View Presentation Video

Kai Zhao¹, Jiahao Lin¹, and Kyung Hyun Hyun Sung^1
1Department of Radiological Sciences, University of California, Los Angeles, Los Angeles, CA, United States

Keywords: Data Analysis, Prostate, super resolution

We proposed a generative models-based method for through-plane super-resolution of multi-slice MRI scans. We proposed a novel slice-profile transformation that synthesizes low/high through-plane resolution slices for model training without accessing to isotropic datasets. Expert ready study, visual and quantitative comparisons reveal our method produce superior super-resolution results in terms of sharpness, artifacts, noise level and overall image quality.

View Presentation Video

Melanie Leguizamon¹, Colin McKnight², Jarrod Eisma¹, Alex Song¹, Jason Elenberger¹, Daniel Claassen¹, Manus Donahue¹, and Kilian Hett^1
1Department of Neurology, Vanderbilt University Medical Center, Nashville, TN, United States, ²Department of Radiology, Vanderbilt University Medical Center, Nashville, TN, United States

Keywords: Data Analysis, Neurodegeneration

We apply novel deep learning algorithms using non-contrasted T₂-weighted MRI to test hypotheses regarding arachnoid granulation (AG) hypertrophy in patients with Parkinson’s disease (PD). Using this method, we identify AGs protruding into the superior sinus lumen, which may serve as a site of CSF egress implicated in the neurofluid clearance system. Results suggest a significant increase in AG volume in the parietal and frontal lobes of PD participants compared to age-matched healthy controls, potentially indicative of reduced neurofluid clearance efficiency secondary to macromolecular aggregation along the CSF circuit.

View Presentation Video

Zifei Liang¹, Patryk Filipiak¹, Steven H Baete¹, Yulin Ge¹, Leslie Ying², and Jiangyang Zhang^1
1Radiology, NYU Langone health, new york, NY, United States, ²the State University of New York, Buffalo, NY, United States

Keywords: Visualization, Brain Connectivity, Diffusion MRI tractography

Although diffusion MRI (dMRI) tractography can map brain connectivity non-invasively, accurate tractography in the human brain remains challenging due to inherent and technical limitations. In this study, we demonstrate a deep learning (DL) based approach for improving the estimation of fiber orientation distribution (FOD) from dMRI data. Trained with augmented whole brain tractography results from high-resolution dMRI data, the DL approach outperformed conventional FOD estimation methods in crossing fiber regions with dMRI data at spatial and angular resolutions comparable to routine clinical scans. The approach can potentially shorten the dMRI acquisition necessary for accurate tractography and connectome analysis.

Hongyi Chen¹, Xueling Liu², Yuxin Li^1,2, Puyeh Wu³, and Daoying Geng^1,2
1Academy for Engineering and Technology, Fudan University, Shanghai, China, ²Huashan Hospital, State Key Laboratory of Medical Neurobiology, Fudan University, Shanghai, China, ³GE Healthcare, Beijing, China

Keywords: Data Analysis, Parkinson's Disease

In this study, we constructed a hybrid machine learning model utilizing CNN and radiomics features based on NM-sensitive setMag images. The hybrid features improved the diagnostic performance in distinguishing PD patients from HC, as demonstrated in the SVM classifier, which demonstrated 95.7% accuracy, 92.9% sensitivity, and 100% specificity. The interpretability of the radiomics approach is better because radiomics features provide more interpretable biomarkers, while the CNN approach extracts deeper features from images. Furthermore, visualizing regions that influence classification decisions via saliency map can also enhance the interpretability of the CNN approach.

View Presentation Video

Hesam Abdolmotalleby¹ and Merry Mani^1
1University of Iowa, Iowa City, IA, United States

Keywords: Data Processing, Data Processing

We estimated biophysical model parameters using a dictionary of diffusion signals. The dictionary is generated based on the set of biophysical model parameters. We then rotated this dictionary along fascicle direction derived from spherical deconvolution and tried to fit it to the diffusion signal using non-negative linear least square optimization to find the associated dictionary weights. These weights went through a deep learning network to find model parameters. Our approach shows reasonable accuracy and reliability for biophysical model estimation. It is also possible to extend our approach to more complex and general biophysical models to achieve higher specificity.

View Presentation Video

Ajin Joy¹, Rajakumar Nagarajan¹, Eric Daar^2,3, Jhelum Paul¹, Santosh K Yadav⁴, Mario Guerrera², Paul M. Macey⁵, and M. Albert Thomas^1
1Radiological Sciences, University of California, Los Angeles, Los Angeles, CA, United States, ²Division of HIV Medicine, Lundquist Institute at Harbor-UCLA Medical Center, Los Angeles, CA, United States, ³Medicine, University of California, Los Angeles, Los Angeles, CA, United States, ⁴Department of Radiology and Radiological Science, School of Medicine,, Johns Hopkins University, Baltimore, MD, United States, ⁵School of Nursing and Brain Research Institute, University of California, Los Angeles, Los Angeles, CA, United States

Keywords: Data Processing, Brain

This study was designed to use the structural MRI data to characterize volumetric changes in gray matter and white matter, and to determine variations in cortical thickness, in a group of HIV-infected individuals to understand how their brain structural changes are associated with their neuropsychological state compared to a group of healthy individuals with similar social, behavioral backgrounds. Despite the demographic similarities and cART, we observed reduced gray and white matter volumes, as well as altered cortical thickness in HIV-infected participants compared with healthy controls. In addition, the neuroanatomic changes in HIV-infected patients showed statistically significant correlations with memory scores.

View Presentation Video

Kouhei Kamiya^1,2, Sayori Hanashiro³, Osamu Kano³, Koji Kamagata², Masaaki Hori^1,2, and Shigeki Aoki^2
1Department of Radiology, Toho University, Faculty of Medicine, Tokyo, Japan, ²Department of Radiology, Juntendo University, Faculty of Medicine, Tokyo, Japan, ³Department of Neurology, Toho University, Faculty of Medicine, Tokyo, Japan

Keywords: Data Processing, Diffusion/other diffusion imaging techniques, Image preprocessing

The impacts of different diffusion MRI preprocessing pipelines on the statistical results of clinical studies were investigated in amyotrophic lateral sclerosis. The results illustrate the influences of popular preprocessing tools on the effect size of the disease in DKI metrics.

View Presentation Video

Muhammad Asaduddin¹, Eung Yeop Kim², and Sung-Hong Park^1
1Bio and Brain Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Korea, Republic of, ²Department of Radiology, Samsung Medical Center, Sungkyunkwan University College of Medicine, Seoul, Korea, Republic of

Keywords: Data Processing, DSC & DCE Perfusion

Dynamic susceptibility contrast (DSC) MRI may suffer from artifacts due to long acquisition time. Past methods are limited in their performance and may change the contrast passage timing. In this work, we present a generative diffusion model that can restore signal loss and movement artifacts. We showed the generated DSC MRI images to have proper post-contrast vessel and grey matter structure with accurate contrast agent arrival/washout timing. The brain shape was also accurately generated as shown by the DICE score. This approach could provide a solution to restore a corrupted DSC MRI data while maintaining accurate contrast passage timing.

View Presentation Video

Tsuyoshi Ueyama^1,2,3, Erika Takahashi¹, Naoto Fujita¹, Yuichi Suzuki², Hideyuki Iwanaga², Osamu Abe⁴, and Yasuhiko Terada^1
1Graduate School of Science and Technology, University of Tsukuba, Tsukuba, Ibaraki, Japan, ²Radiology center, The University of Tokyo Hospital, Tokyo, Japan, ³School of Medicine, Stanford University, Palo Alto, CA, United States, ⁴Department of Radiology, The University of Tokyo Hospital, Tokyo, Japan

Keywords: Data Processing, Machine Learning/Artificial Intelligence, Diffusion weighted image/Diffusion tensor image

Although several end-to-end deep neural networks have been proposed to correct image distortion directly from distorted images, no study has verified the distortion correction performance for high b-values diffusion-weighted image (DWI) and diffusion tensor image (DTI) parameters. For example, the U-Net-based Synb0-DisCo was only validated for distortion correction of b0 images. Here, we used two networks, U-Net and Trans-DisCo, to verify distortion correction performance for DWIs and DTI parameter images. Trans-DisCo is our proposed model that replaces the convolutional neural network in U-Net with Swin Transformer, and we have shown that it outperforms U-Net.

View Presentation Video

Liad Doniza¹, Neta Stern², Dvir Radunsky², Coral Helft³, Patrick Fuchs⁴, Anita Karsa⁴, Karin Shmueli⁴, and Noam Ben-Eliezer^2,3,5
1Department of Electrical Engineering, Tel Aviv University, Tel Aviv, Israel, ²The Department of Biomedical Engineering, Tel-Aviv University, Tel-Aviv, Israel, ³Sagol School of Neuroscience, Tel-Aviv University, Tel-Aviv, Israel, ⁴Department of Medical Physics and Biomedical Engineering, University College London, London, United Kingdom, ⁵Center for Advanced Imaging Innovation and Research (CAI2R), New-York University Langone Medical Center, New York, NY, United States

Keywords: Data Processing, Susceptibility, Denoising

Quantitative susceptibility mapping (QSM) has many clinical applications such as distinguishing between acute and chronic multiple sclerosis (MS) lesions, and probing microbleeds in traumatic brain injury. High scan resolutions improve diagnostic quality and reduce partial volume artifacts albeit at a price of a lower signal-to-noise ratio (SNR). In this study, we introduce a principal component analysis (PCA) denoising algorithm for QSM data, showing the ability to generate QSM maps of the human brain at 0.6x0.6x0.6 mm³ resolution at 3T in vivo.

View Presentation Video

Javier Gonzalez-Castillo¹, Isabel Fernandez¹, Daniel A Handwerker¹, Ka-Chun Lam², Francisco Pereira², and Peter A Bandettini^1,3
1Section on Functional Imaging Methods, NIMH, Bethesda, MD, United States, ²Machine Learning Team, NIMH, Bethesda, MD, United States, ³FMRI Core Facility, NIMH, Bethesda, MD, United States

Keywords: Machine Learning/Artificial Intelligence, fMRI, Manifold Learning, TSNE, UMAP, Laplacian Eigenmaps, Intrinsic Dimension

The exploration of time-varying aspects of the human functional connectome (FC) is challenging because the high dimensionality of connectivity matrices precludes direct visual inspection in a meaningful manner. Dimensionality reduction helps circumvent this problem, yet effective application requires a-priori knowledge of the intrinsic dimension of the data and careful selection of algorithmic hyperparameters. Here, we first estimate the intrinsic dimension of FC data. Next, we use data with known cognitive state changes to evaluate the effectiveness of Laplacian Eigenmaps, T-SNE and UMAP to generate informative low dimensional representations of time-varying FC data for explorative and predictive purposes.

View Presentation Video

Deepa Anand¹, Dattesh Shanbhag¹, Chitresh Bhushan², and Uday Patil^3
1GE Healthcare, Bangalore, India, ²GE Healthcare, Niskayuna, NY, United States, ³GE Healthcaer, Bangalore, India

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

Multi-task training is attractive in AI applications (from memory, processing time and potential data reduction), where tasks have commonality in terms of features, but still requires differentiation for individual outputs derived. In this work we present a methodology to implement robust multi-task learning framework considering various strategies (parallel, iterative and sequential). We tested the approach for image quality assessment of spine MRI localizer images. We demonstrate that the sequential training is the most effective, in preserving an accuracy above the acceptable level while allowing for a save in number of model parameters (50%).

View Presentation Video

Sina Amirrajab¹, Yasmina Al Khalil¹, Josien Pluim¹, Marcel Breeuwer^1,2, and Cian M. Scannell^1
1Biomedical Engineering, Eindhoven University of Technology, Eindhoven, Netherlands, ²MR R&D - Clinical Science, Philips Healthcare, Eindhoven, Netherlands

Keywords: Machine Learning/Artificial Intelligence, Artifacts, Respirator Artifact Simulation

To tackle data scarcity for training a deep-learning algorithm for cardiac MR image quality assessment, we develop a k-space method for simulating respiratory motion artifacts with different levels of severity on artifact-free publicly available cardiac MRI data. The benefit of such simulated data is investigated, demonstrating the usefulness of training a feature extractor with the simulated artifacts for image quality classification. Our proposed method achieved the test accuracy of 0.625 and Cohen's Kappa of 0.473 (n=120 images), ranking third in task one for the CMRxMotion challenge of MICCAI 2022.

View Presentation Video

Chitresh Bhushan¹, Dattesh D Shanbhag², Uday Patil², Trevor Kolupar³, and Maggie Fung^3
1AI and Medical Imaging, GE Research, Niskayuna, NY, United States, ²GE Healthcare, Bangalore, India, ³GE Healthcare, Waukesha, WI, United States

Keywords: Machine Learning/Artificial Intelligence, Visualization, Scan Planning, Spine, Reformatting

We present a generalized DL-based intelligent slice placement framework for planes of cervical neural foramina (CF) and lumbar pars interarticularis (PI) for spine MRI. CF and PI scan improves assessment of foraminal stenosis and lumbar spondylolysis respectively, but requires highly skilled operator for accurate prescription. Our approach enables automatic patient-specific scan plane prescription from routine axial T2W images. In our test, it achieves mean error of <0.7 mm and <0.2 degrees and demonstrates similar/better contiguous anatomical visualization to manual scans on retrospective reformatting of 3D data. These results indicate that our approach is accurate and suitable for clinical usage.

View Presentation Video

Ashish Saxena¹, Chitresh Bhushan², Soumya Ghose², Patil Uday¹, Kameswari Padmanabhan³, Sanjay NT³, and Dattesh Shanbhag^3
1GE Research, Bengaluru, India, ²GE Research, Niskayuna, NY, United States, ³GE Healthcare, Bengaluru, India

Keywords: Machine Learning/Artificial Intelligence, Visualization, workflow

Saturation band placement is important for obtaining good quality MRI images in presence of structures which can generate artifacts such as cardiac regions or large pulsating blood vessels. In anatomy such as spine, saturation-band placement can be time consuming since technologist has to ensure that it doesn’t overlap with vertebrae regions. In this study, we demonstrated an automated deep-learning method to accurately place the saturation band on 2D three-plane localizer images with no intervention from the technologist.

View Presentation Video

Kevin M. Koch^1,2 and Andrew S. Nencka^1,2
1Radiology, Medical College of Wisconsin, Milwaukee, WI, United States, ²Center for Imaging Research, Medical College of Wisconsin, Milwaukee, WI, United States

Keywords: Machine Learning/Artificial Intelligence, Vessels, Metal Artifact

Multi-spectral imaging around metallic implants allows for a unique application of subject-specific deep learning applications because both high-resolution standard acquisitions and multi-spectral acquisitions are performed with significant regions of artifact-free overlap. A deep neural network can be trained in that region of artifact-free overlap within a single patient to yield a contrast transform on the multi-spectral images and achieve similar contrast to traditional acquisitions in the region of those acquisitions obscured by metal artifacts. In this proof of concept, a preliminary example of synthesizing time of flight-like contrast from a multi-spectral acquisition with vascular flow voids was shown.

View Presentation Video

Zhuang Xiong¹, Yang Gao², and Hongfu Sun^1
1the University of Queensland, Brisbane, Australia, ²Central South University, Changsha, China

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

Existing supervised deep learning methods for quantitative susceptibility mapping (QSM) could lead to degraded results when applied to phase images acquired with different scan parameters, such as image resolution and acquisition orientation. This work proposes a novel unsupervised learning method incorporating style transfer and deep image prior to enabling the reconstruction of susceptibility maps from local field maps acquired with any scan parameters. To speed up the inference, a pre-training strategy is also proposed, reducing reconstruction time from minutes to seconds.

View Presentation Video

Bangjun Chen¹, Jian Wang¹, Yirong Zhou¹, Yu Hu¹, Shuxian Niu¹, Biao Qu², Di Guo³, Jingjing Xu¹, Jiyang Dong¹, and Xiaobo Qu^1
1Biomedical Intelligent Cloud R&D Center, Department of Electronic Science, National Institute for Data Science in Health and Medicine, Xiamen University, Xiamen, China, ²Department of Instrumental and Electrical Engineering, Xiamen University, Xiamen, China, ³School of Computer and Information Engineering, Xiamen University of Technology, Xiamen, China

Keywords: Machine Learning/Artificial Intelligence, Segmentation

Deep learning and cloud computing technologies have shown remarkable performance in medical imaging research. Using deep learning for clinical research requires a programming foundation, while the annotation of image data is a highly specialized and time-consuming process. In this work, we develop a high-performance online medical image annotation and training platform (CloudBrain-LabelAI). It provides medical image researchers with an efficient image annotation platform for the rapid construction of datasets for deep learning. Meanwhile, it provides codeless image segmentation training and prediction based on cloud computing, which greatly reduces the threshold of medical image segmentation efforts and simplifies the overall workflow.

View Presentation Video

Soudabeh Kargar¹, Ouri Cohen¹, Sungmin Woo², Hebert Alberto Vargas², and Ricardo Otazo^1
1Medical Physics, MSKCC, New York, NY, United States, ²Radiology, MSKCC, New York, NY, United States

Keywords: Machine Learning/Artificial Intelligence, DSC & DCE Perfusion, Deep Learning

This work develops a deep learning technique that is trained according to physics and kinetics for self-contained comprehensive quantification of dynamic contrast-enhanced MRI. In addition to perfusion parameters, patient-specific parameters that affect the quantification are estimated, including bolus arrival time, T1, steady state magnetization, and AIF. The DCE-MRI quantification network was tested on a patient with cervical cancer and demonstrated high concordance between two scans separated by 24 hours. Physics plus kinetics informed network learning (PKNet) enabled the quantification of multiple parameters which has the potential to increase reproducibility of quantitative DCE-MRI, a long-desired goal.

View Presentation Video

Yan Li¹, Ziming Xu¹, Linggen Dong², Yajie Wang¹, Peng Liu², Ming Lv², and Huijun Chen^1
1Center for Biomedical Imaging Research, Tsinghua University, Beijing, China, ²Department of Neurosurgery, Beijing Tiantan Hospital, Beijing Neurosurgical Institute, Capital Medical University, Beijing, China

Keywords: Data Analysis, DSC & DCE Perfusion

Assessment of intracranial aneurysm rupture risk is clinically critical. Recent studies have shown that the aneurysm wall permeability (K^trans) and wall enhancement index are independent risk factors for predicting saccular aneurysm rupture. However, this conclusion has not been confirmed in non-saccular aneurysms. Our study suggested that K^trans was significantly associated with the aneurysm size and PHASES score in non-saccular aneurysms. But K^trans did not associate with the wall enhancement index in non-saccular aneurysms. Therefore, K^trans can provide an independent quantitative indicator for assessing the risk of rupture of in non-saccular intracranialaneurysms.