View Presentation Video

Malvika Viswanathan¹, Leqi Yin², Yashwant Kurmi¹, and Zhongliang Zu^3
1Vanderbilt University Medical Center, Vanderbilt University Institute of Imaging Sciences, Nashville, TN, United States, ²Vanderbilt University, Nashville, TN, United States, ³Department of Radiology and Radiological Sciences, Vanderbilt University Institute of Imaging Sciences, Nashville, TN, United States

Keywords: Machine Learning/Artificial Intelligence, CEST & MT

Machine learning is increasingly applied to address challenges in specifically quantifying APT effect. The models are usually trained on measured data, which, however are usually lack of ground truth and sufficient training data. Synthetically generated data from both measurements and simulations can create training data which mimic tissues better than full simulations, cover all possible variations in sample parameters, and provide the ground truth. We evaluated the feasibility to use synthetic data to train models for predicting APT effect. Results show that the machine learning predicted APT is more close to the ground truth than the conventional multiple-pool Lorentzian fit.

View Presentation Video

Dennis van de Sande¹, Sina Amirrajab¹, Mitko Veta¹, and Marcel Breeuwer^1,2
1Biomedical Engineering - Medical Image Analysis Group, Eindhoven University of Technology, Eindhoven, Netherlands, ²MR R&D - Clinical Science, Philips Healthcare, Best, Netherlands

Keywords: Machine Learning/Artificial Intelligence, Spectroscopy, deep learning, generative modelling

We propose a conditional VAE to synthesize single-voxel MRS data. This deep learning method can be used to enrich in-vivo datasets for other machine learning applications, without using any physics-based models. Our work is a proof-of-concept study which demonstrates the potential of a cVAE for MRS data generation by using a synthetic dataset of 8,000 spectra for training. We evaluate our model by performing a linear interpolation of the latent space, which shows that spectral properties are captured in the latent space, meaning that our model can learn spectral features from the data and can generate new samples.

View Presentation Video

Frank Zijlstra^1,2 and Peter T While^1,2
1Department of Radiology and Nuclear Medicine, St. Olav's University Hospital, Trondheim, Norway, ²Department of Circulation and Medical Imaging, NTNU - Norwegian University of Science and Technology, Trondheim, Norway

Keywords: Machine Learning/Artificial Intelligence, Diffusion/other diffusion imaging techniques, Model fitting, uncertainty, IVIM

Uncertainty is an important aspect of fitting quantitative models to diffusion MRI data, which is often overlooked. This study presents a method for estimating uncertainty intrinsic to a model using a generative deep learning approach (Denoising Diffusion Probabilistic Models (DDPM)). We numerically validate that the approach provides accurate uncertainty estimates, and demonstrate its use in providing signal-specific uncertainty estimates. Furthermore, we show that DDPM can be used as a fitting method that estimates uncertainty, and show both ADC and IVIM fitting on an in vivo brain scan. This shows promise for DDPM, as both an investigate tool and fitting method.

Caohui Duan¹, Xiangbing Bian¹, Kun Cheng¹, Jinhao Lyu¹, Yongqin Xiong¹, Jianxun Qu², Xin Zhou³, and Xin Lou^1
1Department of Radiology, Chinese PLA General Hospital, Beijing, China, ²MR Collaboration, Siemens Healthineers Ltd., Beijing, China, ³Key Laboratory of Magnetic Resonance in Biological Systems, State Key Laboratory of Magnetic Resonance and Atomic and Molecular Physics, National Center for Magnetic Resonance in Wuhan, Wuhan Institute of Physics and Mathematics, Innovation Academy for Precision Measurement Science and Technology, Chinese Academy of Sciences‒Wuhan National Laboratory for Optoelectronics, Wuhan, China

Keywords: Machine Learning/Artificial Intelligence, Brain

Ultra-high field 7T MRI provides exceptional tissue contrast and anatomical details but is often cost-prohibitive and not widely accessible in clinics. A generative adversarial network (SynGAN) was developed to generate synthetic 7T images from the widely used 3T images. The synthetic 7T images achieved improved tissue contrast and anatomical details compared to the 3T images. Meanwhile, the synthetic 7T images showed comparable diagnostic performance to the authentic 7T images for visualizing a wide range of pathology, including cerebral infarction, demyelination, and brain tumor.

View Presentation Video

Matteo Mancini¹, Carolyn McNabb¹, Derek Jones¹, and Mara Cercignani^1
1Cardiff University Brain Research Imaging Centre, Cardiff University, Cardiff, United Kingdom

Keywords: Machine Learning/Artificial Intelligence, Brain

Myelin biomarkers are a fundamental tool for both neuroscience research and clinical applications. Despite several quantitative MRI methods available for their estimation, in several cases qualitative approaches are the only viable solution. To get the best of both the quantitative and qualitative worlds, here we propose an image-to-image translation method to learn the mapping between common routine scans and a quantitative myelin metric. To achieve this goal, we trained a generative adversarial network on a relatively large dataset of healthy subjects. Both the qualitative and quantitative results show good agreement between the predicted and the ground-truth maps.

Sandeep Kaushik^1,2, Cristina Cozzini¹, Mikael Bylund³, Steven Petit⁴, Bjoern Menze², and Florian Wiesinger^1
1GE Healthcare, Munich, Germany, ²Department of Quantitative Biomedicine, University of Zurich, Zurich, Switzerland, ³Umeå University, Umeå, Sweden, ⁴Erasmus MC Cancer Institute, Rotterdam, Netherlands

Keywords: Machine Learning/Artificial Intelligence, Radiotherapy, MR-only RT, Synthetic CT, Multi-task CNN, PET/MR

Many recent works have proposed methods to convert MRI into synthetic CT (sCT). While they have demonstrated a certain level of accuracy, not many have studied the robustness of those methods. In this work, we study the robustness of a multi-task deep learning (DL) model that computes sCT images from fast ZTE MR images under different levels of image noise. We evaluate its impact on radiation therapy planning. The proposed method  demonstrates  resilience against input noise variations. It  makes way for a clinically acceptable dose calculation with a fast input image acquisition. 

View Presentation Video

Matteo Ferrante¹, Marianna Inglese², Ludovica Brusaferri³, Marco L Loggia³, and Nicola Toschi^4,5
1Biomedicine and prevention, University of Rome Tor Vergata, Roma, Italy, ²Biomedicine and prevention, University of Rome Tor Vergata, Rome, Italy, ³Martinos Center For Biomedical Imaging, MGH and Harvard Medical School (USA), Boston, MA, United States, ⁴BioMedicine and prevention, University of Rome Tor Vergata, Rome, Italy, ⁵Department of Radiology,, Athinoula A. Martinos Center for Biomedical Imaging and Harvard Medical school, Boston, MA, USA, Boston, MA, United States

Keywords: Machine Learning/Artificial Intelligence, Neuroinflammation, PET, image synthesis

Chronic pain-related biomarkers can be found using a specific binding radiotracer called [11C]PBR28 able to target the translocator protein (TSPO), whose expression is increased in activated glia and can be considered as a biomarker for neuroinflammation. One of the main drawbacks of PET imaging is radiation exposure, for which we attempted to develop a deep learning model able to synthesize PET images of the brain from T1w MRI  only. Our model produces synthetic TSPO-PET images from T1W MRI which are statistically indistinguishable from the original PET images both on a voxel-wise and on a ROI-wise level.

View Presentation Video

Alexandros Patsanis¹, Mohammed R. S. Sunoqrot^1,2, Elise Sandsmark², Sverre Langørgen², Helena Bertilsson^3,4, Kirsten Margrete Selnæs^1,2, Hao Wang⁵, Tone Frost Bathen^1,2, and Mattijs Elschot^1,2
1Department of Circulation and Medical Imaging, Norwegian University of Science and Technology - NTNU, Trondheim, Norway, ²Department of Radiology and Nuclear Medicine, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway, ³Department of Clinical and Molecular Medicine, Norwegian University of Science and Technology - NTNU, Trondheim, Norway, ⁴Department of Urology, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway, ⁵Department of Computer Science, Norwegian University of Science and Technology - NTNU, Gjøvik, Norway

Keywords: Machine Learning/Artificial Intelligence, Data Analysis, Deep Learning

This study investigated automated detection and localization of prostate cancer on biparametric MRI (bpMRI). Conditional Generative Adversarial Networks (GANs) were used for image-to-image translation. We used an in-house collected dataset of 811 patients with T2- and diffusion-weighted MR images for training, validation, and testing of two different bpMRI models in comparison to three single modality models (T2-weighted, ADC, high b-value diffusion). The bpMRI models outperformed T2-weighted and high b-value models, but not ADC. GANs show promise for detecting and localizing prostate cancer on MRI, but further research is needed to improve stability, performance and generalizability of the bpMRI models.

View Presentation Video

Saba Sadatamin^1,2, Steven Robbins³, Richard Tyc³, Adam C. Waspe^2,4, Lueder A. Kahrs^1,5,6,7, and James M. Drake^1,2,8
1Institute of Biomedical Engineering, University of Toronto, Toronto, ON, Canada, ²Posluns Centre for Image Guided Innovation & Therapeutic Intervention, Hospital of Sick Children, Toronto, ON, Canada, ³Monteris Medical, Winnipeg, MB, Canada, ⁴Department of Medical Imaging, University of Toronto, Toronto, ON, Canada, ⁵Medical Computer Vision and Robotics Lab, University of Toronto, Toronto, ON, Canada, ⁶Department of Mathematical & Computational Sciences, University of Toronto Mississauga, Toronto, ON, Canada, ⁷Department of Computer Science, University of Toronto, Toronto, ON, Canada, ⁸Department of Surgery, University of Toronto, Toronto, ON, Canada

Keywords: Machine Learning/Artificial Intelligence, MR-Guided Interventions, Laser Interstitial Thermal Therapy

MR-guided laser interstitial thermal therapy (MRgLITT) is a minimally-invasive treatment for brain tumors where a surgeon inserts a laser fiber along a fixed trajectory.  Repositioning the laser is invasive and predicting thermal spread close to heat sinks is difficult. To address this problem, MR thermometry prediction using artificial intelligence (AI) modeling will be developed to aid the surgeon to determine whether a selected laser position is ideal before the treatment starts. AI algorithms will be trained to model the nonlinear mapping from anatomical MRI planning images to MR thermometry. A surgeon will choose a better fiber trajectory by AI model.

View Presentation Video

Sukru Samet Dindar¹, Buse Buz-Yalug², Kubra Tan³, Ayca Ersen Danyeli^4,5, Ozge Can^5,6, Necmettin Pamir^5,7, Alp Dincer^5,8, Koray Ozduman^5,7, Yasemin P. Kahya¹, and Esin Ozturk-Isik^2,5
1Electrical and Electronics Engineering, Bogazici University, Istanbul, Turkey, ²Institute of Biomedical Engineering, Bogazici University, Istanbul, Turkey, ³Health Institutes of Turkey, Istanbul, Turkey, ⁴Department of Medical Pathology, Acibadem University, Istanbul, Turkey, ⁵Brain Tumor Research Group, Acibadem University, Istanbul, Turkey, ⁶Department of Medical Engineering, Acibadem University, Istanbul, Turkey, ⁷Department of Neurosurgery, Acibadem University, Istanbul, Turkey, ⁸Department of Radiology, Acibadem University, Istanbul, Turkey

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

Neurofibromatosis type 2 (NF2) gene mutations have been linked to tumorigenesis in meningiomas. This study aims to improve the prediction of NF2 loss in meningiomas using T1-weighted contrast-enhanced MRI augmented by a deep convolutional generative adversarial network (DCGAN). Synthetically generated MRI increased the training accuracy from 78.9% to 93% and test accuracy from 69.4% to 79.5% in this study.

Sandeep Kaushik^1,2, Cristina Cozzini¹, Jonathan J Wyatt³, Hazel McCallum³, Ross Maxwell⁴, Bjoern Menze², and Florian Wiesinger^1
1GE Healthcare, Munich, Germany, ²Department of Quantitative Biomedicine, University of Zurich, Zurich, Switzerland, ³Newcastle University and Northern Centre for Cancer Care, Newcastle upon Tyne, United Kingdom, ⁴Newcastle University, Newcastle upon Tyne, United Kingdom

Keywords: Machine Learning/Artificial Intelligence, Radiotherapy, Synthetic CT, radiation therapy planning, Multi-task network, deep learning

MRI to synthetic CT image conversion is a problem of interest for clinical applications such as MR-radiation therapy planning, PET/MR attenuation correction, MR bone imaging. Many methods proposed for this purpose use different MR inputs. In this work, we compare the sCT generated from different 3D MR inputs, including Zero TE (ZTE), fast spin echo (CUBE), and fast spoiled gradient echo with Dixon-type fat-water separation (LAVA-Flex), using a multi-task deep learning (DL) model. We analyze the qualitative and quantitative accuracy of the generated sCT image from each input and highlight the aspects relevant for different clinical applications.

View Presentation Video

Grace Wen¹, Jake McNaughton¹, Ben Chong¹, Vickie Shim¹, Justin Fernandez¹, Samantha Holdsworth^2,3, and Alan Wang^1,2,3
1Auckland Bioengineering Institute, University of Auckland, Auckland, New Zealand, ²Faculty of Medical and Health Sciences, University of Auckland, Auckland, New Zealand, ³Mātai Medical Research Institute, Tairāwhiti-Gisborne, New Zealand

Keywords: Machine Learning/Artificial Intelligence, Stroke, Image Synthesis

MRI holds an important role in diagnosing brain conditions, however, many patients do not receive an MRI before their diagnosis and onset of treatment. We propose to use deep learning to generate an MRI from a patient's CT and have implemented multiple models to compare their results. Using CT/MRI pairs from 181 stroke patients, we use mutiple deep learning models to generate MRI from the CT images. The model produces high quality images and accurately translates lesions onto the target image.

View Presentation Video

Lu Wang¹, Pujie Zhang¹, Zhen Xing², Congbo Cai¹, Zhong Chen¹, Dairong Cao², Zhigang Wu³, and Shuhui Cai^1
1Department of Electronic Science, Xiamen University, Xiamen, China, ²Department of Radiology, First Affiliated Hospital of Fujian Medical University, Fuzhou, China, ³MSC Clinical & Technical Solutions, Philips Healthcare, Shenzhen, China

Keywords: Machine Learning/Artificial Intelligence, Brain, Hemoperfusion parameter estimation

Hemoperfusion magnetic resonance (MR) imaging derived parameters characterize both endothelial hyperplasia and neovascularization that are associated with tumor aggressiveness and growth. However, the hemoperfusion parameter estimation is still limited by low reliability, high bias, long processing time, and operator experience dependency up to now. In this study, a synthetic data driven learning method for hemoperfusion parameter estimation is proposed. Image analysis shows that the proposed method improves the reliability and precision of hemodynamic parameter estimation in a full-automatic and high-efficient way.

View Presentation Video

Xiao Liang¹, Ti Bai¹, Andrew Godley¹, Chenyang Shen¹, Junjie Wu¹, Boyu Meng¹, Mu-Han Lin¹, Paul Medin¹, Yulong Yang¹, Steve Jiang¹, and Jie Deng^1
1Radiation Oncology, University of Texas Southwestern Medical Center, Dallas, TX, United States

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

Synthetic CT (sCT) image generated from MRI by unsupervised deep learning models tends to have large errors around bone area. To generate better sCT image quality in bone area, we propose to add bony structure constrains in the loss function of the unsupervised CycleGAN model, and modify the single-channel CycleGAN to a multi-channel CycleGAN that takes Dixon constructed MR images as inputs. The proposed model has lowest mean absolute error compared with single-channel CycleGAN with different MRI images as input. We found that it can generate more accurate Hounsfield Unit and anatomy of bone in sCT.

View Presentation Video

Fernando Vega^1,2,3, Abdoljalil Addeh^1,2,3, Ahmed Elmenshawi², and M. Ethan MacDonald^1,2,3,4
1Biomedical Engineering, Schulich School of Engineering, University of Calgary, Calgary, AB, Canada, ²Electrical & Software Engineering, Schulich School of Engineering, University of Calgary, Calgary, AB, Canada, ³Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada, ⁴Department of Radiology, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada

Keywords: Machine Learning/Artificial Intelligence, Alzheimer's Disease, MRI, PET, Image Translation

In this work, an image translation model is implemented to produce synthetic amyloid-beta PET images from structural MRI that are quantitatively accurate. Image pairs of amyloid-beta PET and structural MRI were used to train the model. We found that the synthetic PET images could be produced with a high degree of similarity to truth in terms of shape, contrast  and overall high SSIM and PSNR.  This work demonstrates that performing structural to quantitative image translation is feasible to enable the access amyloid-beta information from only MRI.

View Presentation Video

Daniel Kim¹, Jae-Hun Lee¹, Mohammed A. Al-masni², Jun-ho Kim¹, Yoonseok Choi¹, Eun-Gyu Ha¹, SunYoung Jung³, Young Noh⁴, and Dong-Hyun Kim^1
1Department of Electrical and Electronic Engineering, Yonsei Univ., Seoul, Korea, Republic of, ²Department of Artificial Intelligence, Sejong Univ., Seoul, Korea, Republic of, ³Department of Biomedical Engineering, Yonsei Univ., Wonju, Korea, Republic of, ⁴Department of Neurology, Gachon Univ., Incheon, Korea, Republic of

Keywords: Machine Learning/Artificial Intelligence, Data Processing, Augmentation

Recently, research on the detection of cerebral small vessel disease (CSVD) has been mainly implemented in two-stages (1^st: candidate detection, 2^nd: false-positive reduction). Previous studies presented the difficulty of collecting labeled data as a limitation. Here, we synthesized the lesion through 3D-DCGAN and insert it at different locations on the MR image considering anatomical localization and alpha blending to augment labeled data. Through this, the detecting architecture was simplified to a single-stage, and the precision and recall values were improved by an average of 0.2.

View Presentation Video

Ramtin Babaeipour¹, Maria Mihele², Keeirah Raguram², Matthew Fox^2,3, and Alexei Ouriadov^1,2,3
1School of Biomedical Engineering, The University of Western Ontario, London, ON, Canada, ²Department of Physics and Astronomy, The University of Western Ontario, London, ON, Canada, ³Lawson Health Research Institute, London, ON, Canada

Keywords: Machine Learning/Artificial Intelligence, Segmentation, Deep learning, Transfer learning

Hyperpolarized ¹²⁹Xe lung MRI is an efficient technique used to investigate and assess pulmonary diseases.  However, the longitudinal observation of the emphysema progression using hyperpolarized gas MRI-based ADC can be problematic, as the disease-progression can lead to increasing unventilated-lung areas, which likely excludes the largest ADC estimates.  One solution to this problem is to combine static-ventilation and ADC measurements following the idea of ³He MRI ventilatory ADC (vADC).  We have demonstrated this method adapted for ¹²⁹Xe MRI to help overcome the above-mentioned shortcomings and provide an accurate assessment of the emphysema progression.