View Presentation Video

Matthew L. Sala¹, Jan Lost², Niklas Tillmanns², Sara Merkaj³, Marc von Reppert⁴, Divya Ramakrishnan¹, Khaled Bousabarah⁵, Anita Huttner⁶, Sanjay Aneja⁷, Arman Avesta⁷, Antonio Omuro⁸, and Mariam Aboian^1
1Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, CT, United States, ²University of Düsseldorf, Düsseldorf, Germany, ³University of Ulm, Ulm, Germany, ⁴Leipzig University, Leipzig, Germany, ⁵Visage Imaging, Düsseldorf, Germany, ⁶Pathology, Yale School of Medicine, New Haven, CT, United States, ⁷Therapeutic Radiology, Yale School of Medicine, New Haven, CT, United States, ⁸Neurology, Yale School of Medicine, New Haven, CT, United States

Keywords: Tumors, Machine Learning/Artificial Intelligence, Glioma

Recent development of Machine Learning (ML) tools for analysis of CNS tumors demonstrates great potential benefit to research and clinical practice but has been hindered by a lack of external validation. There is a critical need for open access to large individual hospital-based datasets with expert annotations. Here, we present the Yale Glioma Dataset, a database of 1,033 patients featuring annotated segmentations on FLAIR and T1 post-gadolinium, tumor grading and classification, and further clinical information. Open access of this database will support the development and validation of new AI algorithms for glioma detection and segmentation.

View Presentation Video

Yunzhi Xu¹, Jiaxin Li¹, Xue Feng², Kun Qing³, Dan Wu¹, and Li Zhao^1
1Zhejiang University, Hangzhou, China, ²Biomedical Engineering, University of Virginia, Virginia, VA, United States, ³Department of Radiation Oncology, City of Hope National Center, Los Angeles, CA, United States

Keywords: Machine Learning/Artificial Intelligence, Fetus, Fetal Brain MRI，Point Spread Function

 High apparent resolution of fetal MRIs is provided by slice-to-volume reconstruction pipelines widely. However, the physical resolution of the fetal brain is lower than that. Therefore, we hypothesize that fetal brain segmentation can be performed based on downsampled fetal brain MRI according to its point spread function. In this work, 150 adult brain and 80 fetal brain MRIs were used to validate hypothesize. Using downsampled fetal data with factor of 4, a highly efficient segmentation model achieved similar segmentation accuracies compared to original data, which demonstrated that segmentation models can be developed based on PSF.

 

View Presentation Video

Jakob Meglič^1,2, Mohammed R. S. Sunoqrot^1,3, Tone F. Bathen^1,3, and Mattijs Elschot^1,3
1Department of Circulation and Medical Imaging, Norwegian University of Science and Technology (NTNU), Trondheim, Norway, ²Faculty of Medicine, University of Ljubljana, Ljubljana, Slovenia, ³Department of Radiology and Nuclear Medicine, St. Olavs Hospital, Trondheim University Hospital, Trondheim, Norway

Keywords: Machine Learning/Artificial Intelligence, Prostate

Prostate segmentation is an essential step in computer-aided diagnosis systems for prostate cancer. Deep learning (DL)-based methods provide good performance for prostate segmentation, but little is known about the impact of ground truth (manual segmentation) selection. In this work, we investigated these effects and concluded that selecting different segmentation labels for the prostate gland and zones has a measurable impact on the segmentation model performance.

Zhengyong Huang^1,2, Na Zhang¹, Dong Liang¹, Xin Liu¹, Hairong Zheng¹, and Zhanli Hu^1
1Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, ²University of Chinese Academy of Sciences, Beijing, China

Keywords: Machine Learning/Artificial Intelligence, Data Processing, MRI medical segmentation

Semi-supervised segmentation, using large amounts of unlabeled data and small amounts of labeled data, has achieved great success. This paper proposes a semi-supervised segmentation method based on consistent learning and contrast learning. It mainly uses a mean-teacher framework to add consistency losses and contrast losses based on multiscale features to minimize the distance of model responses under different disturbance inputs. In addition, mean square error loss was used to alternately minimize the gap between the teacher and student models. In 3D left atrium data, a Dice coeffivient of 0.8970 was obtained, which was superior to other methods.

View Presentation Video

Haoyu Lan¹, Arthur W. Toga¹, and Jeiran Choupan^1,2
1University of Southern California, Los Angeles, CA, United States, ²NeuroScope Inc., Scarsdale, NY, United States

Keywords: Machine Learning/Artificial Intelligence, Segmentation, self-supervised learning

Due to the absence of isotropic T2w modality in clinical datasets, it is challenging to enhance the PVS contrast using multiple neuroimage modalities. To overcome this issue, in this work we introduced using self-supervised pre-trained model in the enhanced PVS contrast image space to improve the downstream model segmentation performance when solely using T1w as the training data. The experiment results showed that the proposed method increased segmentation accuracy compared to the model trained from scratch using T1w modality and resulted in faster training and less required training data volume.

View Presentation Video

Joonhyeok Yoon¹, Juhyung Park¹, Yoonho Nam², Chul-Ho Sohn^3,4, Junghwa Kang², Jonghyo Youn¹, Sooyeon Ji¹, Chungseok Oh¹, and Jongho Lee^1
1Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea, Republic of, ²Department of Biomedical Engineering, Hankuk University of Foreign Studies, Gyeonggi-do, Korea, Republic of, ³Institute of Radiation Medicine, Seoul National University Medical Research Center, Seoul, Korea, Republic of, ⁴Department of Radiology, Seoul National University Hospital, Seoul, Korea, Republic of

Keywords: Machine Learning/Artificial Intelligence, Parkinson's Disease, Neuromelanin, Denoising

Neuromelanin (NM) has been considered an associated biomarker of Parkinson’s disease (PD). Conventional NM visualizeing techniques requires about 5~10min which is sub-optimal for scanning PD patient with movement disorders. Recently, SandwichNM is reduced scan time to 5m 30s, but it is still may not enough.

In this research, we approach this issue in the viewpoint of image post processing with denoising techniques to reduce scan time. After 162 NM MRI data acquisition with short scan time, we demonstrated that using denoising technique can improve the distinguishability for NM segmentation.

View Presentation Video

Dilek Mirgun Yalcinkaya^1,2, Khalid Youssef³, Bobak Heydari⁴, Subha Raman^3,5, Rohan Dharmakumar^3,5, and Behzad Sharif^1,3,5
1Laboratory for Translational Imaging of Microcirculation, Indiana University (IU) School of Medicine, Indianapolis, IN, United States, ²Elmore Family School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, United States, ³Krannert Cardiovascular Research Center, IU School of Medicine/IU Health Cardiovascular Institute, Indianapolis, IN, United States, ⁴Stephenson Cardiac Imaging Centre, Department of Cardiac Sciences, University of Calgary, Calgary, AB, Canada, ⁵Weldon School of Biomedical Engineering, Purdue University, West Lafayette, IN, United States

Keywords: Machine Learning/Artificial Intelligence, Segmentation

We proposed and validated Data Adaptive Uncertainty-Guided Spatiotemporal (DAUGS) analysis that leverages the data-driven uncertainty map of the segmentation contours among a pool of trained deep neural networks (DNNs) and automatically selects the segmentation result with the highest level of certainty. Our results suggest that proposed DAUGS and standard DNN-based analysis demonstrated on-par performance on the internal test set which is from the same institution as training set and acquired with FLASH sequence. In contrast, DAUGS analysis considerably outperformed DNN-based analysis on the external test set which was acquired with a bSSFP pulse sequence at a different institution, demonstrating the improved robustness of the proposed method despite limited training data.

Kexin Wang^1
1Capital Medical University, Beijing, China

Keywords: Prostate, Machine Learning/Artificial Intelligence

In this study we evaluate the generalization of the AI algorithms for the classification of the mpMRI image sequences and the segmentation of the prostate gland with multicenter external dataset. A total of 719 patients who underwent multiparametric MRI (mpMRI) of the prostate were collected retrospectively from two hospitals. AI algorithms were tested for classification of the image type and segmentation of the prostate gland. The AI models demonstrated good performance in the external validation in the task of image classification and prostate gland segmentation.

View Presentation Video

Xing Lu¹, Yajun Ma¹, Jiyo Athertya¹, Chun-Nan Hsu², Eric Y Chang^1,3, Amilcare Gentili^1,3, Christine Chung¹, and Jiang Du^1
1Department of Radiology, University of California, San Diego, San Diego, CA, United States, ²Department of Neurosciences, University of California, San Diego, San Diego, CA, United States, ³Radiology Service, Veterans Affairs San Diego Healthcare System, San Diego, CA, United States

Keywords: Osteoarthritis, Quantitative Imaging

To accelerate ultrashort echo time (UTE) based multi-parameter quantitative MRI (qMRI), we performed comparison studies on the input MRI image numbers and the effects on the prediction quality of the MSMQ-Net. The MSMQ-net was modified and trained accordingly with different combinations of inputs. Both image similarity and regional analysis were evaluated. The results demonstrate that 90% accuracy can be achieved for both UTE-T1 and UTE-T1rho mapping when the scan time is reduced by 75%.

View Presentation Video

Kexin Jiang¹, Yuhan Xie², Zhiyong Zhang², Jiaping Hu¹, Shaolong Chen², Zhongping Zhang³, Changzhen Qiu², and Xiaodong Zhang^1
1Department of Medical Imaging, The Third Affiliated Hospital of Southern Medical University, GuangZhou, China, ²Electronics and Communication Engineering, Sun Yat-sen University, GuangZhou, China, ³Philips Healthcare, GuangZhou, China

Keywords: Joints, Joints, meniscus

Quantitative MRI of meniscus morphology (such as MOAKS system) have shown clinical relevance in the diagnosis of osteoarthritis. However, it requires a large workload, and often lead to deviation due to reader’s subjectivity. Therefore, based on automated segmentation of six horns of meniscus using fully and weakly supervised networks, we established two-layer cascaded classification models that can detect the meniscal lesions and further classify them into three types, and finally achieved excellent performance. This can improve the efficiency and accuracy of using quantitative MRI to study KOA in the future.

View Presentation Video

Fares OUADAHI¹, Aurélie PETER¹, Anaïs BERNARD¹, and Julien ROUYER^1
1Research and Innovation Department, Olea Medical, La Ciotat, France

Keywords: Breast, Data Processing, Registration, Motion correction

This work is part of an ongoing study to correct accidental breast motion during dynamic contrast-enhanced imaging. Here, we explored a finite element method to generate “moving” images based on a “fixed” one. The proposed methodology can help in the constitution of realistic dataset with large variety of motion to support the development of a dedicated non-rigid registration model by deep learning.

Hang Yu¹, Zichuan Xie², Lizhi Xie³, Zhiheng Liu¹, Lina Zhang⁴, Siyao Du⁴, Xiangjie Yin¹, Chenyang Li¹, Wenhong Jiang⁴, Yuru Guo¹, and Zhongqi Kang^4
1School of Aerospace Science and Technology, Xidian University, Xi'an, China, ²Guangzhou institute of technology, Xidian University, Guangzhou, China, ³GE Healthcare, Beijing, China, ⁴Department of Radiology, The First Hospital of China Medical University, Shenyang, China

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

The multimodal MRI data is often ultilized for breast cancer analysis, and by now still difficult and inefficient to explored by segmentation algorithms. In this paper, we propose a MP-Unet based on U-net convolutional neural network, which can effectively ensemble multiple inputs of modal data and obtain accuracy segmentation results at the end. In MP-Unet, we reused some good quality modal data for training. The multiple MP-Unet models are further integrated based on Bagging algorithm to improve the segmentation accuracy of lesions. Experiments suggest that our proposed method has a huge performance improvement.

View Presentation Video

Subin Erattakulangara¹, Sarah Gerard¹, Karthika Kelat¹, Katie Burnham², Rachel Balbi², David Meyer², and Sajan Goud Lingala^1,3
1Roy J Carver Department of Biomedical Engineering, University of Iowa, iowa city, IA, United States, ²Janette Ogg Voice Research Center, Shenandoah University, Winchester, IA, United States, ³Department of Radiology, University of Iowa, iowa city, IA, United States

Keywords: Machine Learning/Artificial Intelligence, Head & Neck/ENT

The investigation into the 3D airway area is the prerequisite step for quantitatively studying the anatomical structures and function of the upper airway. Segmentation of upper airway can be considered as one of the stepping stones for this investigation. In this work, we propose a transfer learning-based 3D U-Net model with a ResNet encoder for vocal tract segmentation with small datasets training. We demonstrate its utility on sustained volumetric vocal tract MR scans from the recently released French speaker database.