regarding skewness, values on all sequences were not different between well-differentiated liposarcoma and lipoma (p=0.15-0.74). The two best performing 3D-MRTA parameters were kurtosis on T2WI (AUC, 0.86; 95 % CI, 0.77-0.95) and entropy on FS-T1CE (AUC, 0.83; 95 % CI, 0.74- 0.93). There were good or excellent interobserver agreements for all measurements (ICC, 0.750–0.885).Conclusion
In conclusion, texture metrics in 3D-MRTA parameters were different between well-differentiated liposarcoma and lipoma. Among them, kurtosis on T2WI and entropy on FS-T1CE were most useful for differentiating well-differentiated liposarcoma and lipoma. Therefore, 3D-MRTA shows promise as a tool which may help increase radiological accuracy and confidence in the workup of well-differentiated liposarcoma.Acknowledgements
None.References
1. Murphey MD, Arcara LK, Fanburg-Smith J. From the archives of the AFIP: imaging of musculoskeletal liposarcoma with radiologic-pathologic correlation. Radiographics. 2005;25(5):1371-1395.
2. Brisson M, Kashima T, Delaney D, et al. MRI characteristics of lipoma and atypical lipomatous tumor/well-differentiated liposarcoma: retrospective comparison with histology and MDM2 gene amplification. Skeletal Radiol. 2013;42(5):635-647.
3. Juntu J, Sijbers J, De Backer S, Rajan J, Van Dyck D. Machine learning study of several classifiers trained with texture analysis features to differentiate benign from malignant soft-tissue tumors in T1-MRI images. J Magn Reson Imaging. 2010;31(3):680-689.
4. Robertson EG, Baxter G. Tumour seeding following percutaneous needle biopsy: the real story! Clin Radiol. 2011;66(11):1007-1014.
5. Skrzynski MC, Biermann JS, Montag A, Simon MA. Diagnostic accuracy and charge-savings of outpatient core needle biopsy compared with open biopsy of musculoskeletal tumors. J Bone Joint Surg Am. 1996;78(5):644-649.
6. Ganeshan B, Miles KA. Quantifying tumour heterogeneity with CT. Cancer Imaging. 2013;13:140-149.
7. Miles KA, Ganeshan B, Hayball MP. CT texture analysis using the filtration-histogram method: what do the measurements mean? Cancer Imaging. 2013;13(3):400-406.
Figures
