Non-invasive characterization of BI-RADS 4/5 X-ray mammography findings using unenhanced breast MRI with fusion of diffusion weighted and morphological T2-weighted images
Sebastian Bickelhaupt1, Jana Tesdorff1, Frederik Bernd Laun2, Wolfgang Lederer3, Heidi Daniel4, Tristan Amseln Kuder2, Susanne Teiner3, Stefan Delorme2, and Heinz-Peter Schlemmer2

1Department of Radiology, German Cancer Research Center, Heidelberg, Germany, 2Department of Medical Physics in Radiology, German Cancer Research Center, Heidelberg, Germany, 3Radiological Practice at the ATOS Clinic Heidelberg, Heidelberg, Germany, 4Radiology Center Mannheim (RZM), Mannheim, Germany

Synopsis

Additive unenhanced breast MRI prior to biopsy might reduce false positive x-ray mammograms. Therefore we evaluated the adjunctive diagnostic value of unenhanced breast MRI with image fusion of DWI and T2-weighted images and compared it to a side-by-side analysis of the sequences. 30 patients were included undergoing a 1.5 T breast MRI examination prior biopsy. We found that reading DWI/T2w-fused images alone decreased the accuracy (81.6% ±2.8%) compared to the side-by-side analysis (91.6% ± 2.3%) and significantly reduced the reader confidence significantly (p<0.001). Therefore fused image series should currently be considered always in addition to morphologic images since they could miss important morphologic information.

Purpose:

Within X-ray breast cancer screening programs about 50% of all biopsies reveal a false positive finding in terms of a benign lesion (1). Diffusion weighted breast MRI (DWI-MRM) might allow for a more specific characterization of lesions found on X-ray mammogram prior to biopsy (2). With DWI, the background can be suppressed and diffusion-altered focal lesions or areas will stand out only. It further allows performing image fusion by combining DWI with the information of unenhanced morphological T2w-sequences, which is similar to PET-imaging (3). Image fusion thereby could combine the morphologic information (T2w) and the biophysiological information (DWI) in one postprocessed image sequence. We evaluated the diagnostic value of reading fused MR image series combining diffusion weighted imaging with background suppression (DWIBS) and morphological T2-weighted sequences for lesion characterization of suspicious BIRADS 4-5 lesion in X-ray mammograms prior to biopsy.

Methods:

This retrospective subgroup analysis was performed within an institutional and governmental review board approved prospective ongoing study. Written informed consent was obtained. 30 patients (mean age 58.9 years, SD ±6.8 years, range 50-69 years) with indication for biopsy due to a suspicious findings on screening X-ray mammograms (BIRADS 4-5) were included (December 2014 to February 2015). Preliminary results have been reported. Breast MRI was performed prior biopsy using a 1.5T MR scanner (Ingenia, Philips, Best, the Netherlands; 7-channel breast coil) in prone position. MRI sequences used for image fusion (as part of a full diagnostic protocol including contrast enhanced sequences) were T2w- sequences (transverse, TE 120 msec, TR 3800 msec, Slice thickness 3mm, FOV 340 x 280 mm, Sense x2, TurboFactor 20) and diffusion weighted imaging with background suppression (DWIBS; b-values b=0 s/mm² and b=1500 s/mm², transverse, TE 85 msec, TR 7300 msec, Slice thickness 3 mm, FOV 400 x 340 mm, STIR with Inversion Time 180 msec). Image fusion was performed using OsiriX (V.6.0, Switzerland). Distortion was corrected using point-by-point co-registration. Images were analyzed either using the fused image series or the conventional side-by-side analysis of DWI and T2-weighted images. Two blinded readers scored both image series using a Likert scale from “1= no suspicious lesion detected” to “5= highly suspicious lesion”; “0= no decision possible, additive information needed”. The Likert-like scales were dichotomized to enable analysis of diagnostic accuracies. A Likert-like score of 1, 2 or 3 was considered as “negative”; 0, 4 or 5 was considered as a “positive”. Diagnostic confidence was evaluated using a Likert scale from “0= no confidence” to “5 = recognition with high confidence”. Accuracy (positive predictive value [PPV], negative predictive value [NPV], sensitivity and specificity) was calculated and ANOVA-test was used to compare confidence (SigmaPlot, Version 12.5).

Results:

Biopsy revealed malignant lesions in 16/30 participants (53%) in histopathology. The overall mean accuracy achieved by using the fused image series was 81.6% (± 2.8%,R1 0.80%,R2 0.83%) with a sensitivity of 96.8% (± 4.4%,R2 100%,R1 93.7%) and a specificity of 64.2% (± 10.1%,R1 57.1%,R2 71.4%). In contrast the overall mean accuracy achieved by using the side-by-side analysis of DWI and T2-weighted series was 91.6% (± 2.3%,R1 0.90%,R2 0.93%) with a sensitivity of 97.7% (± 3.2%,R2 100%,R1 93.3%) and a specificity of 85.7% (± 0%, R1 85.7%, R2 85.7%) (Example Figure 1). Dichotomized overall interreader agreement for the fused image series was moderate with a kappa of 0.58 (SE 0.18) and good for the side-by-side analysis (kappa 0.63, SE 0.16). Confidence was significantly (p<0.001) higher in the side-by-side analysis (median 5, 25%-75% CI 4-5) compared to reading the fused series alone (median 4, 25-75% CI 3-4) (Figure 2).

Discussion:

Image fusion of diffusion weighted (DWI) and morphological T2-weighted images in breast imaging could combine morphological information and biophysiological properties of lesions displayed as one image/series (2, 3). Using fused image series as a stand-alone-tool for further characterization of previously known X-ray mammography BIRADS 4/5 lesions however reduced the accuracy and confidence of lesion characterization compared to reading DWI and T2-weighted images side-by-side. However, image fusion in unenhanced MRI protocols can represent an important tool linking morphological and biophysiological information which can add relevant information i.e. for breast biopsies or in other malignant gynecological entities such as for parametrial invasion in cancer of the cervix uteri (3).

Conclusion:

Abbreviated, contrast-agent-free breast MRI consisting of T2-weighted and DWI sequences can help to noninvasively characterize suspicious lesions found on x-ray mammography. However fused image series used for lesion characterization in breast MR should not be read alone but always in addition to morphologic images, since the overly might obscure subtle but diagnostically important morphological details.

Acknowledgements

This work was funded by the Dietmar-Hopp-Foundation.

References

1. Mammographie. K. Evaluationreport 2010. Results of the Mammography-Screening-Program in Germany. 2014.

2. Bickelhaupt S, Laun F, Tesdorff J, et al. Fast and non-invasive characterization of suspicious lesions detected on X-ray breast cancer screening– capability of diffusion weighted MRI with maximum intensity projections. Radiology. 2015 ;"DOI: 10.1148/radiol.2015150425".

3. Park JJ, Kim CK, Park SY, et al. Parametrial Invasion in Cervical Cancer: Fused T2-weighted Imaging and High-b-Value Diffusion-weighted Imaging with Background Body Signal Suppression at 3 T. Radiology. 2014:140920.

Figures

Figure 1: Example of image fusion using biophysiological Diffusion weighted Imaging (DWI, orange overlay) and morphological T2-weighted images. The lesion is demarcated in orange color (white arrow).

Figure 2: Plot demonstrating the different individual scores given for the reader confidence in each of the evaluated patients.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3049