View Presentation Video

Deepa Anand¹, Dattesh Shanbhag¹, Chitresh Bhushan², Kavitha Manickam², and RAdhika Madhavan^2
1GE Healthcare, Bangalore, India, ²GE Healthcare, Niskayuna, NY, United States

Keywords: Machine Learning/Artificial Intelligence, Artifacts, Metal implants

In MRI, presence of metal can cause blooming artifacts on MRI images. Automatically detecting the presence of metal and localizing it on the scout images itself can help streamline the MRI imaging workflow decisions downstream. In this work, we demonstrate a DL framework for hot spotting of metal/metal affected regions based on 2D three-plane MRI scout images of spine. The results indicate that the solution not only detects metal regions well on spine images (on which it was trained), but also is generalizable enough to work on other anatomies such as knee which was not part of the training data.

View Presentation Video

Sean C. Epstein^1,2, Timothy J. P. Bray^3,4, Margaret Hall-Craggs^3,4, and Hui Zhang^1,2
1Computer Science, UCL, London, United Kingdom, ²Centre for Medical Image Computing, UCL, London, United Kingdom, ³Centre for Medical Imaging, UCL, London, United Kingdom, ⁴Imaging, UCLH, London, United Kingdom

Keywords: Machine Learning/Artificial Intelligence, Data Analysis

We compare modern deep learning (DL)-based parameter-estimation methods to their traditional maximum-likelihood estimation (MLE) counterparts by evaluating each approach’s performance in two clinical classification tasks. This is motivated by recent work demonstrating the inherent bias-variance trade-off that differentiates different DL-based approaches. Results show how these trade-offs manifest in the ‘real world’ of tissue classification, and how they compare to the performance achievable with conventional iterative MLE.

View Presentation Video

Siavash Khallaghi¹, Lucas Porto¹, Sean London², Yosef Chodakiewitz², Rajpaul Attariwal¹, and Sam Hashemi^1
1Voxelwise Imaging Technology Inc., Vancouver, BC, Canada, ²Prenuvo, Vancouver, BC, Canada

Keywords: Machine Learning/Artificial Intelligence, Machine Learning/Artificial Intelligence, Spine, spondyloarthropathy, spondylosis

We present a fully automatic system for the quantitative assessment of discs and vertebrae using convolutional neural networks. The proposed algorithm works in three stages: 1) segmentation/identification of spinal anatomy; 2) curvature analysis; and 3) detection of pathological conditions of intervertebral discs. We validate the proposed approach on a large dataset of 1,500 subjects with sagittal T2-weighted whole spine MRI, obtained as part of a whole body MRI protocol in a preventative health screening program.

View Presentation Video

Christof Karmonik¹, Lily Buckner², Kayla Wilhoit¹, and Trisha Roy^3
1Houston Methodist Research Institute, Houston, TX, United States, ²Houston Methodst Research Institute, Houston, TX, United States, ³Houston Methodist Hospital, Houston, TX, United States

Keywords: Machine Learning/Artificial Intelligence, High-Field MRI, ultra-short TE, 7 Tesla

To differentiate soft versus hard peripheral arterial disease (PAD) plaques, a dedicated ultra-high field magnetic resonance imaging (MRI) histology protocol using three different image contrast (ultra-short TE, T1-weighted and T2-weighted) was used to scan on a clinical 7 Tesla human MRI scanner six amputated legs immediately after surgery that were harboring PAD lesions.  A 2D convolutional network (CNN) variational autoencoder (VAE) was created. For each lesion, average classification values along the long axis of the artery were determined from latent space and ranged from partially occluded to patent (three lesions), soft-plaque occluded (two) and hard-plaque occluded (one).

View Presentation Video

Joshua Eisenstat¹, Matthias Wagner², Logi Vidarsson³, Birgit Ertl-Wagner^2,4, and Dafna Sussman^1,5,6
1Department of Electrical, Computer and Biomedical Engineering, Faculty of Engineering and Architectural Sciences, Toronto Metropolitan University, Toronto, ON, Canada, ²Division of Neuroradiology, The Hospital for Sick Children, Toronto, ON, Canada, ³Department of Diagnostic Imaging, The Hospital for Sick Children, Toronto, ON, Canada, ⁴Department of Medical Imaging, Faculty of Medicine, University of Toronto, Toronto, ON, Canada, ⁵Institute for Biomedical Engineering, Science and Technology (iBEST), Toronto Metropolitan University and St. Michael’s Hospital, Toronto, ON, Canada, ⁶Department of Obstetrics and Gynecology, Faculty of Medicine, University of Toronto, Toronto, ON, Canada

Keywords: Machine Learning/Artificial Intelligence, Machine Learning/Artificial Intelligence, Convolutional Neural Networks, Fet-Net

Fetal orientation determines the mode of delivery. It is also important for sequence planning in fetal MRI. This abstract proposes Fet-Net, a deep-learning algorithm, which uses a novel convolutional neural network (CNN) architecture, to automatically detect fetal orientation from a 2-dimensional (2D) magnetic resonance imaging (MRI) slice. 6,120 2D MRI slices displaying vertex, breech, oblique and transverse fetal orientations were used for training, validation and testing. Fet-Net achieved an average accuracy and F1 score of 97.68%, and a loss of 0.06828. Fet-Net was able to detect and classify fetal orientation, which may serve to accelerate fetal MRI acquisition.

View Presentation Video

Huixin Tan¹ and Yantu Huang^1
1Siemens Shenzhen Magnetic Resonance Ltd., China, Shenzhen, China

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

Pilot Tone (PT) respiratory signals are susceptible to interference like patient bulk motion or radio frequency interference. The signal curve is prone to bad when remains strong interference after suppression processing in the learning phase, resulting in low triggering accuracy and efficiency. With real-time quality assessment in the learning phase, a good-quality signal can be selected to learn better processing parameters. To ensure robustness and inference time, a tiny CNN is utilized to classify the signal into good and bad quality. Experimental results demonstrate that the proposed method has a strong ability to assess the quality of PT respiratory signals.

View Presentation Video

Jayse Merle Weaver^1,2, Marissa DiPiero^2,3, Patrik Goncalves Rodrigues², Hassan Cordash², and Douglas C Dean III^1,2,4
1Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ²Waisman Center, University of Wisconsin-Madison, Madison, WI, United States, ³Neuroscience Training Program, University of Wisconsin-Madison, Madison, WI, United States, ⁴Pediatrics, University of Wisconsin-Madison, Madison, WI, United States

Keywords: Machine Learning/Artificial Intelligence, Artifacts, Quality Control

A three-dimensional convolutional neural network was trained to detect motion artifacts on a volume level for two pediatric diffusion MRI datasets acquired between 1 month and 3 years of age. Accuracies of 95% and 98% were achieved between the two datasets. Additionally, the effects of motion-corrupted volumes on quantitative parameter estimation was examined. Data was processed without quality control and with quality control performed by the neural network. DTI and NODDI metrics were calculated and compared between methods. Significant differences were found for both individual and group results.

View Presentation Video

Jinfa Ren¹, Dongming Han¹, Xiaoyang Zhai¹, Huijia Yin¹, Ruifang Yan¹, and Kaiyu Wang^2
1Department of MR, The First Affiliated Hospital of Xinxiang Medical University, Weihui, China, ²GE Healthcare, MR Research China, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Radiomics

Detecting true tumor recurrence and treatment-related effects in glioma after treatment is crucial for patient managements and challenging via conventional MRI for differentiation. Radiomics can be used to access the details in the images in an objective way. We constructed models based on multiple modalities by using radiomics features of the postoperative enhanced and edematous regions to find key features for identifying true tumor recurrence. Features from CE-T1WI and enhanced regions have excellent classification performance, and the model of multimodality with whole regions is the best, which may aid clinicians in developing individualized treatment strategies.

View Presentation Video

Shahrzad Moinian^1,2, Nyoman Kurniawan ¹, Shekhar Chandra ³, Viktor Vegh^1,2, and David Reutens^1,2
1Centre for Advanced Imaging, The University of Queensland, St Lucia, Brisbane, Australia, ²Australian Research Council Training Centre for Innovation in Biomedical Imaging Technology, The University of Queensland, Brisbane, Australia, ³School of Information Technology and Electrical Engineering, The University of Queensland, Brisbane, Australia

Keywords: Machine Learning/Artificial Intelligence, Motion Correction

A primary challenge for in vivo kidney MRI is the presence of different types of involuntary physiological motion, affecting the diagnostic utility of acquired images due to severe motion artifacts. Existing prospective and retrospective motion correction methods remain ineffective when dealing with complex large amplitude nonrigid motion artifacts. We introduce an unsupervised deep learning-based method for in vivo kidney MRI motion correction. We demonstrate that our deep learning model achieved the average structural similarity index measure (SSIM) of 0.76±0.06 between the reconstructed motion-corrected and ground truth motion-free images, showing an improvement of about 0.33 compared to the corresponding motion-corrupted images.

View Presentation Video

Jon Haitz Legarreta¹, Laurent Petit², Pierre-Marc Jodoin¹, and Maxime Descoteaux^1
1Computer Science, Université de Sherbrooke, Sherbrooke, QC, Canada, ²Université Bordeaux, Bordeaux, France

Keywords: Machine Learning/Artificial Intelligence, Tractography & Fibre Modelling

Current tractography methods have a limited ability to accurately reconstruct the long-range brain white matter fiber pathways. Local orientation propagation methods provide tractograms with a non-negligible amount of implausible streamlines. In this work, we propose an artificial intelligence model to recover long-range white matter tracks that are potentially missed in conventional streamline propagation. Our method uses the generative ability of an autoencoder to propose new, plausible streamlines that are subsequently exchanged, according to a given similarity index, with the implausible streamlines in a given tractogram. This allows to potentially improve the reliability of the reconstructed fiber pathways.

View Presentation Video

Ke Lei¹, Ali B. Syed², Xucheng Zhu³, John M. Pauly¹, and Shreyas V. Vasanawala^2
1Electrical Engineering, Stanford University, Stanford, CA, United States, ²Radiology, Stanford University, Stanford, CA, United States, ³GE Healthcare, Menlo Park, CA, United States

Keywords: Machine Learning/Artificial Intelligence, Machine Learning/Artificial Intelligence, Region of Interest

Manual prescription of the field of view (FOV) by MRI technologists is variable and prolongs the scanning process. Often, the FOV is too large or crops critical anatomy. We propose a deep-learning framework, trained by radiologists’ supervision, for predicting region of interest (ROI) and automating FOV prescription. The proposed ROI prediction model achieves an average IoU of 0.867, significantly better (P<0.05) than two baseline models and not significantly different from a radiologist (P>0.12). The FOV prescribed by the proposed framework achieves an acceptance rate of 92% from an experienced radiologist. 

View Presentation Video

Keigo Yamazaki^1,2, Yoshitaka Masutani³, Wataru Uchida¹, Koji Kamagata¹, Koh Sasaki^4,5, and Shigeki Aoki^1
1Department of Radiorogy, Juntendo University graduate School of Medicine, Tokyo, Japan, ²Department of Radiological Science, Tokyo Metropolitan University, Tokyo, Japan, ³Tohoku University graduate School of Medicine, Miyagi, Japan, ⁴Graduate School of Infomation Sciences, Hiroshima City University, Hiroshima, Japan, ⁵Hiroshima Heiwa Clinic, Hiroshima, Japan

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

Free water imaging (FWl), among the diffusion MRI (dMRI) family, is an extended version of the single diffusion tensor model by adding the isotropic diffusion compartment. Generally, FWI parameters have been estimated by fitting of signal model to measured DWI signals. Recently, machine learning techniques have shown promising results also in dMRI parameter inference.In this study, we aimed at development of a hybrid approach for FWI parameter estimation based on synthetic q-space learning (synQSL) and conventional fitting.Our approach was validated by comparison with the conventional fitting method based on quantitative and visual evaluation and computation time.

View Presentation Video

Maximilian Zubke^1,2, Robin A Müller^1,2, Marius Wernz^1,2, Filip Klimeš^1,2, Frank Wacker^1,2, and Jens Vogel-Claussen^1,2
1Institute of Diagnostic and Interventional Radiology, Hannover Medical School, Hannover, Germany, ²Biomedical Research in Endstage and Obstructive Lung Disease Hannover (BREATH), German Center for Lung Research (DZL), Hannover, Germany

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

Currently, MR-based ventilation imaging relying on radial 3D-stack-of-stars spoiled gradient echo sequence requires a fairly long acquisition time of 8 minutes, which may impact clinical translation. Therefore, a shorter acquisition time is desired. In this study, a novel deep learning approach called transformer was evaluated for image restauration of radial undersampled lung images from 16 post-COVID-19 patients. For each patient, images resulting from 4- and 8 minutes acquisitions were provided. A transformer was trained to translate the 4-minute-version to the corresponding 8-minute-version and led to a significant image quality improvement, demonstrated by three complementary image similarity metrics. 

View Presentation Video

Subham Kumar¹ and Harikrishna Rai^1
1GE Healthcare, Bengaluru, India

Keywords: Machine Learning/Artificial Intelligence, Data Analysis

Estimating patient height and weight is used to determine the optimum SAR (Signal Modelling) for the patient. We used a deep learning approach to first generate the point cloud of the person (Data Processing) followed by prediction of the height and weight (Data Analysis). This tackles the problems of heavy occlusion which occurs in the MR imaging scenario in the form of coils/blankets and different positions in which the patient will be placed. We achieved a MAE score of 4.9 cm on the height and 6.2 kg on the weight. This is a promising solution to an important problem.

Qiongge Li¹, Yayan Yin¹, and Jie Lu^1
1Xuanwu Hospital Capital Medical University, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Diffusion/other diffusion imaging techniques

Deep learning-based reconstruction may improve the image signal to noise ratio without impacting the image contrasts. Intravoxel incoherent motion (IVIM) often require multiple b values with multiple averages that give rise to prolonged scan time. In this work, deep learning reconstruction is used to reduce the overall IVIM scan time. Based on qualitative and quantitative analysis, deep learning reconstruction may significantly improve the results of IVIM and make an abbreviated IVIM feasible.

View Presentation Video

Li Guo Liu¹, Yue Xin Zhang¹, Nan Zhang¹, Feng Hua Xiao¹, Jing xin Chen¹, and Lin Ma^1
1Department of Radiology, The First Medical Center, Chinese PLA General Hospital, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Multimodal

In this study, we proposed a promising method to distinguish the double expression lymphoma (DEL) from the non-double expression lymphoma (non-DEL) in primary central nervous system lymphoma (PCNSL) by using multiparametric MRI-based machine learning. The results showed that clinical characteristics and MR imaging features had no significant differences in distinguishing DEL from non-DEL . However, radiomics features could differentiate the two status and the best model in this study was SVMlinear with the combined four sequence group (AUCmean = 0.89±0.04). So multiparametric MRI based machine learning is promising in predicting DEL status in PCNSL.

View Presentation Video

Valentin H Prevost¹, Shelton Caruthers¹, Khadra Fleury², Wissam AlGhuraibawi³, and Kensuke Shinoda^1
1Canon Medical Systems Corporation, Otawara, Japan, ²Canon Medical Systems France, Paris, France, ³Canon Medical Systems USA, Tustin, CA, United States

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

In MRI, edge sharpness is one of the main criteria to allow structure’s delineation and relevant clinical diagnosis. One way to improve the sharpness is to artificially increase the reconstructed matrix size with methods such as zero-padding interpolation (ZIP). In a previous study, we created a deep learning reconstruction (DLR) pipeline, combining ZIP with two CNN’s: the first one trained to reduce image noise, and the second one to reduce ringing artifacts. The goal of this work was to evaluate the clinical impact of this DLR pipeline on pathological knee images performed at 1.5T and 3T, compared to standard reconstructions.

View Presentation Video

Yang Jing¹, Li Qiong Ge¹, Wu Tao², Qi Zhi Gang¹, Zhao Cheng¹, and Lu Jie^1
1Department of Radiology and Nuclear Medicine, xuanwu hospital, Capital Medical University, Beijing, China, ²General Electric Medical System Trade Development (Shanghai) Co., Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Brain, Hippocampus, Deep learning reconstruction, Epilepsy

Deep learning reconstruction (DL Recon) method can improve the image signal-to-noise ratio and contrast-to-noise ratio without prolonging the scanning time and affecting the signal intensity difference of bilateral hippocampus. It can improve the image quality of hippocampus toe and surrounding small lesions dramatically as in Fig.1. 

View Presentation Video

Matthias Jung¹, Marco Reisert², Susanne Rospleszcz^3,4, Annette Peters^3,4, Johanna Nattenmüller¹, Christopher L. Schlett¹, Fabian Bamberg¹, and Jakob Weiss^1
1Department of Diagnostic and Interventional Radiology, University Medical Center Freiburg, Freiburg, Germany, ²Medical Physics, Department of Diagnostic and Interventional Radiology, University Medical Center Freiburg, Freiburg, Germany, ³Institute of Epidemiology, Helmholtz Zentrum München, Neuherberg, Germany, ⁴Institute for Medical Information Processing, Biometry and Epidemiology, Ludwig-Maximilians-University München, München, Germany

Keywords: Screening, Diabetes

Diabetes remains a major challenge for healthcare systems, making screening and early detection desirable. We used deep learning to quantify myosteatosis as 1) skeletal muscle fat fraction (SMFF) and 2) intramuscular adipose tissue (IMAT) normalized for SM mass and assessed their association with impaired glucose metabolism. SMFF had a higher discriminatory capacity for impaired glucose metabolism than IMAT. In multivariable logistic regression adjusted for baseline demographics and cardiometabolic risk factors, only SMFF remained an independent predictor of impaired glucose metabolism. Deep learning-based MR phenotyping enables opportunistic screening of myosteatosis and may identify individuals at high risk for impaired glucose metabolism.

View Presentation Video

Xinzeng Wang¹, Yedaun Lee^2,3, Joonsung Lee⁴, Sagar Mandava⁵, Ty A. Cashen⁶, Xucheng Zhu⁷, and Arnaud Guidon^8
1GE Healthcare, Houston, TX, United States, ²Department of Radiology, Haeundae Paik Hospital, Busan, Korea, Republic of, ³Inje University College of Medicine, Busan, Korea, Republic of, ⁴GE Healthcare, Seoul, Korea, Republic of, ⁵GE Healthcare, Atlanta, GA, United States, ⁶GE Healthcare, Madison, WI, United States, ⁷GE Healthcare, Menlo Park, CA, United States, ⁸GE Healthcare, Boston, MA, United States

Keywords: Cancer, Image Reconstruction, Liver, Pancreas

Stack of Star acquisition is one of the most frequently used non-Cartesian k-space sampling methods due to its fast speed and robustness to motion. To further reduce the scan time or increase the temporal resolution, Stack of Star is often down-sampled with fewer spokes and advanced sampling patterns, such as golden angle acquisition. However, this makes Stack of Star prone to noise and streak artifacts, limiting the in-plane resolution and degrading diagnostic quality. In this work, we evaluated a deep-learning based stack-of-star method for free-breathing abdominal imaging and it shows improved diagnostic quality.