View Presentation Video

View Presentation Video

View Presentation Video

Kazufumi Kikuchi¹, Osamu Togao², Koji Yamashita³, Makoto Obara⁴, and Kousei Ishigami^1
1Department of Clinical Radiology, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan, ²Department of Molecular Imaging & Diagnosis, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan, ³Department of Radiology Informatics & Network, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan, ⁴Philips Japan, Tokyo, Japan

Keywords: Tumors, Machine Learning/Artificial Intelligence, Brain metastases

This study aimed to improve vessel visibility by modifying k-space filling and to verify the usefulness of volume isotropic simultaneous interleaved bright- and black-blood examination (VISIBLE) in detecting brain metastases using machine learning (ML). We tested three types of VISIBLE in different k-space fillings, and counted the number of vessels. We also tested the ML model by using VISIBLE. The number of vessels was lower in Centric and Reversed centric sequences than that in MPRAGE, but comparable in the Startup echo 30 sequence. Our ML model was achieved high sensitivity (97%) and there were no differences among three sequences.

View Presentation Video

Yoonseok Choi¹, Mohammed A Al-masni², Hyeok Park¹, Jun-ho Kim¹, Dong-Hyun Kim¹, and Roh-Eul Yoo^3
1Yonsei University, SEOUL, Korea, Republic of, ²Sejong Univiersity, Seoul, Korea, Republic of, ³Seoul National University Hospital, Seoul, Korea, Republic of

Keywords: Tumors, Brain

Accurately segmenting contrast-enhancing brain tumors plays an important role in surgical planning of high-grade gliomas. Also, precisely stratifying malignancy risk within non-enhancing T2 hyperintense area helps control the radiation dose according to the malignancy risk and prevent normal brain tissue from being unnecessarily exposed to radiation. In this work, we 1) segment brain tumors using deep learning, and 2) provide more detailed segmentation results that can show the malignancy risk within the T2 high region. We utilize a two-stage framework where we make images with restricted ROI through foreground cropping so that the model can focus on only tumor part.

View Presentation Video

Jiahong Ouyang¹, Kevin T. Chen², Jarrett Rosenberg¹, and Greg Zaharchuk^1
1Stanford University, Stanford, CA, United States, ²National Taiwan University, Taipei, Taiwan

Keywords: Machine Learning/Artificial Intelligence, PET/MR

PET is a widely used imaging technique, but it requires exposing subjects to radiation and is not offered in the majority of medical centers in the world. Here, we proposed to synthesize FDG-PET images from multi-contrast MR images using a U-Net-based network with attention modules and transformer blocks. The experiments on a dataset with 87 brain lesions in 59 patients demonstrated that the proposed method was able to generate high-quality PET from MR images without the need for radiotracer injection.  We also demonstrate methods to handle potential missing or corrupted sequences.

View Presentation Video

Sena Azamat^1,2, Buse Buz-Yaluğ¹, Alpay Ozcan³, Ayça Ersen Danyeli^4,5,6, Necmettin Pamir^5,7, Alp Dinçer^4,8, Koray Ozduman^4,7, and Esin Ozturk-Isik^1,4
1Institute of Biomedical Engineering, Bogazici University, Istanbul, Turkey, ²Department of Radiology, Basaksehir Cam and Sakura City Hospital, Istanbul, Turkey, ³Electric and Electronic Engineering Department, Bogazici University, Istanbul, Turkey, ⁴Brain Tumor Research Group, Acibadem University, Istanbul, Turkey, ⁵Center for Neuroradiological Applications and Reseach, Acibadem University, Istanbul, Turkey, ⁶Department of Medical Pathology, Acibadem University, Istanbul, Turkey, ⁷Department of Neurosurgery, Acibadem University, Istanbul, Turkey, ⁸Department of Radiology, Acibadem University, Istanbul, Turkey

Keywords: Tumors, Machine Learning/Artificial Intelligence

Meningiomas are the most common primary extra-axial intracranial tumors in adults. Grade of meningioma helps to predict the patient prognosis. Sixty-two patients with preoperative MRI were included in this IRB approved study. The whole tumor volumes were segmented from FLAIR, followed by co-registration onto SWI. A pretrained convolutional neural network (CNN) was employed to classify meningiomas into high and low-grades based on SWI, CE-T1W and FLAIR MRI. The pretrained CNN with data augmentation resulted in an accuracy of 80.2% (sensitivity=82.6% and specificity=78.1%) for identifying grades in meningiomas.

View Presentation Video

Darui Li¹, Wanjun Hu¹, Tiejun Gan¹, Guangyao Liu¹, Laiyang Ma¹, Kai Ai², and Jing Zhang^1
1Department of Magnetic Resonance, Lanzhou University Second Hospital, Lanzhou, China, ²Philips Healthcare, Xi'an, China

Keywords: Tumors, Machine Learning/Artificial Intelligence, Deep learning

The preoperative accurate and non-invasive prediction of glioma grading remains challenging. To accurately predict high-or low-grade gliomas, we constructed a 3D-ResNet101 deep learning model with data from a multicenter. These data were obtained from the Second Hospital of Lanzhou University, with 708 glioma patients, and the TCIA database, with 211 patients. The areas under the curve of the 3D-ResNet-101 deep learning model are 0.97 and 0.89 in the test cohort and external test cohort, respectively. This new method can be used for non-invasive prediction of glioma grading before surgery.

View Presentation Video

Abdullah Bas¹, Banu Sacli-Bilmez¹, Buse Buz-Yalug¹, Esra Sumer¹, Sena Azamat¹, Gokce Hale Hatay¹, Ayca Ersen Danyeli^2,3, Ozge Can^2,4, Koray Ozduman^2,5, Alp Dincer^2,6, and Esin Ozturk-Isik^1,2
1Institute of Biomedical Engineering, Bogazici University, Istanbul, Turkey, ²Brain Tumor Research Group, Acibadem University, Istanbul, Turkey, ³Department of Medical Pathology, Acibadem University, Istanbul, Turkey, ⁴Department of Biomedical Engineering, Acibadem University, Istanbul, Turkey, ⁵Department of Neurosurgery, Acibadem University, Istanbul, Turkey, ⁶Department of Radiology, Acıbadem University, Istanbul, Turkey

Keywords: Tumors, Machine Learning/Artificial Intelligence, Deep Learning

Intelligent Radiological Imaging Systems (IRIS)-DL is a deep learning software tool that includes libraries for segmenting tumor regions and identifying several genetic mutations in gliomas and meningiomas. The tool has three modules, which are “Model Library”, “Trainer”, and “Plotter”. In the “Model Library”, the users could run pre-trained models on their local data. The “Trainer” module is for creating custom AI (conventional machine learning, artificial neural networks, and deep learning) models on the user data. Lastly, “Plotter” module is for data visualization and explorative data analysis.  

View Presentation Video

Hongxi Yang¹, Ankang Gao², Yida Wang¹, Yong Zhang², Jingliang Cheng², Yang Song³, and Guang Yang^1
1Shanghai Key Laboratory of Magnetic Resonance, East China Normal University, Shanghai, China, ²Department of MRI, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, China, ³MR Scientific Marketing, Siemens Healthcare, Shanghai, China

Keywords: Tumors, Brain

We retrospectively enrolled 243 patients to develop a multi-task radiomics approach to predict IDH mutation status and early recurrence simultaneously in patients with WHO II-IV gliomas from preoperative multi-parametric MRI (mp-MRI). Firstly, multi-task LASSO was performed to find features shared between the two tasks, which were then combined with task-specific features selected recursively to build radiomics models. The models achieved test AUCs of 0.826 and 0.770 for IDH mutation status identification and early recurrence prediction, respectively. Shared features enabled the models to achieve satisfactory performance with minimum number of features, avoiding overfitting and making the models more interpretable.

View Presentation Video

Louis Gagnon¹, Diviya Gupta¹, Nathan S White², Vaness Goodwill³, Carrie McDonald⁴, Thomas Beaumont⁵, Tyler M Seibert^1,4,6, Jona Hattangadi-Gluth⁴, Santosh Kesari⁷, Jessica Schulte⁸, David Piccioni⁸, Anders M Dale^1,8, Nikdokht Farid¹, and Jeffrey D Rudie^1
1Department of Radiology, University of California San Diego, La Jolla, CA, United States, ²Cortechs.ai, San Diego, CA, United States, ³Department of Pathology, University of California San Diego, La Jolla, CA, United States, ⁴Department of Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, CA, United States, ⁵Department of Neurological Surgery, University of California San Diego, La Jolla, CA, United States, ⁶Department of Bioengineering, University of California San Diego, La Jolla, CA, United States, ⁷Department of Translational Neurosciences, Pacific Neuroscience Institute and Saint John’s Cancer Institute at Providence Saint Johns’ Health Center, Santa Monica, CA, United States, ⁸Department of Neuroscience, University of California San Diego, La Jolla, CA, United States

Keywords: Tumors, Diffusion/other diffusion imaging techniques

Differentiating recurrent tumor from post-treatment changes is challenging in glioblastoma. Using restriction spectrum imaging (RSI) and deep learning, we were able to accurately identify and segment residual and recurrent enhancing and non-enhancing cellular tumor in post-treatment brain MRIs. Including RSI in the deep learning model improved tumor segmentation due to the ability of RSI to separate cellular tumor from peritumoral edema and treatment related enhancement. The volume of cellular tumor was also predictive of survival. Our results suggest that combining deep learning and RSI may identify recurrent tumor in glioblastoma patients, which could improve targeted treatments and guide clinical decision-making. 

View Presentation Video

Keywords: Tumors, Radiotherapy

Using pre-radiotherapy anatomical, diffusion, and metabolic MRI from 42 patients newly-diagnosed with GBM, we first used Random Forest models to identify voxels that later exhibit either contrast-enhancing or T2 lesion progression. We then applied convolutional encoder-decoder neural networks to pre-radiotherapy imaging to segment subsequent tumor progression and found that the resulting predicted region better covered the actual tumor progression while sparing normal brain compared to the standard uniform 2cm expansion of the anatomical lesion to define the radiation target volume. This shows that multi-parametric MRI with deep learning has the potential to assist in future RT treatment planning.

View Presentation Video

Shraddha Pandey^1,2, Tugce Kutuk³, Matthew N Mills⁴, Mahmoud Abdalah⁵, Olya Stringfield⁵, Kujtim Latifi⁴, Wilfrido Moreno¹, Kamran Ahmed⁴, and Natarajan Raghunand^2
1Electrical Engineering, University of South Florida, Tampa, FL, United States, ²Department of Cancer Physiology, H.Lee Moffitt Cancer Center and Research Institute, Tampa, FL, United States, ³Department of Radiation Oncology, Baptist Health South Florida, Miami, FL, United States, ⁴Department of Radiation Oncology, H.Lee Moffitt Cancer Center and Research Institute, Tampa, FL, United States, ⁵Quantitative Imaging Shared Service, H.Lee Moffitt Cancer Center and Research Institute, Tampa, FL, United States

Keywords: Tumors, Radiotherapy, Image Prediction

Stereotactic radiosurgery (SRS) can provide effective local control of breast cancer metastases to the brain while limiting damage to surrounding healthy tissues. Knowledge-based algorithms have been reported that can alleviate the manual aspects of radiation dose planning, but these do not currently provide voxel-level dose prescriptions that are optimized for tumor control and avoidance of radionecrosis and associated toxicity. On the assumption that a voxelwise relationship exists between pre-SRS MR images, the RT dose map, and the resulting post-SRS MR images, we have investigated a deep learning framework to predict the latter from the former two.