View Presentation Video

Louisa Fay^1,2, Bin Yang², Sergios Gatidis^1,3, and Thomas Kuestner^1
1Medical Image and Data Analysis (MIDAS.lab), Department of Diagnostic and Interventional Radiology, University Hospital of Tuebingen, Tuebingen, Germany, ²Institute of Signal Processing and System Theory, University of Stuttgart, Stuttgart, Germany, ³Max Planck Institute for Intelligent Systems, Tuebingen, Germany

Keywords: Machine Learning/Artificial Intelligence, Data Analysis

Deep Learning methods can detect patterns in data such as MR images but are incapable of determining causal relationships. However, causal understanding is crucial in medical applications, since the presence of confounders (e.g. scan conditions) obscure the causal relationship and create spurious-correlations. State-of-the-art models purely rely on correlated patterns which can result in wrong conclusions or diagnoses when spurious-correlations change (e.g. new scanner). We propose a deep learning framework that is robust in the presence of spurious-correlations by decreasing mutual information between learned features of MR images and leads to improved performance under distribution shifts.

View Presentation Video

Simon Graf¹, Nora Küchler¹, Walter Wohlgemuth¹, and Andreas Deistung^1
1University Hospital Halle (Saale), Halle (Saale), Germany

Keywords: Machine Learning/Artificial Intelligence, Quantitative Susceptibility mapping

Deep learning models used for solving dipole inversion of quantitative susceptibility mapping typically lack an integration of a-priori information. We show for the first time that information of voxel-size and field-of-view orientation with respect to B₀ can be incorporated into network models by deploying adaptive convolution. Various network models were trained for 170 epochs to solve dipole inversion on synthetic data with arbitrary orientation and voxel-sizes. Adaptive convolution models outperform conventional models in computing susceptibility maps from arbitrarily oriented field distributions with anisotropic voxel sizes and allows a reduction of training time.

View Presentation Video

Mevan Ekanayake^1,2, Zhifeng Chen^1,3, Kamlesh Pawar¹, Gary Egan^1,4, Mehrtash Harandi², and Zhaolin Chen^1,3
1Monash Biomedical Imaging, Monash University, Melbourne, Australia, ²Department of Electrical and Computer Systems Engineering, Monash University, Melbourne, Australia, ³Department of Data Science and AI, Monash University, Melbourne, Australia, ⁴School of Psychological Sciences, Monash University, Melbourne, Australia

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction, self-supervised learning, contrastive learning, accelerated reconstruction, fastMRI

Most deep learning methods for MR image reconstruction heavily depend on supervised learning on fully sampled reference data, hence, lack generalizability on out-of-distribution inputs such as higher accelerations. To reduce the models’ dependency on fully sampled reference data and to leverage large cohorts of undersampled MR measurements, we propose a self-supervised framework that extracts contrastive features between different accelerations of a given scan and the rest of the scans in the dataset, which we then utilize for the downstream reconstruction task. Our quantitative and qualitative analysis demonstrates the superiority of the proposed framework for highly accelerated MR image reconstruction.

View Presentation Video

Lavanya Umapathy^1,2, Taylor Brown^2,3, Mark Greenhill^2,3, J'rick Lu^2,3, Diego Martin⁴, Maria Altbach², and Ali Bilgin^1,2,5,6
1Department of Electrical and Computer Engineering, University of Arizona, Tucson, AZ, United States, ²Department of Medical Imaging, University of Arizona, Tucson, AZ, United States, ³College of Medicine, University of Arizona, Tucson, AZ, United States, ⁴Department of Radiology, Houston Methodist Hospital, Houston, TX, United States, ⁵Program in Applied Mathematics, University of Arizona, Tucson, AZ, United States, ⁶Department of Biomedical Engineering, University of Arizona, Tucson, AZ, United States

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

The availability of limited labeled data motivates the use of self-supervised pretraining techniques for deep learning (DL) models. Here, we propose a novel contrastive loss that pushes/pulls local representations within an image based on representational constraints from co-registered multi-contrast MR images that share similar underlying parameters. For multi-organ segmentation tasks in T2-weighted images, pretraining a DL model using the proposed loss function with constraints from co-registered echo images from a radial TSE acquisition, can help reduce annotation burden by 60%. On two independent datasets, proposed pretraining improved Dice scores compared to random initialization and pretraining with conventional contrastive loss.

View Presentation Video

Onat Dalmaz^1,2, Muhammad Usama Mirza^1,2, Gokberk Elmas^1,2, Muzaffer Ozbey^1,2, Salman UH Dar^1,2, Emir Ceyani³, Salman Avestimehr³, and Tolga Çukur^1,2,4
1Department of Electrical and Electronics Engineering, Bilkent University, Ankara, Turkey, ²National Magnetic Resonance Research Center (UMRAM), Bilkent University, Ankara, Turkey, ³University of Southern California, Los Angeles, CA, United States, ⁴Neuroscience Program, Aysel Sabuncu Brain Research Center, Bilkent University, Ankara, Turkey

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

MRI contrast translation enables image imputation for missing sequences given acquired sequences in a multi-contrast protocol. Training of learning-based translation models requires access to large, diverse datasets that are challenging to aggregate centrally due to patient privacy risks. Federated learning (FL) is a promising solution that mitigates privacy concerns, but naive FL methods suffer from performance losses due to implicit and explicit data heterogeneities. Here, we introduce a novel FL-based personalized MRI translation method (pFLSynth) that effectively addresses implicit and explicit heterogeneity in multi-site datasets. FL experiments conducted on multi-contrast MRI datasets show the effectiveness of the proposed approach.

View Presentation Video

Tahsin Rahman¹, Sergio D. Cabrera¹, and Ali Bilgin^2
1Electrical and Computer Engineering, The University of Texas at El Paso, El Paso, TX, United States, ²Electrical and Computer Engineering, University of Arizona, Tucson, AZ, United States

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Vision transformers (ViT) are increasingly utilized in computer vision and have been shown to outperform CNNs in many tasks. In this work, we explore the use of Shifted Window (Swin) transformers for accelerated MRI reconstruction.  Our proposed SwinV2-MRI architecture enables the use of multi-coil data and k-space consistency constraints with Swin transformers. Experimental results show that the proposed architecture outperforms CNNs even when trained on a limited dataset and without any pre-training.

View Presentation Video

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

With the ageing of the population, Parkinson's disease (PD) has presented a severe challenge to public health. Here, a deep-learning framework named the AMDGM model was proposed to predict PD patients at an early stage. Firstly, multi-modal image-based models were respectively generated using the AMDGM model. Then, a weighted ensemble network was created as the final model. The proposed method achieved the best AUC performance of 0.872 in the testing cohort, better than others. And the proposed method can predict PD patients early to help clinical radiologists formulate more targeted treatments in the future. 

View Presentation Video

Hwihun Jeong¹, Dong Un Kang¹, Jiye Kim¹, and Jongho Lee^1
1Department of electrical and computer engineering, Seoul national university, Seoul, Korea, Republic of

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

We propose MRFlow, which is a normalizing flow-based neural network for the MRI harmonization framework. With the normalizing flow trained only with the target domain (e.g., 3T image) data, we harmonize the image from the source domain (e.g., 1.5T image) to the target domain by alternately reducing the norm of the latent variable and increasing similarity between the source domain and harmonized images. When MRFlow is applied to synthesized source domain images, the harmonized images showed lower errors than the source domain images. In the prospective study, the harmonized images became more similar to the target domain images after MRFlow.

View Presentation Video

Redha Ali¹, Hailong Li¹, Huixian Zhang¹, Wen Pan², Scott B. Reeder³, David T. Harris³, William R. Masch⁴, Anum Alsam⁴, Krishna P. Shanbhogue⁵, Anas Bernieh⁶, Sarangarajan Ranganathan⁶, Nehal A. Parikh⁶, Jonathan R. Dillman¹, and Lili He^1
1Department of Radiology, Cincinnati children's hospital medical center, Cincinnati, OH, United States, ²Department of Radiology, Cincinnati children's hospital medical center, 45429, OH, United States, ³University of Wisconsin-Madison, Madison, WI, United States, ⁴Michigan Medicine, University of Michigan, Ann Arbor, MI, United States, ⁵New York University Langone Health, New York, NY, United States, ⁶Cincinnati children's hospital medical center, Cincinnati, OH, United States

Keywords: Machine Learning/Artificial Intelligence, Liver

Magnetic resonance elastography (MRE) provides a noninvasive method to quantify liver stiffening, a surrogate biomarker for monitoring liver fibrosis. However, the availability of MRE remains limited, especially outside the United States, in part due to cost. This study aims to develop a deep learning-based approach for stratifying liver stiffness using multiparametric MRI images from pediatric and adult patients from multiple sites. We performed multi-site ten-fold cross-validation and achieved an AUROC of 0.80 for liver stiffness stratification. These results demonstrate that our proposed deep learning model may provide a means for categorical estimation of liver stiffening without dedicated elastography.

View Presentation Video

Wei Jiang¹, Yang Gao¹, and Hongfu Sun^1
1University of Queensland, Brisbane, Australia

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction, MRI, diffusion models

High-resolution MR images are advantageous for medical diagnosis. However, they may require longer scan time and more powerful hardware. Significant effort has been made to carry out super-resolution methods to synthesize higher-resolution MR images from lower-resolution acquisitions using deep learning approaches. However, most current methods are biased. In this work, we propose a new super-resolution method based on the state-of-the-art generative model (i.e., variational diffusion model), using the lower-resolution K-space measurements as a condition for guiding the super-resolution process. This unsupervised approach achieved high performance for reconstructing MR images from arbitrary lower resolutions without retraining the model.

View Presentation Video

Gawon Lee¹, Jun-Hyeok Lee¹, Dong Hye Ye², and Se-Hong Oh^1
1Biomedical Engineering, Hankuk University of Foreign Studies, Yongin-si, Korea, Republic of, ²Computer Science, Georgia State University, Atlanta, GA, United States

Keywords: Machine Learning/Artificial Intelligence, Brain, Data Harmonization

For multi-site research in MR studies, a harmonization process is necessary. Due to the lack of paired dataset for harmonization, the unsupervised-based learning has been suggested. However, this method is susceptible to causing the error in anatomical information and performing well in the unseen domain. Thus, we suggest a deep neural network for multi-site harmonization based on MR signal physics with the paired traveling dataset. We implemented Quantitative Maps Generators and Denoising Network with the Bloch Equation module. We proved that using the Bloch Equation enhances the accuracy of harmonization.

View Presentation Video

Keywords: Machine Learning/Artificial Intelligence, Cancer, federated learning

Prostate cancer screening and diagnosis from MRI is extremely challenging, and current machine learning algorithms suffer in cross-institutional generalizability. Federated learning is a way to alleviate these issues by combining multi-center data without aggregating or homogenizing data. To enable this for prototype-stage algorithms, we introduce FLtools, a lightweight python library with re-usable federated learning components available freely at https://federated.ucsf.edu. We use this federated learning system to train a 3D UCNet on bi-parametric MRI and paired prostate biopsy data from two University of California hospitals, demonstrating dramatic improvements in cross-site generalization accuracy in clinically-significant lesion classification.

View Presentation Video

Muzaffer Ozbey^1,2, Onat Dalmaz^1,2, Salman UH Dar^1,2, Hasan Atakan Bedel^1,2, Şaban Öztürk^1,2, Alper Güngör^1,2, and Tolga Çukur^1,2,3
1Department of Electrical and Electronics Engineering, Bilkent University, Ankara, Turkey, ²National Magnetic Resonance Research Center (UMRAM), Bilkent University, Ankara, Turkey, ³Neuroscience Program, Aysel Sabuncu Brain Research Center, Bilkent University, Ankara, Turkey

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

Synthesis of missing contrasts in an MRI protocol via translation from acquired contrasts can reduce costs associated with prolonged exams. Current learning-based translation methods are predominantly based on generative adversarial networks (GAN) that implicitly characterize the distribution of the target contrast, with limits fidelity of synthesized images. Here we present SynDiff, a novel conditional adversarial diffusion model for computationally efficient, high-fidelity contrast translation.  SynDiff enables training on unpaired datasets, thanks to its cycle-consistent architecture with coupled diffusion processes. Demonstrations on multi-contrast MRI datasets indicate the superiority of SynDiff against competing GAN and diffusion models. 

View Presentation Video

Keywords: Bioeffects & Magnetic Fields, Safety

Both the Tier-3 and Tier-4 methods proposed in ISO/TS 10974 for assessing the radiofrequency safety of active implantable medical devices have limitations in practice. The transfer function proposed in the Tier-3 method can only be measured individually for each implant and only for homogeneous tissue environments, while Tier-4 requires significant high-speed computing resources. In this study, machine learning algorithms are proposed to predict the transfer function both in homogeneous and heterogeneous tissue environment. The results show that this approach is feasible and performs well in the prediction task.

View Presentation Video

Sidharth Kumar¹, Asad Aali¹, and Jonathan I Tamir^1,2,3
1Chandra Family Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, TX, United States, ²Oden Institute for Computational Engineering and Sciences, University of Texas at Austin, Austin, TX, United States, ³Department of Diagnostic Medicine, University of Texas at Austin, Austin, TX, United States

Keywords: Machine Learning/Artificial Intelligence, Image Reconstruction

Score-based generative modeling has emerged as a powerful tool for modeling image priors and has recently been used to solve ill-posed inverse problems in various domains including MRI reconstruction. Here we extend the framework to reconstruct multi-contrast 3D fast spin-echo (FSE), i.e. T2 Shuffling data. This is achieved by constraining the posterior sampling reconstruction to a low-dimensional subspace and training a score model on images from this subspace. We demonstrate a proof-of-principal reconstruction of data with no model mismatch, i.e. generated from the forward model.