Jeewon Kim¹, Kinam Kwon¹, Seohee So², Byungjai Kim¹, Wonil Lee¹, and HyunWook Park^1
1Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea, Republic of, ²Korea Institute of Science and Technology, Seoul, Korea, Republic of

We propose a new correction scheme using a deep neural network with unsupervised learning to correct Nyquist ghosts and geometry distortions occurring in EPI. The proposed scheme includes NGAC-net and GDC-net. First, the NGAC-net estimates the phase error of k-space with the help of a ghost formulation operator and correlation loss. The NGAC-net produces two Nyquist ghost corrected images obtained by dual-polarity phase-encoding gradients. The GDC-net is trained to estimate the frequency-shift map using the two output images from the NGAC-net. Afterwards, an MR image generation operator utilizes the estimated frequency-shift map to obtain the geometry distortion corrected images.

Srivathsa Pasumarthi Venkata¹, Ben Andrew Duffy¹, and Keshav Datta^1
1R&D, Subtle Medical Inc, Menlo Park, CA, United States

Image registration is a crucial preprocessing step for many downstream analysis tasks. Existing iterative methods for affine registration are accurate but time consuming. We propose a deep learning (DL) based unsupervised affine registration algorithm that executes orders of magnitude faster when compared to conventional registration toolkits. The proposed algorithm aligns 3D volumes from the same modality (e.g. T1 vs T1-CE) as well as different modalities (e.g. T1 vs T2). We train the model and perform quantitative evaluation using a pre-registered brain MRI public dataset.

Abrar Faiyaz¹, Nhat Hoang¹, Alan Finkelstein¹, Jianhui Zhong¹, Marvin Doyley¹, Henry Wang², Md Nasir Uddin^1,2, and Giovanni Schifitto^1,2
1University of Rochester, Rochester, NY, United States, ²University of Rochester Medical Center, Rochester, NY, United States

Brain arterial-blood-vessels can carry important information regarding cerebrovascular pathogenesis. Due to non-invasiveness, 3D-time-of-flight MR-angiography is widely used to depict arteries in clinical-exams. However, deemed limited for qualitative assessment in clinical setting. Although several post-processing approaches exist such as tool-based-manual segmentation and diameter-marking or deep-learning-based segmentations--these require huge-time and experienced-eyes or large manually-segmented training-sets, making performance variable. Since brain-artery-annotations are anatomically-inspired, anatomical-confidant-points identified with a Bayesian approach can be used to annotate major-brain-arteries, which circumvents the need for large-dataset from learning requirement. The approach will allow clinical evaluations of major-arteries and  biomarker identifications for cerebrovascular pathogenesis with limited resource and time. 

Haejoon Lee¹, Mohammed A. Al-masni¹, Jun-Ho Kim¹, Seul Lee¹, Kyu-Jin Jung¹, Woo-Ram Kim², Young Noh^2,3, and Dong-Hyun Kim^1
1Department of Electrical & Electronic Engineering, Yonsei University, Seoul, Korea, Republic of, ²Neuroscience Research Institute, Gachon University, Incheon, Korea, Republic of, ³Department of Neurology, Gachon University College of Medicine, Gil Medical Center, Incheon, Korea, Republic of

 Automated detection of Cerebral Microbleeds (CMBs) in Magnetic Resonance (MR) images can be clinically useful because of the small size of CMBs and the presence of numerous mimics. In this study, we propose an end-to-end Triplanar Ensemble Detection Network (TPE-Det) that demonstrates notable performance with only a single stage. Via the proposed TPE-Det, we combined CMBs detection results from axial, sagittal, and coronal planes. When considering the additional planes, the average False Positives per patient (FP_avg) decreased from 29.16 to 0.92 compared to the case of using only the axial plane, preserving the high sensitivity of 95.94%.

Soumick Chatterjee^1,2,3, Hadya Yassin⁴, Florian Dubost⁵, Andreas Nürnberger^2,3,6, and Oliver Speck^1,6,7,8
1Department of Biomedical Magnetic Resonance, Otto von Guericke University Magdeburg, Magdeburg, Germany, ²Data and Knowledge Engineering Group, Otto von Guericke University Magdeburg, Magdeburg, Germany, ³Faculty of Computer Science, Otto von Guericke University Magdeburg, Magdeburg, Germany, ⁴Institute for Medical Engineerin, Otto von Guericke University Magdeburg, Magdeburg, Germany, ⁵Department of Biomedical Data Science, Stanford University, Stanford, CA, United States, ⁶Center for Behavioral Brain Sciences, Magdeburg, Germany, ⁷German Centre for NeurodegenerativeDiseases, Magdeburg, Germany, ⁸Leibniz Institute for Neurobiology, Magdeburg, Germany

Deep learning pipelines typically require manually annotated training data and the complex reasoning done by such methods make them appear as “black-boxes” to the end-users, leading to reduced trust. Unsupervised or weakly-supervised techniques could be a possible candidate for solving the first issue, while explainable classifiers or applying post-hoc interpretability-explainability methods on opaque classifiers may solve the second issue. This research addresses both problems by segmenting brain tumours without segmentation labels for training, using an explainable deep learning-based classifier. The classifier combined with a global pooling operation with a segmentation model to train and obtain classification results from this method.

Soumick Chatterjee^1,2,3, Alessandro Sciarra^1,4, Max Dünnwald^3,4, Pavan Tummala³, Shubham Kumar Agrawal³, Aishwarya Jauhari³, Aman Kalra³, Steffen Oeltze-Jafra^4,5, Oliver Speck^1,5,6,7, and Andreas Nürnberger^2,3,5
1Department of Biomedical Magnetic Resonance, Otto von Guericke University Magdeburg, Magdeburg, Germany, ²Data and Knowledge Engineering Group, Otto von Guericke University Magdeburg, Magdeburg, Germany, ³Faculty of Computer Science, Otto von Guericke University Magdeburg, Magdeburg, Germany, ⁴MedDigit, Department of Neurology, Medical Faculty, University Hospital, Magdeburg, Germany, ⁵Center for Behavioral Brain Sciences, Magdeburg, Germany, ⁶German Centre for NeurodegenerativeDiseases, Magdeburg, Germany, ⁷Leibniz Institute for Neurobiology, Magdeburg, Germany

Deep learning methods are typically trained in a supervised with annotated data for analysing medical images with the motivation of detecting pathologies. In the absence of manually annotated training data, unsupervised anomaly detection can be one of the possible solutions. This work proposes StRegA, an unsupervised anomaly detection pipeline based on a compact ceVAE and shows its applicability in detecting anomalies such as tumours in brain MRIs. The proposed pipeline achieved a Dice score of 0.642±0.101 while detecting tumours in T2w images of the BraTS dataset and 0.859±0.112 while detecting artificially induced anomalies.

Soumick Chatterjee^1,2,3, Himanshi Bajaj³, Mohammad Istiyak Hossain Siddiquee³, Nandish Bandi Subbarayappa³, Steve Simon³, Suraj Bangalore Shashidhar³, Oliver Speck^1,4,5,6, and Andreas Nürnberger^2,3,5
1Department of Biomedical Magnetic Resonance, Otto von Guericke University Magdeburg, Magdeburg, Germany, ²Data and Knowledge Engineering Group, Otto von Guericke University Magdeburg, Magdeburg, Germany, ³Faculty of Computer Science, Otto von Guericke University Magdeburg, Magdeburg, Germany, ⁴German Centre for NeurodegenerativeDiseases, Magdeburg, Germany, ⁵Center for Behavioral Brain Sciences, Magdeburg, Germany, ⁶Leibniz Institute for Neurobiology, Magdeburg, Germany

Deep Learning based deformable registration techniques such as Voxelmorph, ICNet, FIRE, do not explicitly encode global dependencies and track large deformations. This research attempts to encode semantics, i.e. structure and overall view of the anatomy in the supplied image, by incorporating self-constructing graph network in the latent space of a UNet model. It also attempts to track larger deformations through multiscale architecture and maintains consistent deformations through cycle consistency. The proposed method was compared against Voxelmorph and ANTs for T1 intramodal and T1-T2 Intermodal registration on IXI Dataset. The experiments show that the proposed model outperforms the baselines.

Parna Eshraghi Boroojeni¹, Yasheng Chen¹, Cihat Eldeniz¹, Paul Commean¹, Gary Skolnick¹, Kamlesh Patel¹, and Hongyu An^1
1Washington University in Saint Louis, Saint Louis, MO, United States

Head trauma is common in the pediatric population. Craniosynostosis is abnormal early fusion of a cranial suture, causing an irregular-shaped cranium.3D high-resolution head CT scans are commonly used in these pediatric patients to identify skull fractures and sutures. However, CT exposespediatric patients to ionizing radiationand increases risk of cancer. We developed a robust and automated deep learning method to convert MR images to pseudo-CT (pCT) that can facilitate translating MR cranial bone imaging into clinical practice.  An average Dice Coefficient of 0.89 and mean absolute error of 72.45 HU between pCT and CT were achieved.

Soohyun Jeon¹, Kanghyun Ryu², and Dong-Hyun Kim^1
1Electrical and Electronic Engineering, Yonsei University, Seoul, Korea, Republic of, ²Radiology, Stanford University, Stanford, CA, United States

MR phase information has been used in many MRI applications, recently, so it is very important to estimate the correct MR phase. Since the MR phase is encoded in complex exponential, the actual phase information obtained is in the form of a wrapped signal. Unfortunately, conventional MR phase unwrap methods are highly related to SNR and require a lot of computation, so an efficient method is needed.. Here, we propose ViT-PU-Net to perform 3D phase unwrap using ViT-based encoder architecture. Therefore, we investigated whether this architecture could be applied to 3D phase unwrap task by applying it to various data.

Alexandra Grace Roberts¹, Pascal Spincemaille², Ilhami Kovanlikaya², Apostolos John Tsiouris², Thanh Nguyen², and Yi Wang^2,3
1Electrical and Computer Engineering, Cornell University, Ithaca, NY, United States, ²Radiology, Weill Cornell Medicine, New York, NY, United States, ³Biomedical Engineering, Cornell University, Ithaca, NY, United States

Susceptibility-Weighted Imaging Super-Resolution (SWISeR) increases the apparent resolution of input images and improves image quality, measured by mean-squared error and clinical scoring. This method generalizes to healthy subjects scanned at 3T and 7T and subjects imaged at 1.5T and 3T.

Victoria Y Yu¹, Kathryn R Tringale², Ricardo Otazo¹, and Ouri Cohen^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

In MR fingerprinting, quantitative maps are obtained by matching the measured signal to a pre-computed dictionary. However, a key constraint of dictionary matching is the exponential growth of the dictionary with the number of parameters. A deep learning method named DRONE overcomes this constraint by using deep learning to map the magnitude-valued signal to the underlying tissue parameters. Here we describe an extension of DRONE that jointly estimates a phase term to enable mapping complex-valued signals and improve the quantitative accuracy. We test the accuracy in the ISMRM NIST phantom and demonstrate the clinical utility in patients with brain metastases.