View Presentation Video

Huixian Zhang¹, Redha Ali¹, Hailong Li¹, Wen Pan¹, Scott B. Reeder², David T. Harris², William R. Masch³, Anum Alsam³, Krishna P. Shanbhogue⁴, Nehal A. Parikh¹, Jonathan R. Dillman¹, and Lili He^1
1Cincinnati Children's Hospital, Cincinnati, OH, United States, ²University of Wisconsin-Madison, Madison, WI, United States, ³Michigan Medicine, University of Michigan, Ann Arbor, MI, United States, ⁴NYU Langone Health, New York, NY, United States

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

To develop an AI system for precise and fully automated, simultaneous segmentation of the liver and spleen from T1-weighted and T2-weighted MRI data. Our study compares the performance of the U-Net Transformer (UNETR) and the standard 3D U-Net model for simultaneous liver and spleen segmentation using MRI images from pediatric and adult patients from multiple institutions. Our work demonstrates that the UNETR shows a statistical improvement over 3D U-Net in both liver and spleen segmentations.

View Presentation Video

Yiheng Li¹ and Ryan Chamberlain^1
1Subtle Medical Inc., Menlo Park, CA, United States

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

To optimize DL-based medical registration models to produce a more realistic flow field and reduce the over-smoothing effect of deformable registration while keeping the generalizability of the multi-contrast registration model to work on all anatomical structures and contrast of MRI, we proposed an algorithm that adopts the image synthetic framework from SynthMorph and optimized with cycle-consistent loss. By experimenting with the jacobian loss, bidirectional loss, and cycle-consistent loss, we managed to further optimize the results of registered images. The evaluation of two MRI image datasets, the BraTS dataset, and the LSpine dataset, demonstrated the increased SSIM, PSNR, and LocalNCC.

View Presentation Video

Reza Kalantar¹, Jessica Mary Winfield^1,2, Mihaela Rata¹, Gigin Lin³, Susan Lalondrelle^1,4, Christina Messiou^1,2, Matthew David Blackledge¹, and Dow-Mu Koh^1,2
1Radiotherapy and Imaging, The Institute of Cancer Research, London, United Kingdom, ²Radiology, The Royal Marsden Hospital, London, United Kingdom, ³Medical Imaging and Intervention, Chang Gung Memorial Hospital at Linkou and Chang Gung University, Taoyuan, Taiwan, ⁴Gynaecological Unit, The Royal Marsden Hospital, London, United Kingdom

Keywords: Machine Learning/Artificial Intelligence, Cancer

Quantitative magnetic resonance imaging (qMRI) can provide additional information for diagnosis and response assessment, but adoption of multi-parametric qMRI techniques has been hindered by long acquisition times and labor-intensive processing steps. Magnetic resonance fingerprinting (MRF) provides quantitative maps in a single acquisition but MRF deployment in clinical studies still requires manual delineation of volumes of interest. A spatial-adaptive deep learning framework was developed to segment cervical cancer on MRI and quantify T₁ relaxation times of the tumor pre- and post-radiotherapy treatment. Our results suggest that automated segmentation models may be promising tools for quantitative tumor evaluation and treatment response assessment.

View Presentation Video

Michael Carl¹, Kaustaub Lall², Armin Jamshidi², Eric Chang^3,4, Sheronda Statum^3,4, Anja Brau¹, Christine B Chung^3,4, Maggie Fung¹, and Won C. Bae^3,4
1General Electric Healthcare, Menlo Park, CA, United States, ²Electrical and Computer Engineering, University of California, San Diego, La Jolla, CA, United States, ³Radiology, University of California, San Diego, La Jolla, CA, United States, ⁴Radiology, VA San Diego Healthcare System, San Diego, CA, United States

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

Evaluation of 3D bone morphology of the glenohumeral joint is necessary for pre-surgical planning. Zero echo time (ZTE) MRI provides excellent bone contrast, and we developed a deep learning model to perform automated segmentation of major bones (i.e., humerus and others) from ZTE to aid evaluation. Axial ZTE images of normal shoulders (n=31) acquired at 3T were annotated for training with a 2D U-Net, and the trained model was validated with testing data (n=10 normal shoulder, n=6 symptomatic). Testing accuracy was around 80 to 90% (Dice score) for either cohort, except for a few failed cases with very low scores.

View Presentation Video

Andrew Seohwan Yu^1,2,3, Sibaji Gaj^1,2, William Holden^1,2, Richard Lartey^1,2, Jeehun Kim^1,2,4, Carl Winalski^1,2,5, Naveen Subhas^1,2,5, and Xiaojuan Li^1,2,5
1Program of Advanced Musculoskeletal Imaging (PAMI), Cleveland Clinic, Cleveland, OH, United States, ²Department of Biomedical Engineering, Lerner Research Institute, Cleveland Clinic, Cleveland, OH, United States, ³Cleveland Clinic, Department of Computer and Data Sciences, Case Western Reserve University, Cleveland, OH, United States, ⁴Department of Electrical, Computer, and Systems Engineering, Case Western Reserve University, Cleveland, OH, United States, ⁵Department of Diagnostic Radiology, Imaging Institute, Cleveland Clinic, Cleveland, OH, United States

Keywords: Machine Learning/Artificial Intelligence, Segmentation

Quantitative assessment of the bone marrow edema-like lesions and their association with osteoarthritis requires a consistent and unbiased segmentation method, which is difficult to obtain in the presence of human annotators. This study proposes an unsupervised approach using Bayesian deep learning and conditional generative adversarial networks that detects and segments anomalies without human intervention. The full pipeline has a lesion-wide sensitivity of 0.86 on unseen scans. This approach is expected to be generalizable to other lesions and/or modalities.

View Presentation Video

Maysam Orouskhani¹, Shaojun Xia², Mahmud Mossa-Basha¹, and Chengcheng Zhu^1
1Department of Radiology, University of Washington, Seattle, WA, United States, ²Peking University Cancer Hospitals & Institution, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Machine Learning/Artificial Intelligence, Compound Loss function, Deep Neural Networks, nnU-Net

In the segmentation of intracranial aneurysm, deep neural networks are equipped with modified loss functions to penalize the training weights for aneurysm false predictions and conduct unbiased learning. In this paper, we used a new compound loss function to capture the different aspects of embedding as well as diverse features. The proposed loss was given to a 3D full resolution nnU-Net to segment imbalanced TOF-MRA images from ADAM dataset. The proposed loss outperformed commonly used losses in terms of Dice, Sensitivity, and Precision. 

View Presentation Video

Yue Wu^1,2, Jinyuan Zhang^3,4, Zhixin Li^3,4, Yishuang Yang^3,4, Yan Zhuo^3,4, Xudong Zhao^2,3, and Zihao Zhang^2,3
1AHU-IAI AI Joint Laboratory, Anhui University, Hefei, China, ²Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, Hefei, China, ³State Key Laboratory of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences, Beijing, China, ⁴University of Chinese Academy of Sciences, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Vessels

Time-of-flight MR angiography (TOF-MRA) allows noninvasive visualization of the mouse intracranial vessels. However, there is no available tool for vessel segmentation in mouse 3D TOF-MRA. In this study, we developed an automated vessel segmentation method based on a convolutional neural network and multiscale vessel enhancement filtering algorithm. The method is potentially useful in the preclinical studies of cerebrovascular diseases.

View Presentation Video

Tong Wu¹, Santiago Estrada², Renza Gils¹, Ruisheng Su¹, Vincent Jaddoe³, Edwin Oei¹, and Stefan Klein^1
1Department of Radiology and Nuclear Medicine, Erasmus MC, Rotterdam, Netherlands, ²Image Analysis, German Center for Neurodegenerative Diseases (DZNE), Bonn, Germany, ³Department of Pediatrics, Erasmus MC, Rotterdam, Netherlands

Keywords: Machine Learning/Artificial Intelligence, Segmentation, Dixon MRI

This study is embedded in the Generation R Study, a population-based prospective cohort study in the Netherlands. Whole-body Dixon MRI scans were performed at age 13 years. A previous neural network (CDFNet) has been published that was trained on adults. We aimed to retrain this network on our MRIs in the Generation R Study. The obtained segmentations are in strong agreement with expert-generated manual segmentations and can therefore greatly reduce the manual workload. Therefore, for accurate abdominal fat quantification, segmentation of both subcutaneous and visceral adipose tissue, on Dixon MRIs of adolescents is feasible with a retrained convolutional neural network.

View Presentation Video

Lara Schlaffke¹, Niklas Doliwa¹, Marius Markmann¹, Marlena Rohm^1,2, Robert Rehmann^1,3, and Martijn Froeling^4
1Neurology, University Clinic Bergmannsheil Bochum gGmbH, Bochum, Germany, ²Heimer Institute for Muscle Research, University Clinic Bergmannsheil Bochum gGmbH, Bochum, Germany, ³Neurology, Klinikum Dortmund, University Witten-Herdecke, Dortmund, Germany, ⁴Radiology, UMC Utrecht, Utrecht, Netherlands

Keywords: Machine Learning/Artificial Intelligence, Muscle, Segmentation

Quantitative magnetic resonance imaging offers promising surrogate biomarkers for the early diagnosis and monitoring of pathological changes in leg muscles in patients with neuromuscular disorders. To use this in a clinical routine, automatic segmentation of muscles is needed. To investigate precision and accuracy of various Computational neuronal networks for automated segmentation of upper leg muscles, different loss function were used and quantitative outcome compared to the gold-standard of manual segmentation. The best segmentation results were achieved, with the Fokal Tversky loss function and a single input approach. 

Meng Dou^1,2, Ying Zhao ³, Tao Lin³, Yu Yao^1,2, and Ailian Liu^3
1Chengdu institute of computer application, Chinese academy of sciences, Chengdu, China, ²University of Chinese Academy of Sciences, Beijing, China, ³Department of Radiology, First Affiliated Hospital of Dalian Medical University, Dalian, China

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

In this paper, we propose a deep learning-based liver tumor segmentation algorithm in enhanced multi-phase MRI images. The experimental results show that the proposed method can segment liver and tumor in enhanced multi-phase MRI images with a smaller resource occupation, and outperforms the comparison method.

View Presentation Video

Soumya Ghose¹, Chitresh Bhushan¹, Dattesh Shanbhag², Desmond Teck Beng Yeo³, and Thomas K. Foo^3
1AI and Computer Vision, GE Research, Niskayuna, NY, United States, ²GE Healthcare, Bangalore, India, ³Biology & Physics, GE Research, Niskayuna, NY, United States

Keywords: Machine Learning/Artificial Intelligence, Segmentation, Vertebra segmentation, T2w, Deep Learning, Intensity Transformation

Deep learning (DL) models have been successful in solving segmentation problems that involve large, balanced, and labeled datasets. However, in the medical imaging domain, it is rare to find manually annotated datasets that capture the entire spectrum of heterogeneity. Novel datasets with significantly different intensities than training datasets, may adversely affect DL model performance. In this work, we present a hybrid framework for tissue-specific, non-linear intensity transformation of pediatric T2w images similar to that of the adults training dataset and demonstrate an improved performance for vertebra segmentation of pediatric datasets without the need for DL network re-training/re-tuning. 

View Presentation Video

Dongnan Liu^1,2, Mariano Cabezas², Dongang Wang^2,3, Zihao Tang^1,2, Geng Zhan^2,3, Kain Kyle^2,3, Linda Ly^2,3, James Yu^2,3, Chun-Chien Shieh^2,3, Ryan Sullivan⁴, Fernando Calamante^4,5, Michael Barnett^2,3, Wanli Ouyang⁶, Weidong Cai¹, and Chenyu Wang^2,3
1the School of Computer Science, University of Sydney, Sydney, Australia, ²Brain and Mind Centre, University of Sydney, Sydney, Australia, ³Sydney Neuroimaging Analysis Centre, Sydney, Australia, ⁴the School of Biomedical Engineering, University of Sydney, Sydney, Australia, ⁵Sydney Imaging, University of Sydney, Sydney, Australia, ⁶the School of Electrical and Information Engineering, University of Sydney, Sydney, Australia

Keywords: Machine Learning/Artificial Intelligence, Segmentation

Multiple sclerosis (MS) is a neurodegenerative disease of the central nerve system (CNS), which has the potential to cause a neurological disability, particularly for young adults. Recently, deep learning-based techniques are important for MS diagnosis and treatment, since they can segment the lesions caused by MS automatically and accurately. However, their applicability to multi-center scenarios is limited, due to the privacy and security issues in data sharing. To tackle these limitations, a decentralized deep learning framework is designed in this work, which can bring accurate multi-center MS lesion segmentation performance without sharing the raw data.

Wenyun Liu¹, Chinting Wong², Cheng Li³, Di Yu¹, Yi Zhu⁴, Ke Jiang⁴, and Huimao Zhang^1
1Radiology, The First Hospital of Jilin University, Changchun, China, ²Nuclear Medicine, The First Hospital of Jilin University, Changchun, China, ³Cardiovascular Center, The First Hospital of Jilin University, Changchun, China, ⁴Philips Healthcare, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Myocardium, Cardiomyopathy,Data Analysis

Cardiac MRI is considered reference standard for the noninvasive assessment of ventricular volumes and function. However, long multibreath-hold acquisition time can prove difficult in patients and lead to poor image quality. In this study, we investigated the use of a deep learning-based reconstruction algorithm, named Compressed SENSE Artificial Intelligence(CS-AI), to accelerate two-dimensional cine bSSFP for cardiac MRI. The purpose of this study was to compare the image quality and performance of a CS-AI-based cine sequence between reference and accelerated methods: SENSE, Compressed-SENSE, and CS-AI, and then to investigate the impact of images reconstructed  by deep learning on AI segmentation model.

View Presentation Video

Alejo Costanzo^1,2, Birgit Ertl-Wagner^3,4, and Dafna Sussman^1,5,6
1Department of Electrical, Computer and Biomedical Engineering, Toronto Metropolitan University, Toronto, ON, Canada, ²Institute for Biomedical Engineering, Science and Technology, Toronto Metropolitan University and St. Michael’s Hospital, Toronto, ON, Canada, ³Division of Neuroradiology, 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, 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, Segmentation, Convolutional Neural Network

Amniotic Fluid Volume (AFV) is an important fetal biomarker when diagnosing certain fetal abnormalities. We aim to implement a novel Convolutional Neural Network (CNN) model for amniotic fluid (AF) segmentation which can facilitate clinical AFV evaluation. The model, called AFNet was trained and tested on a radiologist–validated AF dataset. AFNet improves upon ResUNet++ through the efficient feature mapping in the attention block, and transpose convolutions in the decoder. Experimental results show that our AFNet model achieved a 93.38% mean Intersection over Union (mIoU) on our dataset. We further demonstrate that AFNet outperforms state-of-the-art models while maintaining a low model size.

View Presentation Video

Namho Jeong¹, Beomgu Kang¹, and Hyunwook Park^1
1Korea Advanced Institute of Science and Technology, Daejeon, Korea, Republic of

Keywords: Machine Learning/Artificial Intelligence, Tumor

Abundant data of healthy subjects is available and thus it would be helpful if normal data can boost the brain tumor segmentation. We proposed the out-of-distribution (OOD) feature based on the learned distribution of normal data to discriminate abnormal tumors. Once the neural network was trained to synthesize other contrast MR images with normal data, it failed to properly synthesize the tumor tissues. This OOD characteristic was used for the generation of new feature that can help the tumor segmentation. It was verified that the OOD features contributed to an improvement of segmentation through experiments.

View Presentation Video

Yunkyoung Jun¹, Jiwoo Jeong¹, Seokha Jin¹, Noehyun Myung¹, Jimin Lee², and Hyungjoon Cho^1
1BME, UNIST, Ulsan, Korea, Republic of, ²Nuclear Engineering, UNIST, Ulsan, Korea, Republic of

Keywords: Machine Learning/Artificial Intelligence, Animals

The proposed research implemented an automatic tumor segmentation application on an orthotopic breast tumor model. This application can segment the tumors accurately and monitor tumor growth and the therapeutic effect of Doxorubicin for treatment. Also, the outputs from the application can reconstruct into 3D rendering and offer the visualization of shape and volume. As a result, the application can be applied to orthotopic breast tumor model research.

View Presentation Video

Ananya Anand¹, Daniel Nussey², and Dafna Sussman^3
1Toronto Metropolitan University, Fremont, CA, United States, ²Toronto Metropolitan University, Toronto, ON, Canada, ³Electrical, Computer, and Biomedical Engineering, Toronto Metropolitan University, Toronto, ON, Canada

Keywords: Machine Learning/Artificial Intelligence, Magnetization transfer, Multi-Modal

This study presents an evaluation of utilizing MTR and FLAIR as compared to T1wCE and FLAIR images for automatic segmentation of brain metastases, due to the unknown consequences posed by accumulated contrast enhancement. Numerous combinations of FLAIR, T1wCE, and MTR MRI images were tested on three top-performing segmentation models (MS-FCN, U-NET, and SegresNet) to evaluate the feasibility of using MTR and FLAIR images for tumor segmentation. Overall, the U-Net produced the best similarity scores of 0.4±0.15 using MTR and FLAIR images. A future study will test the utility of 3D MTR images for tumor segmentation using  convolutional and full-connected models.

View Presentation Video

Chenghao Zhang¹, Mikhail Morgan¹, Yanting Yang¹, Ye Tian¹, and Jia Guo^1
1Columbia University, New York, NY, United States

Keywords: Machine Learning/Artificial Intelligence, Brain

Deformable image registration is a crucial part of the medical image process that seeks to estimate an optimal spatial transformation to align two images. Traditional methods neglect the inverse-consistent property and topology preservation of the transformation, which can lead to errors in registration results. To address these issues, we propose the cycle inverse consistent transformer-based deformable medical image registration model. We adopt Swin-UNet to achieve higher registration performance and comprehensively consider the inverse consistency loss function to guarantee a more accurate registration. Our pipeline can be trained to create study-specific templates of images for diagnostic and/or educational purposes. 

Liying An¹, Fanyang Meng¹, Kaixin Yu², Lei Zhang¹, and Huimao Zhang^1
1Radiology, The First Hospital of Jilin University, Changchun, China, ²Shukun (Beijing) Technology Co., Ltd, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Vessels

Accurate cerebrovascular segmentation plays an important role in clinical diagnosis and the related research. Magnetic Resonance angiography (MRA) postprocessing manually recognized by technologists is extremely labor intensive and error prone. However, automatic segmentation of brain vessels remains challenging because of the variable vessel shape and high complex of vessel geometry. We propose an artificial intelligence brain vessel segmentation system supported by an optimized physiological anatomical-based 3D convolutional neural network that can automatically achieve brain MRA vessel segmentaion in healthcare services. The overall segmentation accuracy of the independent testing dataset is 0.931.

View Presentation Video

Claudia Chinea Hammecher^1,2, Karin van Garderen^1,3, Marion Smits^1,3, Pieter Wesseling^4,5, Bart Westerman⁶, Pim French⁷, Mathilde Kouwenhoven⁸, Roel Verhaak^9,10, Frans Vos^1,2, Esther Bron¹, and Bo Li^1
1Department of Radiology & Nuclear Medicine, Erasmus MC, Rotterdam, Netherlands, ²Department of Imaging Physics, Delft University of Technology, Delft, Netherlands, ³Medical Delta, Delft, Netherlands, ⁴Department of Pathology, Amsterdam UMC / VU Medical Center, Amsterdam, Netherlands, ⁵Laboratory for Childhood Cancer Pathology, Princess Máxima Center for Pediatric Oncology, Utrecht, Netherlands, ⁶Department of Neurosurgery, Amsterdam UMC / VU Medical Center, Amsterdam, Netherlands, ⁷Department of Neurology, Erasmus MC Cancer Institute, Rotterdam, Netherlands, ⁸Department of Neurology, Amsterdam UMC / VU Medical Center, Amsterdam, Netherlands, ⁹The Jackson Laboratory for Genomic Medicine, Farmington, CT, United States, ¹⁰Department of Neurosurgery, Amsterdam UMC/VUmc, Amsterdam, Netherlands

Keywords: Machine Learning/Artificial Intelligence, Data Analysis, Image Registration

Glioma growth may be quantified with longitudinal image registration. However, the large mass-effects and tissue changes across images suposse an added challenge. Here, we propose a longitudinal, learning-based and groupwise registration method for the accurate and unbiased registration of glioma MRI. We evaluate on a dataset from the Glioma Longitudinal AnalySiS consortium and compare to classical registration methods. We achieve comparable Dice coeffients, with more detailed registrations, while significantly reducing the runtime to under a minute. The proposed methods may serve as an alternative to classical toolboxes, to provide further insight into glioma growth .