Jiejie Zhou¹, Yang Zhang², Kai-Ting Chang³, Kyoung Eun Lee⁴, Ouchen Wang¹, Jiance Li¹, Yezhi Lin⁵, Zhifang Pan⁵, Peter Chang³, Daniel Chow³, Meihao Wang¹, and Min-Ying Su^3
1First Affiliate Hospital of Wenzhou Medical University, Wenzhou, China, ²University of California, Irvine, CA, United States, ³University of California, Irvine, Irvine, CA, United States, ⁴Inje University Seoul Paik Hospital, Seoul, Korea, Republic of, ⁵Wenzhou Medical University, Wenzhou, China

A total of 91 malignant/62 benign lesions were used for training, and 48 malignant/26 benign lesions for independent testing. Deep learning with ResNet50 were performed for differential diagnosis. To investigate the contribution of peri-tumor tissue, the tumor alone, smallest bounding box, and 1.2, 1.5, 2.0 times enlarged boxes were used as inputs. For per-lesion diagnosis, The accuracy was 91% for smallest bounding box, 84% for tumor alone and 1.2 times box, and further to 73% for 1.5 times box and 69% for 2.0 times box. In the independent testing dataset, the highest accuracy was 89% for the smallest bounding box.
 

Nan Wang^1,2, Yibin Xie¹, Lixia Wang^1,3, Sen Ma^1,2, Stephen L. Shiao⁴, Anthony G. Christodoulou¹, and Debiao Li^1,2
1Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, CA, United States, ²Department of Bioengineering, University of California, Los Angeles, Los Angeles, CA, United States, ³Beijing Chaoyang Hospital, Beijing, China, ⁴Radiation Oncology, Cedars-Sinai Medical Center, Los Angeles, CA, United States

DCE MRI is a well-accepted tool in the management of breast cancer, but continues to face technical challenges and concerns regarding gadolinium deposition. In this work, we proposed a novel Multitasking DCE technique, which enables adequate breast coverage, 0.9-mm isotropic spatial resolution, 1.5-s temporal resolution, dynamic T1 mapping throughout all DCE phases, and reduced dose of 0.02mmol/kg for the imaging of breast cancer. The in vivo studies demonstrated that the low-dose Multitasking DCE showed equivalent tumor delineation compared to standard DCE. The quantitative DCE parameters were repeatable in vivo and significantly different between normal breast and breast cancer. 

Nan Meng¹, Xuejia Wang², Dongming Han², Jing Sun³, Wenling Liu², Kaiyu Wang⁴, and Meiyun Wang*^5
1Department of Radiology, Zhengzhou University People’s Hospital & Henan Provincial People’s Hospital, Academy of Medical Sciences, Zhengzhou University, Zhengzhou, China, ²Department of MRI, the First Affiliated Hospital of Xinxiang Medical University, Weihui, China, ³Department of Pediatrics, Zhengzhou Central Hospital, Zhengzhou University, Zhengzhou, China, ⁴GE Healthcare, MR Research China, Beijing, China, ⁵Department of Radiology, Zhengzhou University People’s Hospital & Henan Provincial People’s Hospital, Zhengzhou, China

Amide proton transfer-weighted imaging (APTWI) has unique advantages in displaying the metabolism of diseased proteins. Diffusion kurtosis imaging (DKI) can quantify the diffusion state water molecules in tissues with a non-gaussian model, thus correcting the deviation of the DWI model and improving the detection of lesions. Our results show that compared with APTWI, the DKI is more effective in the diagnosis of benign and malignant breast tumors.

Shizhan Gong¹, Matthew Muckley¹, Nan Wu¹, Taro Makino¹, Gene Kim¹, Laura Heacock¹, Linda Moy¹, Florian Knoll¹, and Krzysztof Geras^1
1New York University, New York, NY, United States

In this paper we trained an end-to-end classifier using a deep convolutional neural network on a large data set of 8632 3D MR exams. Our model can achieve an AUC of 0.8486 in identifying malignant cases on a test set reflecting the full spectrum of the patients who undergo the breast MRI examination. We studied the effect of the data set size and the effect of using different T1-weighted images in the series on the performance of our model. This work will serve as a guideline for optimizing future deep neural networks for breast MRI interpretation.

Weibo Gao¹, Quanxin Yang¹, Xin Chen¹, Xiaocheng Wei², Qiujuan Zhang¹, Honghong Sun¹, Xiaohui Li¹, Lin Wang¹, Xiao na Zhang¹, Baobin Guo¹, Ali Shang¹, and Xiao xia Lu^1
1The second affiliated hospital of xi 'an jiaotong university, xi an, China, ²MR Research, GE Healthcare, Beijing, China

In this study, we aim to investigate the feasibility of quantitative measurements obtained from magnetic resonance image compilation (MAGIC) MRI technology in the breast lesions and to further evaluate the application value of quantitative measurements in differentiating malignant from benign breast lesions. It was concluded that quantitative T2 relaxation time and PD value measured by MAGIC-MRI sequence can be applied in the breast lesions. The measured lower quantitative T2 and PD value is closely related to breast malignancy, which worth further study. 

Kango Kawase¹, Masako Y Kataoka², Tomohiro Takemura¹, Takuto Fukutome¹, Kojiro Yano³, Maya Honda², Mami Iima², Dominik Marcel Nickel⁴, Tatsuki Kataoka⁵, Masakazu Toi⁶, and Kaori Togashi^2
1Faculty of Medicine, Kyoto University, Kyoto, Japan, ²Department of Diagnostic Imaging and Nuclear Medicine,, Kyoto University Graduate School of Medicine, Kyoto, Japan, ³Osaka Institute of Technology, Osaka, Japan, ⁴Siemens Healthcare GmbH, Erlangen, Germany, ⁵Department of Diagnostic Pathology, Kyoto University Graduate School of Medicine, Kyoto, Japan, ⁶Department of Breast Surgery, Kyoto University, Kyoto, Japan

For quantitative evaluation of tumor-related vessels, convolution filter processing was applied to ultrafast dynamic contrast-enhanced (UF-DCE) MRI of 51 lesions to obtain vessel length and crossing count as quantitative markers. The current analysis showed that these imaging markers were associated with a subtype of invasive breast cancer, the Ki-67 index, among invasive breast cancers. This opens a new approach to the evaluation of tumor-related vessels and tumor microenvironment.

Muge Karaman^1,2, Shunan Che³, Rahul Mehta^1,2, Guangyu Dan^1,2, Zheng Zhong^1,2, Han Ouyang³, X. Joe Zhou^1,4, and Xinming Zhao^3
1Center for Magnetic Resonance Research, University of Illinois at Chicago, Chicago, IL, United States, ²Department of Bioengineering, University of Illinois at Chicago, Chicago, IL, United States, ³Department of Radiology, Cancer Hospital, Chinese Academy of Medical Sciences, Beijing, China, ⁴Departments of Radiology and Neurosurgery, University of Illinois at Chicago, Chicago, IL, United States

An early imaging assessment of breast cancer’s response to neoadjuvant chemotherapy (NAC) is critical for timely planning of treatment strategies. In this study, we develop a machine-learning-based approach to investigate whether the combined features obtained from the intravoxel incoherent motion and continuous-time random-walk diffusion models provide an early prediction of pathologic response in patients receiving NAC. Our results have shown that a gradient boosting classifier trained with the early-treatment parametric changes within tumor can predict the response with an accuracy that is 96% of the accuracy achieved by using the post-treatment parametric changes.

Gabrielle Baxter¹, Andrew J Patterson², Leonardo Rundo¹, Ramona Woitek¹, Reem Bedair², Julia Carmona-Bozo¹, Roido Manavaki¹, Mary A McLean³, Scott A Reid⁴, Martin J Graves², and Fiona J Gilbert^1
1Department of Radiology, University of Cambridge, Cambridge, United Kingdom, ²Department of Radiology, Addenbrooke's Hospital, Cambridge, United Kingdom, ³Cancer Research UK, Cambridge, United Kingdom, ⁴GE Healthcare, Amersham, United Kingdom

This study investigated the prediction of pathological complete response (pCR) to neoadjuvant chemotherapy in breast cancer using radiomics features derived from pre-treatment DCE-MRI. 121 women with biopsy-confirmed breast cancers (44 pCR and 77 non-pCR) were imaged before treatment. 384 radiomics features were extracted from 5 post-contrast images. A logistic regression model trained on 21 of these features was able to predict pCR with an AUC of 0.78. The highest AUC (0.85) was achieved by using 7 features from only the 3^rd post-contrast time point. Clinical and pathological features should be included to improve the accuracy of prediction.

Wen Li¹, Natsuko Onishi¹, Vignesh Arasu¹, David C. Newitt¹, Alex Nguyen¹, Jessica Gibbs¹, Lisa J. Wilmes¹, Ella F. Jones¹, John Kornak¹, Bonnie N. Joe¹, The I-SPY 2 Investigator Network², Laura J. Esserman¹, and Nola M. Hylton^1
1University of California, San Francisco, San Francisco, CA, United States, ²'Quantum Leap Healthcare Collaborative, San Francisco, CA, United States

This study tested the additive value of quantitative background parenchymal enhancement (BPE) assessed in the contralateral breast in the prediction of treatment response for patients with locally advanced breast cancer undergoing neoadjuvant chemotherapy (NAC). BPE predictors were added to the prediction models together with functional tumor volume predictors. Our results showed that combined model achieved better prediction for pathologic complete response in HER2-positive HR-negative cancer subtype after 3 weeks of NAC. The additive values of BPE were also observed at inter-regimen (12-week) for triple negatives, and at the pre-surgery for HR-positive subtypes.

Maya Honda¹, Masako Kataoka¹, Mami Iima¹, Kanae Kawai Miyake¹, Akane Ohashi¹, Ayami Ohno Kishimoto¹, Rie Ota¹, Marcel Dominik Nickel², Tatsuki Kataoka³, Masakazu Toi⁴, and Kaori Togashi^1
1Department of Diagnostic Imaging and Nuclear Medicine, Graduate School of Medicine, Kyoto University, Kyoto, Japan, ²MR Application Predevelopment, Siemens Healthcare GmbH, Erlangen, Germany, ³Department of Diagnostic pathology, Graduate School of Medicine, Kyoto University, Kyoto, Japan, ⁴Department of breast surgery, Graduate School of Medicine, Kyoto University, Kyoto, Japan

The study evaluated the accuracy to predict pathologic complete response (pCR) after neo-adjuvant chemotherapy (NAC) using ultrafast dynamic contrast-enhanced (UF-DCE) MRI. The sensitivity to predict pCR was higher on UF-DCE MRI compared with conventional dynamic contrast-enhanced (DCE) MRI. The difference in image and pathological sizes on UF-DCE MRI was smaller than on conventional DCE MRI. UF-DCE MRI potentially assesses post-NAC status in breast cancer patients accurately in a shorter acquisition time.