Shuyang Zhang¹, Min Zhang², Xinhua Cao³, Geoffrey S Young², and Xiaoyin Xu^2
1University of Michigan-Ann Arbor, Ann Arbor, MI, United States, ²Brigham and Women's Hospital and Harvard Medical School, Boston, MA, United States, ³Boston Children’s Hospital, Boston, MA, United States

Cancer metastases to the brain is a major cause of fatality in patients. Finding all the metastases is crucial to clinical treatment planning as today’s radiation therapy can target up to 20 individual metastases, making it necessary for clinicians to detect and marking multiple metastases in practice. Detecting brain metastases, however, is very challenging because the objects are small and of low contrast. Computer-aided detection of metastases can be highly valuable to improve the accuracy and efficiency of a human reader. In this work, we developed a deep learning-based pipeline for finding metastases on brain MRI.

Jhimli Mitra¹, Soumya Ghose¹, David Mills¹, Lowell Scott Smith¹, Sarah Frisken², Alexandra Golby², Thomas K. Foo¹, and Desmond Teck-Beng Yeo^1
1General Electric Research, Niskayuna, NY, United States, ²Brigham and Women's Hospital, Boston, MA, United States

Multimodality fusion in neurosurgical guidance aids neurosurgeons in making critical clinical decisions regarding safe maximal resection of tumors. It is challenging to have registration methods that automatically update pre-surgical MRI on intra-operative ultrasound, adjusting for the brain-shift for surgical guidance. A 3D deep learning-based convolutional network was developed for fast, multimodal alignment of pre-surgical MRI and intra-operative ultrasound volumes. The neural network is a combination of some well-known deep-learning architectures like FlowNet, Spatial Transformer Networks and UNet to achieve fast alignment of multimodal images. The CuRIOUS 2018 challenge training data was used to evaluate the accuracy of the developed method.

Tanay Chougule¹, Rakesh Gupta², Jitender Saini³, Shaleen Agarwal⁴, Rana Patir⁴, and Madhura Ingalhalikar^1
1Symbiosis Center for Medical Image Analysis, Symbiosis International University, Pune, India, ²Department of Radiology, Fortis Hospital, Gurgaon, India, ³Department of Radiology, National Institute of Mental Health and Neurosciences, Bangalore, India, ⁴Radiation Oncology and Neurosurgery, Fortis Hospital, Gurgaon, India

Standard post-operative radiation therapy in glioblastoma delivers radiation uniformly across the hyper-intense areas from pre-operative FLAIR images and does not account for the regions where the infiltration might relapse. This work creates a non-invasive prognostic signature of the extent of recurrent hyper-intense FLAIR using radiomics features extracted from multi-modal MRI. Results demonstrate that the area of recurrence can be accurately predicted earlier with some radiomic features as beacon of recurrence than others when tested temporally across multiple time-points.

Zijian Zhou¹, Jeremiah W. Sanders¹, Jason M. Johnson², Tina M. Briere³, Mark D. Pagel⁴, Jing Li⁵, and Jingfei Ma^1
1Imaging Physics, The University of Texas MD Anderson Cancer Center, Houston, TX, United States, ²Diagnostic Radiology, The University of Texas MD Anderson Cancer Center, Houston, TX, United States, ³Radiation Physics, The University of Texas MD Anderson Cancer Center, Houston, TX, United States, ⁴Cancer Systems Imaging, The University of Texas MD Anderson Cancer Center, Houston, TX, United States, ⁵Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, United States

Manual delineation of brain metastases for stereotactic radiosurgery (SRS) is time consuming and labor intensive. We successfully constructed a deep learning ensemble, including a single shot detector and U-Net, to detect and subsequently segment brain metastases in MRI for SRS treatment planning. Postcontrast 3D T1-weighted gradient echo MR images from 266 patients were randomly split by 212:54 for model training-validation and testing. For the testing group, an overall sensitivity of 80.4% (189/235 metastases) with 4 false positives per patient, and a median segmentation Dice of 77.9% (61.4% - 86.3%) for the detected metastases were achieved.

Banu Sacli-Bilmez¹, Zeynep Firat², Melih Topcuoglu², C. Kaan Yaltirik³, Uğur Türe³, and Esin Ozturk-Isik^1
1Institute of Biomedical Engineering, Bogazici University, Istanbul, Turkey, ²Department of Radiology, Yeditepe University, Istanbul, Turkey, ³Department of Neurosurgery, Yeditepe University, Istanbul, Turkey

Glioblastoma (GBM) is the most common primary brain tumor in adults with 15 months median overall survival. The purpose of this study was to identify overall survival of GBM patients based on clinical and Visually AcceSAble Rembrandt Images (VASARI) features using machine learning. According to our results, a support vector machine (SVM) model worked better for categorical data classification. With the help of adaptive synthetic (ADASYN) oversampling, a fine Gaussian SVM model identified short overall survival at 12 and 24 months thresholds with 99.78% and 88.80% accuracies, respectively.

Julia Cluceru^1,2, Paula Alcaide-Leon¹, Valentina Pedoia¹, Joanna Phillips³, Devika Nair¹, Yannet Interian⁴, Susan Chang⁵, Javier E. Villanueva-Meyer¹, Tracy Luks¹, Annette Molinaro⁵, Mitchel Berger⁵, and Janine Lupo^1,6
1Radiology and Biomedical Imaging, UCSF, San Francisco, CA, United States, ²Bioengineering and Therapeutic Sciences, UCSF, San Francisco, CA, United States, ³UCSF, Neurological Surgery, CA, United States, ⁴Data Science, USF, San Francisco, CA, United States, ⁵Neurological Surgery, UCSF, San Francisco, CA, United States, ⁶Graduate Program in Bioengineering, UCSF/UC Berkeley, San Francisco and Berkeley, CA, United States

In this study, we leverage a promising new centrally restricted diffusion pattern1 together with modern advances in deep learning to create a novel method for detecting treatment-related injury in the context of suspected recurrent glioblastoma. We report a 5-fold cross-validation average AUC ROC of 0.83 +/- 0.2 for the classification of lesions into two categories: those induced by treatment, and those that are true incidences of recurrent glioblastoma.

Shiteng Suo¹, Mengqiu Cao¹, Xiaoqing Wang¹, Wei Yang², Jianrong Xu¹, and Yan Zhou^1
1Renji Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai, China, ²Guangdong Provincial Key Laboratory of Medical Image Processing, School of Biomedical Engineering, Southern Medical University, Guangzhou, China

Preoperative noninvasive prediction of IDH mutation status is crucial for prognosis and therapeutic decision making. In this study, we evaluated the qualitative and quantitative MRI features, namely, Visually Accessible Rembrandt Images (VASARI) features and apparent diffusion coefficient radiomics features in identifying IDH1 mutation status in lower-grade gliomas (WHO grade II-III). Results by machine learning methods showed that the combination achieved a better prediction performance. Our model may have the potential to serve as an alternative to the conventional workflow for the noninvasive identification of the molecular profiles.

Julia Cluceru^1,2, Yannet Interian³, Joanna Phillips⁴, Devika Nair¹, Susan Chang⁴, Paula Alcaide-Leon¹, Javier E. Villanueva-Meyer¹, and Janine Lupo^1,5
1Radiology and Biomedical Imaging, UCSF, San Francisco, CA, United States, ²Bioengineering and Therapeutic Sciences, UCSF, San Francisco, CA, United States, ³Data Science, USF, San Francisco, CA, United States, ⁴UCSF, Neurological Surgery, CA, United States, ⁵Graduate Program in Bioengineering, UCSF/UC Berkeley, San Francisco and Berkeley, CA, United States

In this abstract, we use diffusion and anatomical MR imaging together with a pre-trained RGB ImageNet to classify patients into major genetic entities defined by the WHO. We achieved 91% accuracy on our validation set with high per-class accuracy, precision, and recall; and 81% accuracy on a separate test dataset.

Sohaib Naim¹, Chi Wah Wong², Eemon Tizpa¹, Hannah Jade Young¹, Kimberly Jane Bonjoc¹, Seth Michael Hilliard¹, Aleksandr Filippov ¹, Saman Tabassum Khan¹, Christine Brown³, Behnam Badie⁴, and Ammar Ahmed Chaudhry^1
1Diagnostic Radiology, City of Hope National Medical Center, Duarte, CA, United States, ²Applied AI and Data Science, City of Hope National Medical Center, Duarte, CA, United States, ³Hematology & Hematopoietic Cell Transplantation and Immuno-Oncology, City of Hope National Medical Center, Duarte, CA, United States, ⁴Surgery, City of Hope National Medical Center, Duarte, CA, United States

High grade gliomas (HGG) is the most common malignant primary brain tumors in adults. In this study, 61 patients with recurrent HGGs underwent surgical resection and chimeric antigen receptor-T cell therapy. Volumetric segmentations of contrast-enhanced (CE) and non-enhanced tumors (NET) using T1-weighted CE MR images were used to identify shape- and texture-based features from these regions of interest. We evaluated radiomic characteristics of these HGGs to determine novel imaging biomarkers to predict treatment response. Exponentially-filtered textural radiomic features based on Neighboring Gray Tone Difference Matrix and Gray Level Co-occurrence Matrix derived from NET were the strongest predictors of overall survival.