View Presentation Video

Siyue Li¹, Shutian Zhao¹, Fan Xiao², Dόnal G. Cahil³, Kevin Ki-Wai Ho⁴, Michael Tim-Yin Ong⁴, Queeie Chan⁵, James F Griffith³, Jin Hong⁶, and Weitian Chen^1
1CU Lab for AI in Radiology (CLAIR), Department of Imaging and Interventional Radiology, The Chinese University of Hong Kong, Hong Kong, China, ²Department of Radiology, Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shang hai, China, ³Department of Imaging and Interventional Radiology, The Chinese University of Hong Kong, Hong Kong, China, ⁴Department of Orthopaedics & Traumatology, The Chinese University of Hong Kong, Hong Kong, China, ⁵Philips Healthcare, Hong Kong, China, ⁶Guangdong Provincial People's Hospital, Guangdong Academy of Medical Sciences, Guang Zhou, China

Keywords: Segmentation, Machine Learning/Artificial Intelligence

Automatic meniscus segmentation is highly desirable for quantitative analysis of knee joint diseases. As three-dimensional Fast Spin Echo (3D FSE) is a promising MR (magnetic resonance) imaging technique to evaluate the tissues of the knee joint. In this study, we explore meniscal segmentation on 3D FSE MRI. Manually annotating 3D knee images is challenging since it is time-consuming and requires clinical expertise. In this study, we propose a domain adaption-based few-shot learning method for meniscal segmentation on 3D FSE images using only one annotated MRI data. We demonstrate that the proposed method outperformed the fully supervised segmentation model.

View Presentation Video

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

Keywords: Segmentation, Segmentation

Object motion during the acquisition of magnetic resonance images can negatively impact image quality by introducing inconsistencies in the k-space data and in turn, blurring and ghosting artifacts on the acquired images. Such artifacts represent significant challenges in the clinical deployment of automated segmentation algorithms. In this work, we present an ensemble of approaches aimed at developing a robust and generalizable segmentation model, particularly tailored to handle the appearance of respiratory motion artifacts. We achieve this by introducing k-space based simulation for augmentation, as well as by reducing common errors across basal and apical slices with a region-focused segmentation approach.

 

View Presentation Video

Jaime Barranco^1,2, Hamza Kebiri^1,2, Óscar Esteban³, Raphael Sznitman⁴, Oliver Stachs⁵, Sönke Langner^6,7, Benedetta Franceschiello^8,9,10, and Meritxell Bach Cuadra^1,2,10
1CIBM Center for Biomedical Imaging, Lausanne, Switzerland, ²Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, ³Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland, ⁴ARTORG Center for Biomedical Engineering, University of Bern, Bern, Switzerland, ⁵Ophthalmology, Rostock University Medical Center, Rostock, Germany, ⁶Institute for Diagnostic and Interventional Radiology, Rostock University Medical Center, Rostock, Germany, ⁷Diagnostic Radiology and Neuroradiology, University of Greifswald, Greifswald, Germany, ⁸School of Engineering, Institute of Systems Engineering, HES-SO Valais-Wallis, Sion, Switzerland, ⁹The Sense Innovation and Research Center, Lausanne and Sion, Switzerland, ¹⁰These authors provided equal last-authorship contribution, Lausanne, Switzerland

Keywords: Segmentation, Neuro, Eye, ophthalmology

This work comparatively evaluates two approaches for the automated segmentation of eye structures from 3D T1-weighted MRI data of the whole human head (N=1210). Quantitative results on a validation sub-set with manual annotations provide accurate results for lens and globe and set the first median Dice (DSC) benchmarks for optic-nerve (0.91), muscles (0.58 to 0.76) and fat (0.67 and 0.75). The ability of our framework to automatically extract state-of-the-art measurements, such as the axial length, paves the way to accurately identify and compute new biomarkers of the eye via MRI.

View Presentation Video

Carlos Velasco¹, Roman Jakubicek², Alina Hua¹, Anastasia Fotaki¹, René M. Botnar^1,3, and Claudia Prieto^1,3
1School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, ²Department of Biomedical Engineering, Brno University of Technology, Brno, Czech Republic, ³Institute for Biological and Medical Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile

Keywords: Segmentation, Myocardium

The high amount of data obtained from a single 3D whole heart multiparametric scan (up to ~40 slices per parametric map) increases considerably the time required to segment and analyse the quantitative maps. Thus, an automated segmentation tool for these maps is desirable to perform this otherwise prohibitively laborious task. In this work, we leverage the potential of nnU-Net to perform fast, automated segmentation of 3D whole-heart simultaneous T₁ and T₂ maps and show its feasibility to predict segmentation masks with comparable quality while shortening the segmentation and analysis time by ~100x.

Luxuan Guo¹, Simran Kukran¹, Krithika Balaji¹, and Neal Bangerter^1
1Imperial College London, London, United Kingdom

Keywords: Segmentation, Cartilage, Osteoarthritis

7T 3D knee MRI shows great promise for the quantification of cartilage volume and thickness to assess osteoarthritis, but manual segmentation is time consuming. Segmented 7T 3D knee datasets to train machine learning techniques are limited. Here, we trained a network on a larger dataset of segmented 3T 3D knee MRI scans and used transfer learning to create an automatic segmentation network of 7T MRI knee cartilage using a small number of segmented 7T images. The resulting network constructed demonstrated vastly improved automatic segmentation of the knee cartilage at 7T compared to the network trained with limited 7T data only.

Hang Yu¹, Yuru Guo¹, Lizhi Xie², Zhiheng Liu¹, Zichuan Xie³, Chenyang Li¹, and Suiping Zhou^1
1School of Aerospace Science and Technology，Xidian university, xi'an, China, ²GE Healthcare, Beijing, China, ³Guangzhou institute of technology，Xidian University, Guangzhou, China

Keywords: Segmentation, Breast

In breast MRI, overall breast segmentation is a key step in performing breast cancer risk assessment.To achieve automatic and accurate breast segmentation in breast MR images, we propose a breast segmentation model based on U-Net and multi-head self-attention mechanism, which adds global information through multi-head self-attention module and changes the cascade structure in U-Net network to pixel-by-pixel summation.On the breast MRI dataset, the proposed model can achieve accurate, effective and fast breast segmentation with an average DSC and an average MIOU of 97.28 % and 92.01 %, respectively, which are 4.7 % and 5.56 % higher compared to U-Net, respectively.

View Presentation Video

Junhui Huang^1,2, Zhou Liu³, Liyan Zou³, Yingying Chen³, Zhanli Hu⁴, Dong Liang⁴, Xin Liu⁴, Hairong Zheng⁴, and Na Zhang^4
1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, ShenZhen, China, ²Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming, China, ³Department of Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital & Shenzhen Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Shenzhen, China, ⁴Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

Keywords: Segmentation, Machine Learning/Artificial Intelligence

Accurate segmentation of nasopharyngeal tumor lesions from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) facilitates subsequent diagnosis and treatment. However, current segmentation methods do not incorporate the pathological properties of the tumor. Therefore, this paper proposes a multimodal DCE-MRI segmentation method that uses the pharmacokinetic features of NPC, K_trans, as modal information to assist nasopharyngeal tumor segmentation. We validated our method in several classical deep learning segmentation networks, and DCE-MRI with fused K_trans eigenmodes had higher Dice coefficients than DCE-MRI with a single modality. The best segmentation results were obtained by this method on the ResUNet model (dice=74.39).

View Presentation Video

Cheng Li^1,2, Yousuf Babiker M. Osman^1,3, Weijian Huang^1,3,4, Zhenzhen Xue^1,2, Hua Han^1,3, Hairong Zheng¹, and Shanshan Wang^1,2,4
1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, ²Guangdong Provincial Key Laboratory of Artificial Intelligence in Medical Image Analysis and Application, Guangzhou, China, ³University of Chinese Academy of Sciences, Beijing, China, ⁴Peng Cheng Laboratory, Shenzhen, China

Keywords: Segmentation, Body

T1-weighted in-phase and opposed-phase gradient-echo imaging is a routine component in abdominal MR imaging. Organ segmentation with the acquired images plays an important role in identifying various diseases and making treatment plans. Despite the promising performance achieved by existing deep learning models, further investigation is still needed to effectively exploit the information provided by different imaging parameters. Here, we propose a complementarity-aware multi-parametric MR image feature fusion network to extract and fuse the information of paired in-phase and opposed-phase MR images for enhanced abdominal multi-organ segmentation. Extensive experiments are conducted, and better results are achieved when compared to existing methods.

View Presentation Video

Karthikeyan Subramanian¹, Pavan Kumar Jella¹, Feifei Qu¹, Tinnakorn Chaiworapongsa^2,3, and Mark E Haacke^1
1Department of Radiology, Wayne State University, Detroit, MI, United States, ²Perinatology Research Branch, NICHD/NIH/DHHS, Detroit, MI, United States, ³Department of Obstetrics and Gynecology, Wayne State University, Detroit, MI, United States

Keywords: Segmentation, Fetus

Mapping the vasculature, flow and tissue properties of the placenta and umbilical cord can serve as a means to study placental and fetal health. The goal of this work is to use a rapid, multi-echo, interleaved GRE sequence to minimize motion artifacts and cover the entire abdomen of the mother in a few minutes. To better map out the vasculature, we propose to both separate the arteries and veins in the umbilical cord using an R2* mapping thresholding technique and use vessel tracking to create 3D renderings. We have successfully done this in a series of 10 fetuses. 

View Presentation Video

Soumick Chatterjee^1,2,3, Karthikesh Varma Chintalapati¹, Chethan Radhakrishna¹, Sri Chandana Hudukula Ram Kumar¹, Raviteja Sutrave¹, Hendrik Mattern³, Oliver Speck^3,4,5, and Andreas Nürnberger^1,2,5
1Faculty of Computer Science, Otto von Guericke University Magdeburg, Magdeburg, Germany, ²Data and Knowledge Engineering Group, Otto von Guericke University Magdeburg, Magdeburg, Germany, ³Department of Biomedical Magnetic Resonance, Otto von Guericke University Magdeburg, Magdeburg, Germany, ⁴German Center for Neurodegenerative Disease, Magdeburg, Germany, ⁵Center for Behavioral Brain Sciences, Magdeburg, Germany

Keywords: Segmentation, Blood vessels

Vessel Segmentation with deep learning is a challenging task that involves not only learning high-level feature representations but also the spatial continuity of the features across dimensions. Semi-supervised patch-based approaches have been effective in identifying small vessels of 1-2 voxels in diameter but failed to maintain vessel continuity. This study focuses on improving the segmentation quality by considering the spatial correlation of the features using the maximum intensity projection (MIP) as an additional loss criterion. It was observed that the proposed method quantitatively improves the segmentation while also improving vessel continuity, as evident in the visual examinations of ROIs.

View Presentation Video

Catarina Rua¹, Mari Lambrechts¹, Mark Tanner¹, James Davies¹, Ali Ghayoor², Howard Dobson², and Lino Becerra^2
1Invicro LLC, A Konica Minolta Company, London, United Kingdom, ²Invicro LLC, A Konica Minolta Company, Needham, MA, United States

Keywords: Segmentation, Neurofluids, CSF segmentation spine

Computational fluid dynamics (CFD) has been used to model the behavior of cerebral-spinal-fluid (CSF) flow along the spine in research of intrathecal drug delivery. However, the model geometry highly impacts the CFD-based prediction of CSF flow. Enhanced images of the spinal CSF can be obtained with T2-weighted MRI, typically collected in 3-4 stations. Here we present a method to stitch spine images for CSF segmentation and compare it to the scanner’s stitching technique. Our offline approach appears more robust to movement and field-of-view adjustments while scanning and presents acceptable CSF signal intensity homogeneity across the spine for tissue segmentation.

View Presentation Video

Jue Jiang¹, Ramesh Paudyal¹, Bill H. Diplas², James Han², Nadeem Riaz², Vaios Hatzoglou ³, Nancy Lee², Joseph Deasy ¹, Amita Shukla-Dave ^1,3, and Harini Veeraraghavan ^1
1Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, United States, ²Radiation Oncology, Memorial Sloan Kettering Cancer Center, New York, NY, United States, ³Radiology, Memorial Sloan Kettering Cancer Center, New York, NY, United States

Keywords: Segmentation, Cancer

Manual segmentation of normal and tumor tissues on MRI is a traditional approach that is still used, but it is a very challenging and time-consuming method requiring a high level of precision and has shown inter-reader contouring variability. Therefore, semi- or fully automated segmentation algorithms are essential to segment tumors such as neck nodal metastases. The present study aimed to apply the previously developed deep learning-based self-distilled masked image transformer method for auto-segmenting neck nodal metastases on longitudinal T₂-weighted MR images.

View Presentation Video

Ben Viggiano¹, Aashna Desai², Elka Rubin³, Andrew Schmidt³, Robert Boutin³, Kathryn J Stevens³, Garry E Gold³, Christopher Ré⁴, Akshay S Chaudhari^1,3, and Arjun D Desai^3,5
1Biomedical Data Science, Stanford University, Stanford, CA, United States, ²Department of Neuroscience, University of California Berkeley, Berkeley, CA, United States, ³Department of Radiology, Stanford University, Stanford, CA, United States, ⁴Department of Computer Science, Stanford University, Stanford, CA, United States, ⁵Department of Electrical Engineering, Stanford University, Stanford, CA, United States

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction, Segmentation, Classification

Deep learning (DL) techniques have shown promise for both reconstruction and image analysis stages of MRI workflows. However, traditional benchmarking methods evaluate each stage separately. As a result, the impact of reconstruction on downstream image analysis tasks and biomarker quantification remains unknown. In this study, we explore how changing aspects of upstream reconstruction affects the downstream analysis. We find that insights from evaluating reconstruction models as a component of a broader end-to-end workflow do not correlate with conventional, task-specific image quality metrics. We use these findings to motivate the discussion of evaluating DL methods at the workflow level.

View Presentation Video

Ahmed Gouda¹, Saqib Basar¹, Yosef Chodakiewitz², Rajpaul Attariwala¹, Sean London², and Sam Hashemi^1
1Voxelwise Imaging Technology Inc., Vancouver, BC, Canada, ²Prenuvo, Vancouver, BC, Canada

Keywords: Data Analysis, Aging, Volumetric Analysis, Organs, Visceral Fat, Skeletal Muscles

One of the vital indicators of normal functioning physiology and anatomy, is the volume and size of body organs and muscles. Overtime due to aging and chronic illnesses, the volume of organs and muscles decreases, and quantifying this reduction would help us in predicting the rate of decline, and making efforts to improve this rate. Artificial intelligence 3D segmentation techniques allow us to measure the average age-related volume loss over a large population. They provide statistical norms which help radiologists to identify abnormal volume changes, and establish a normal age related rate of volumetric decline.

View Presentation Video

Zhenxing Huang¹, Mengxiao Geng¹, Liyun Zheng², Yongming Dai³, Na Zhang¹, Dong Liang¹, Hairong Zheng¹, and Zhanli Hu^1
1Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, ²Shenzhen United Imaging Research Institute of Innovative Medical Equipment, Shenzhen, China, ³Central Reasearch Institute, United Imaging Healthcare, Shanghai, China

Keywords: Data Analysis, Segmentation, 3T/5T MRI

Image segmentation is a complex and core technique in the medical image domain. However, low-quality images, such as images with weak edges, may bring considerable challenges for radiologists. In this paper, we propose an adaptive weighted curvature-based active contour model by coupling heat kernel convolution and adaptively weighted high-order total variation to improve diagnosis effectiveness. The numerical experimental results on 3T/5T MRI datasets demonstrate that the proposed model is quite efficient and robust compared with several traditional segmentation methods, which would exert great value in quantitative image evaluation of MRI diagnosis for the same person.

View Presentation Video

Xiaodi Li¹, Peng Li¹, and Yue Hu^1
1Harbin Institute of Technology, Harbin, China

Keywords: Data Processing, Segmentation

Due to the inherent locality of convolutional operations, convolution neural network (CNN) often exhibits limitations in explicitly modeling long-distance dependencies. In this paper, we propose a novel hybrid multi-level graph neural (HMGN) network that combines the CNN and graph neural network to capture both local and non-local image features at multiple scales. With the proposed patch graph attention module, the HMGN network can capture image features over a large receptive field, resulting in more accurate segmentation of cardiac structures. Experiments on two public datasets show the proposed method obtains improved segmentation performance over the state-of-the-art methods.

View Presentation Video

Dang Bich Thuy Le¹, Ram Narayanan¹, Meredith Sadinski¹, Aleksandar Nacev¹, and Srirama Venkataraman^1
1Research and Design, Promaxo Inc, Oakland, CA, United States

Keywords: Radiomics, Cancer, Prostate

By quantifying pixel relationships from frequencies of local signal intensity spatial variations, Haralick texture features have shown promise for prostate cancer detection. In this study, axial, T2-weighted MR images combined with extracted Haralick texture feature maps were used in a deep learning framework to identify lesion locations and predict Gleason Grade. Results demonstrate potential of Haralick texture features to segment and classify prostate lesions with AUC/sensitivity/specificity of 0.87/0.923/0.776 on patient-level evaluation.

View Presentation Video

Francesco Santini^1,2, Jakob Wasserthal², Abramo Agosti³, Xeni Deligianni^1,2, Kevin R Keene⁴, Hermien E Kan⁵, Stefan Sommer^6,7,8, Christoph Stuprich⁹, Fengdan Wang¹⁰, Claudia Weidensteiner^1,11, Giulia Manco¹², Valentina Mazzoli¹³, Arjun Desai¹⁴, and Anna Pichiecchio^12,15
1Basel Muscle MRI, Department of Biomedical Engineering, University of Basel, Basel, Switzerland, ²Research Coordination Team, Department of Radiology, University Hospital Basel, Basel, Switzerland, ³Department of Mathematics, University of Pavia, Pavia, Italy, ⁴Department of Neurology, Leiden University Medical Center, Leiden, Netherlands, ⁵C.J. Gorter MRI Centre, Department of Radiology, Leiden University Medical Center, Leiden, Netherlands, ⁶Siemens Healthineers International AG, Zurich, Switzerland, ⁷Swiss Center for Musculoskeletal Imaging (SCMI), Balgrist Campus, Zurich, Switzerland, ⁸Advanced Clinical Imaging Technology (ACIT), Siemens Healthineers International AG, Lausanne, Switzerland, ⁹University Hospital Erlangen, Erlangen, Germany, ¹⁰Peking Union Medical College, Beijing, China, ¹¹Radiological Physics, Department of Radiology, University Hospital Basel, Basel, Switzerland, ¹²Advanced Imaging and Radiomics Center, IRCCS Mondino Foundation, Pavia, Italy, ¹³Department of Radiology, Stanford University, Stanford, CA, United States, ¹⁴Departments of Electrical Engineering & Radiology, Stanford University, Stanford, CA, United States, ¹⁵Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy

Keywords: Software Tools, Machine Learning/Artificial Intelligence

An open-source, federated-learning-based segmentation software termed Dafne (Deep Anatomical Federated Network) is presented. This software continuously adapts the deep learning models used for the segmentation (currently for the muscles of the leg and thigh) based on the input of the users, who are in multiple institutions. This software was validated through data usage statistics of more than 50 users and through a retrospective study on 38 datasets of patients with suspected myositis, showing that the continuous learning approach is able to improve and generalize the performance of the original models.

View Presentation Video

Ahmed Gouda¹, Saqib Basar¹, Yosef Chodakiewitz², Rajpaul Attariwala¹, Sean London², and Sam Hashemi^1
1Voxelwise Imaging Technology Inc., Vancouver, BC, Canada, ²Prenuvo, Vancouver, BC, Canada

Keywords: Liver, Fat, Hepatic Steatosis Quantification, Iron Deposition Detection, Dixon, FSF, Dual-echo MRI

Over the years, Magnetic Resonance Imaging (MRI) has become the optimal noninvasive method to quantify liver steatosis and to detect hepatic iron deposition. The conventional manual sampling technique of liver fat quantification at multiple regions is complex, inefficient and a time-consuming process. In addition, it may produce varying results for the heterogeneous fat deposition, which is prone to radiologist’s subjectivity. In this research, we propose a fully automated artificial intelligence (AI) based method for hepatic steatosis quantification and hepatic iron deposition detection using whole liver and spleen volume segmentation.

View Presentation Video

Adam DESCARPENTRIES¹, Florence FERET², and Julien ROUYER^1
1Research and Innovation Department, Olea Medical, La Ciotat, France, ²Clinical solution department, Olea Medical, La Ciotat, France

Keywords: Breast, Cancer, CAD

The aim of our work is to detect any enhancing object of interest for reporting purposes. A deep learning approach combined with a multi-constructor and multi-centric database enabled to initiate the development of a versatile tool in line with clinical real life. The detection problem was addressed using a two-stage three-dimensional cascaded U-Net architecture. A total of 610 single-breast images were used for the model development. Results present interesting  score in term of Dice similarity index (0.83) which agree well with the recent literature. Discussion section focuses on the potential benefit in the use of a recently reported loss function.