A5043312634- Radiomics and Machine Learning in Medical Imaging
- Advanced Neural Network Applications
- Seismic Imaging and Inversion Techniques
- Brain Tumor Detection and Classification
- Neural Networks and Applications
- Medical Imaging Techniques and Applications
- Geological Modeling and Analysis
- AI in cancer detection
- Model Reduction and Neural Networks
- COVID-19 diagnosis using AI
- Advanced X-ray and CT Imaging
- Reservoir Engineering and Simulation Methods
- Medical Imaging and Analysis
- Medical Image Segmentation Techniques
- Hepatocellular Carcinoma Treatment and Prognosis
- Advanced MRI Techniques and Applications
- Liver Disease Diagnosis and Treatment
- Hydrocarbon exploration and reservoir analysis
- Methane Hydrates and Related Phenomena
- Generative Adversarial Networks and Image Synthesis
- CO2 Sequestration and Geologic Interactions
- MRI in cancer diagnosis
- Glioma Diagnosis and Treatment
- Graph theory and applications
- Cholangiocarcinoma and Gallbladder Cancer Studies
Rice University
2021-2024
The University of Texas MD Anderson Cancer Center
2021-2024
Total (Belgium)
2024
Total (United States)
2023-2024
University of Washington
2021
University of Houston
2021
Sandia National Laboratories
2021
Medical imaging deep learning models are often large and complex, requiring specialized hardware to train evaluate these models. To address such issues, we propose the PocketNet paradigm reduce size of by throttling growth number channels in convolutional neural networks. We demonstrate that, for a range segmentation classification tasks, architectures produce results comparable that conventional networks while reducing parameters multiple orders magnitude, using up 90% less GPU memory,...
Physics-informed neural networks (PINNs) have gained significant attention for solving forward and inverse problems related to partial differential equations (PDEs). While advancements in loss functions network architectures improved PINN accuracy, the impact of collocation point sampling on their performance remains underexplored. Fixed methods, such as uniform random equispaced grids, can fail capture critical regions with high solution gradients, limiting effectiveness complex PDEs....
In recent years, there has been an increasing interest in using deep learning and neural networks to tackle scientific problems, particularly solving partial differential equations (PDEs). However, many network-based methods, such as physics-informed networks, depend on automatic differentiation the sampling of collocation points, which can result a lack interpretability lower accuracy compared traditional numerical methods. To address this issue, we propose two approaches for discontinuous...
We introduce two algorithms that invert simulated gravity data to 3D subsurface rock/flow properties. The first algorithm is a data-driven, deep learning-based approach, and the second also data-driven but considers temporal evolution of surface events. target application these proposed prediction CO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> plumes as complementary tool for monitoring sequestration deployments. Each outperforms...
Abstract Image segmentation of the liver is an important step in treatment planning for cancer. However, manual at a large scale not practical, leading to increasing reliance on deep learning models automatically segment liver. This manuscript develops generalizable model T1-weighted MR images. In particular, three distinct architectures (nnUNet, PocketNet, Swin UNETR) were considered using data gathered from six geographically different institutions. A total 819 images both public and...
Within medical imaging segmentation, the Dice coefficient and Hausdorff-based metrics are standard measures of success for deep learning models. However, modern loss functions image segmentation often only consider or similar region-based during training. As a result, architectures trained over such run risk achieving high accuracy but low metrics. Low on can be problematic applications as tumor where benchmarks crucial. For example, scores accompanied by significant Hausdorff errors could...
Accurate medical imaging segmentation is critical for precise and effective interventions. However, despite the success of convolutional neural networks (CNNs) in image segmentation, they still face challenges handling fine-scale features variations scales. These are particularly evident complex challenging tasks, such as BraTS multilabel brain tumor challenge. In this task, accurately segmenting various sub-components, which vary significantly size shape, remains a significant challenge,...
Complex data processing and curation for artificial intelligence applications rely on high-quality sets training analysis. Manually reviewing images their associated annotations is a very laborious task existing quality control tools review are generally limited to raw only. The purpose of this work was develop an imaging informatics dashboard the easy fast processed magnetic resonance (MR) sets; we demonstrated its ability in large-scale review.We developed custom R Shiny that displays key...
Abstract Background Magnetic resonance imaging (MRI) scans are known to suffer from a variety of acquisition artifacts as well equipment‐based variations that impact image appearance and segmentation performance. It is still unclear whether direct relationship exists between magnetic (MR) quality metrics (IQMs) (e.g., signal‐to‐noise, contrast‐to‐noise) accuracy. Purpose Deep learning (DL) approaches have shown significant promise for automated brain tumors on MRI but depend the input...
Image segmentation of the liver is an important step in several treatments for cancer. However, manual at a large scale not practical, leading to increasing reliance on deep learning models automatically segment liver. This manuscript develops model T1w MR images. We sought determine best architecture by training, validating, and testing three different architectures using total 819 images gathered from six datasets, both publicly internally available. Our experiments compared each...
Precise automated delineation of post-operative gross tumor volume in glioblastoma cases is challenging and time-consuming owing to the presence edema deformed brain tissue resulting from surgical resection. To develop a model for volumes glioblastoma, we proposed novel 3D double pocket U-Net architecture that has two parallel U-Nets. Both U-Nets were trained simultaneously with different subsets MRI sequences output models was combined do final prediction. We strategically input (T1, T2,...
Medical imaging segmentation is a highly active area of research, with deep learning-based methods achieving state-of-the-art results in several benchmarks. However, the lack standardized tools for training, testing, and evaluating new makes comparison difficult. To address this, we introduce Imaging Segmentation Toolkit (MIST), simple, modular, end-to-end medical framework designed to facilitate consistent evaluation methods. MIST standardizes data analysis, preprocessing, pipelines,...
We introduce a fully 3D, deep learning-based approach for the joint inversion of time-lapse surface gravity and seismic data reconstructing subsurface density velocity models. The target application this proposed is prediction CO2 plumes as complementary tool monitoring sequestration deployments. Our technique outperforms gravity-only seismic-only models, achieving improved reconstruction, accurate segmentation, higher R-squared coefficients. These results indicate that an effective CO$_2$...
Medical imaging deep learning models are often large and complex, requiring specialized hardware to train evaluate these models. To address such issues, we propose the PocketNet paradigm reduce size of by throttling growth number channels in convolutional neural networks. We demonstrate that, for a range segmentation classification tasks, architectures produce results comparable that conventional networks while reducing parameters multiple orders magnitude, using up 90% less GPU memory,...
Recent advances in machine learning have enabled image-based prediction of local tissue pathology gliomas, but the clinical usefulness these predictions is unknown. We aimed to evaluate prognostic ability imaging-based estimates cellular density for patients with comparison gold standard reference World Health Organization grading.Data from 1181 (207 grade II, 246 III, 728 IV) previously untreated gliomas a single institution were analyzed. A pretrained random forest model estimated...
Deep neural networks with multilevel connections process input data in complex ways to learn the information.A learning efficiency depends not only on network architecture but also training images.Medical image segmentation deep for skull stripping or tumor from magnetic resonance images enables both global and local features of images.Though medical are collected a controlled environment,there may be artifacts equipment based variance that cause inherent bias set.In this study, we...
Accurate medical imaging segmentation is critical for precise and effective interventions. However, despite the success of convolutional neural networks (CNNs) in image segmentation, they still face challenges handling fine-scale features variations scales. These are particularly evident complex challenging tasks, such as BraTS multi-label brain tumor challenge. In this task, accurately segmenting various sub-components, which vary significantly size shape, remains a significant challenge,...
In recent years, there has been a growing interest in leveraging deep learning and neural networks to address scientific problems, particularly solving partial differential equations (PDEs). However, current network-based PDE solvers often rely on extensive training data or labeled input-output pairs, making them prone challenges generalizing out-of-distribution examples. To mitigate the generalization gap encountered by conventional methods estimating solutions, we formulate fully...
In recent years, there has been a growing interest in leveraging deep learning and neural networks to address scientific problems, particularly solving partial differential equations (PDEs). However, current network-based PDE solvers often rely on extensive training data or labeled input-output pairs, making them prone challenges generalizing out-of-distribution examples. To mitigate the generalization gap encountered by conventional methods estimating solutions, we formulate fully...
We introduce three algorithms that invert simulated gravity data to 3D subsurface rock/flow properties. The first algorithm is a data-driven, deep learning-based approach, the second mixes learning approach with physical modeling into single workflow, and third considers time dependence of surface monitoring. target application these proposed prediction CO$_2$ plumes as complementary tool for monitoring sequestration deployments. Each outperforms traditional inversion methods produces...