Introduction

Tumor-related vessels in dynamic contrast-enhanced (DCE) MRI are known to be associated with ipsilateral malignant breast tumors and considered to reflect neovascularization [1,2]. Image-derived information of these tumor-related vessels can be a biomarker of tumor growth. However, most of such trials depend on visual evaluation, and only a few studies have employed a quantitative approach [2]. Ultrafast dynamic contrast-enhanced (UF-DCE)-MRI of the breast enables clear visualization of the feeding artery and drainage veins of the breast tumor [3]. Image processing may help to quantify smaller tumor-related abnormal vessels. In this study we aimed to
1) Evaluate the effect of convolution filtering tailored for 3D-voxels, on tumor-related vessels.
2) Obtain the quantitative value of the vessel features including vessel length and complexity through skeletonization and crossing counting, and
3) Investigate the potential of these values in predicting a pathological diagnosis or proliferative biomarkers, including Ki-67 index, among breast cancers.

Methods

Our study population consisted of female patients suspected of breast cancer who underwent breast MRI with UF-DCE-MRI from December 2015 to March 2018. Those with poor image quality and without any enhancing lesion were excluded. A total of 51 lesions (46 malignant:5 benign) from 46 patients were analyzed. MR images were obtained with 3T MR units (MAGNETOM Prisma/Skyra, Siemens Healthcare, Erlangen, Germany). Gadobutrol/Gadoteridol was intravenously infused (2.0ml/sec). UF-DCE-MRI (15sec before 60sec after the contrast injection, 3.7sec×20frames) was followed by standard DCE-MRI. UF-DCE-MRI was acquired by a prototype based on the 3D gradient-echo VIBE sequence using a CS reconstruction (TR/TE 5.0/2.5ms, FA 15deg, FOV 360×360mm2, matrix 384×269, thickness 2.5mm, CS acceleration=16.5) with 30 iterations.
In-house image analysis included pre-processing using convolution filtering with an additional maximum intensity projection (MIP). Convolution filtering, an established 2D image technology, was applied to the 3D voxels. In order to make the most of the 3D data, we developed an 3D-convolution filter that utilizes neighboring pixels not only in the given slice but also in the adjacent slices. An extended Gaussian filter (reducing noise) and repeated edge enhancement filters (detecting subtle structures) were used. Finally, a MIP was used to collapse the voxel into 2D. The frame just after the feeding arteries started to enhance was selected. The tumor and vessels were separately segmented by three readers, supervised by an experienced breast radiologist. A segmented image was trimmed to a 128x128-pixel square containing tumor-related vessels. Segmented vessels were skeletonized to minimize the effect of vessel thickness. The pixel count of skeletonized vessels was used to represent total vessel length. Then, crossing points of these skeletonized vessels were identified and counted as a marker of vessel network complexity (fig.1). These parameters (skeletonized vessel length, number of crossing points) and their association with pathology/Ki-67 index were analyzed. Invasive cancers were grouped into high (≥14) and low (<14) Ki-67 index groups. ROC analysis was performed to examine the performance of vessel length and number of crossing point in diagnosing cancer with high Ki-67 index.

Results

Impact of convolution filtering: Tumor-related vessels become more visible, with mean total vessel length increased from 179 to 269 pixels (Wilcoxon signed-rank test: p<0.001) after filtering. The mean number of crossing points also increased from 5.7 to 10.4 after filtering (p<0.001).
Association with pathology and subtypes (fig.2): Among invasive breast cancers (n=37), non-luminal A type (n=19: luminal B, Triple negative, HER2+, and Luminal HER2+) showed significantly longer vessel length than luminal A type (n=18) before and after filtering (Wilcoxon rank-sum test: p<0.001 for both). The number of crossing points was significantly higher among non-luminal A type compared with those among luminal A type, which became more obvious after filtering (p=0.007 and p<0.001, respectively).
Association with Ki-67 index (fig.3): Vessel length was significantly longer among the high Ki-67 index group both before and after filtering (Wilcoxon rank-sum test: p<0.001). Crossing points were significantly increased among high Ki-67 index group, which became more obvious after filtering (p=0.003 and p<0.001, respectively).
In ROC analysis, AUC of vessel length and number of crossing points after filtering were 0.92/0.90 in predicting breast cancer with high Ki-67 index.

Discussion

The current study demonstrated the usefulness of convolution filtering to enhance detailed vascular structures, with significant increase in vessel length and crossing points. This preliminary comparison indicates that the enhanced detectability can be explained by the 3D convolution filter applied before the MIP, since any operation will fail to utilize the 3rd dimension once the images are collapsed into a MIP.
The two quantitative markers were associated with invasive cancer subtypes and the Ki-67 index, a marker of proliferation. ROC analysis indicated that these markers can be used to identify high Ki-67 index cancers without the need of tissue sampling.
Tumor-related vessels are tortuous, saccular, chaotic, and leaky [4]. These vascular features reflect the tumor microenvironment. Our proposed image processing and quantification can be used to evaluate and monitor these abnormal tumor-related vessels noninvasively (see fig.4).
Limitations include retrospective design, small sample size, and manual segmentation of the vessels.

Conclusion

With the use of convolution filter technology, tumor-related vessels on UF-DCE MRI can be quantitatively evaluated by vessel length and the number of crossings. These markers are associated with tumor subtype and Ki-67 index.

Acknowledgements

KAKENHI-JP15K09922