Yan Wu¹, Marc Alley ¹, Keshav Datta¹, Zhitao Li ¹, Christopher Sandino ¹, Zhifei Wen², Michael Lustig³, John Pauly¹, and Shreyas Vasanawala^1
1Stanford University, Stanford, CA, United States, ²Hoag Hospital, Newport Beach, CA, United States, ³University of California, Berkeley, Berkeley, CA, United States

We design a data-driven method to generate water/fat images from dual-echo complex Dixon images, aimed at near-instant water-fat separation with high robustness. A hierarchical convolutional neural network is employed, where ground truth images are obtained using a binary quadratic optimization approach. With IRB approval and informed consent, 9281 image sets are collected from 30 pediatric patients to train and test networks, with the application of six-fold cross validation. In addition to high fidelity and significantly reduced processing time, the predicted images are superior to the ground truth in mitigation of water/fat swaps and correction of artifacts introduced by metallic implants. 

Bragi Sveinsson^1,2 and Matthew S Rosen^1,2,3
1Radiology, A. A. Martinos Center, Massachusetts General Hospital, Boston, MA, United States, ²Radiology, Harvard Medical School, Boston, MA, United States, ³Physics, Harvard University, Cambridge, MA, United States

With an aging population, decreased muscle mass from conditions such as sarcopenia can be expected to lead to frequently decreased mobility, lowering patient quality of life. To develop effective treatments to slow down mobility loss, it is essential to obtain robust, objective mobility measurements that ideally do not require patient tasks. In this work, we explore the feasibility of predicting patient mobility by applying a neural network on sagittal knee MR images and accelerometry data from the Osteoarthritis Initiative.

Chaojie Zhang¹, Haresh Rengaraj Rajamohan², Kyunghyun Cho², Gregory Chang¹, and Cem M. Deniz^1
1Department of Radiology, New York University Langone Health, New York, NY, United States, ²Center for Data Science, New York University, New York, NY, United States

Vision transformers were used to predict total knee replacement within 9 years from magnetic resonance images. Inspired by MRNet, 2D slices of an MR image were encoded by a vision transformer and these encodings were aggregated to provide a single prediction outcome from a 3D MR volume. Our results suggest that the prediction performance of vision transformers was comparable with the models based on convolutional neural networks for the outcome prediction task. Moreover, training models with stochastic gradient descent optimizer provided a better performance compared with the Adam optimizer.
 

Anthony A Gatti¹, Kuan-Chieh Wang², Garry E. Gold¹, Scott L. Delp³, and Akshay C. Chaudhari^1
1Radiology, Stanford University, Stanford, CA, United States, ²Computer Science, Stanford University, Stanford, CA, United States, ³Bioengineering, Stanford University, Stanford, CA, United States

Three-dimensional statistical shape models built from MRI data can predict future disease and distinguish between groups. However, these models do not leverage the major advantage of MRI – the ability to visualize and quantify soft tissues such as cartilage. This study built three MRI-based statistical shape models, 1) bone shape, 2) cartilage thickness, 3) both bone shape and cartilage thickness. We showed that bone shape and bone shape + cartilage thickness models predicted sex with an R² of ~0.9, significantly outperforming cartilage thickness alone (0.8). However, cartilage thickness alone was significantly better at predicting radiographic knee osteoarthritis.

Aditya Ashok Bhosale¹, Leslie L Ying¹, and Xiaoliang Zhang^1
1Biomedical Engineering, University at Buffalo, Buffalo, NY, United States

This study proposes a 13-channel hybrid array system consisting of Microstrip transmission line resonators and half-birdcage coil and demonstrates the high-permittivity dielectric material’s positive effect on the B1 field distribution.

Yan Wen¹, Ek Tsoon Tan², Maggie Fung¹, and Darryl B. Sneag^2
1GE Healthcare, New York, NY, United States, ²Department of Radiology and Imaging, Hospital for Special Surgery, New York, NY, United States

The 3D FSE sequence is widely used in clinical MRI for its capability to collect many phase encoding lines in each RF excitation. However, the signals during the long echo train readout are modulated by the T2 decay, resulting in blurring. An algorithm is presented to retrospectively demodulate estimated T2 decay from the actual signal. In the preliminary data, the proposed algorithm demonstrated success in deblurring the images.

Saeed Jerban¹, Alecio Lombardi¹, Yajun Ma¹, Amir M Afsahi¹, Dina Moazamian¹, Jiyo Athertya¹, Hyungseok Jang¹, Christine B Chung¹, Jiang Du¹, and Eric Y Chang^1,2
1Department of Radiology, University of California, San Diego, San Diego, CA, United States, ²Radiology Service, VA San Diego Healthcare System, San Diego, CA, United States

Entheses are transition zones connecting flexible tissues such as tendons and ligaments to bone. Conventional MRI sequences detect little or no signal from entheses as a result of their short T2* values. Alternatively, ultrashort echo time MRI (UTE-MRI) with TE<50 μs can be used to image tissues with short T2 and T2* for quantitative assessment. The feasibility of using quantitative UTE-MRI techniques for the evaluation of entheses regions of Achilles tendons in cadaveric ankle specimens was investigated. Using 20 cadaveric ankle specimens, high quality images were obtained with significant entheses signal which enabled excellent UTE-T1, UTE-Adiab-T1r, and UTE-MT model fittings.