A5013691634- Medical Imaging Techniques and Applications
- Radiomics and Machine Learning in Medical Imaging
- Advanced MRI Techniques and Applications
- Medical Image Segmentation Techniques
- Radiation Detection and Scintillator Technologies
- Advanced X-ray and CT Imaging
- Image and Signal Denoising Methods
- Advanced Radiotherapy Techniques
- Advanced Image Processing Techniques
- Sparse and Compressive Sensing Techniques
- AI in cancer detection
- Advanced Neural Network Applications
- Radiopharmaceutical Chemistry and Applications
- Cell Image Analysis Techniques
- Photoacoustic and Ultrasonic Imaging
- Head and Neck Cancer Studies
- MRI in cancer diagnosis
- Spinal Cord Injury Research
- Atomic and Subatomic Physics Research
- Cardiac Structural Anomalies and Repair
- Nuclear Physics and Applications
- Multimodal Machine Learning Applications
- COVID-19 diagnosis using AI
- Radiation Dose and Imaging
- Generative Adversarial Networks and Image Synthesis
Massachusetts General Hospital
2017-2025
Harvard University
2017-2025
University of Florida
2023-2025
Gordon Center for Medical Imaging
2017-2024
IPS Research (United States)
2023
Philips (United States)
2023
Johns Hopkins University
2023
Boston University
2022
Zhejiang University
2021
State Key Laboratory of Modern Optical Instruments
2021
PET image reconstruction is challenging due to the ill-poseness of inverse problem and limited number detected photons. Recently, deep neural networks have been widely successfully used in computer vision tasks attracted growing interests medical imaging. In this paper, we trained a residual convolutional network improve quality by using existing inter-patient information. An innovative feature proposed method that embed iterative framework for representation, rather than it as...
Recently, deep neural networks have been widely and successfully applied in computer vision tasks attracted growing interest medical imaging. One barrier for the application of to imaging is need large amounts prior training pairs, which not always feasible clinical practice. This especially true image reconstruction problems, where raw data are needed. Inspired by framework, this paper, we proposed a personalized network method no pairs needed, but only patient's own information. The...
Positron emission tomography (PET) is a functional imaging modality widely used in clinical diagnosis. In this paper, we trained deep convolutional neural network to improve PET image quality. Perceptual loss based on features derived from pretrained VGG network, instead of the conventional mean squared error, was employed as training function preserve details. As number real patient data set for limited, propose pretrain using simulation and fine-tune last few layers sets. Results...
Motivated by the great potential of deep learning in medical imaging, we propose an iterative positron emission tomography reconstruction framework using a learning-based prior. We utilized denoising convolutional neural network (DnCNN) method and trained full-dose images as ground truth low dose reconstructed from downsampled data Poisson thinning input. Since most published networks are at predetermined noise level, level disparity training testing is major problem for their applicability...
In radiation therapy, the accurate delineation of gross tumor volume (GTV) is crucial for treatment planning. However, it challenging head and neck cancer (HNC) due to morphology complexity various organs in head, low targets background contrast potential artifacts on conventional planning CT images. Thus, manual GTV anatomical images extremely time consuming suffers from inter-observer variability that leads uncertainty. With wide use PET/CT imaging oncology, complementary functional...
Machine learning has found unique applications in nuclear medicine from photon detection to quantitative image reconstruction. While there have been impressive strides detector development for time-of-flight positron emission tomography, most detectors still make use of simple signal processing methods extract the time and position information signals. Now with availability fast waveform digitizers, machine techniques applied estimate arrival high-energy photons. In reconstruction, used...
Positron emission tomography (PET) is a functional imaging modality widely used in neuroscience studies. To obtain meaningful quantitative results from PET images, attenuation correction necessary during image reconstruction. For PET/MR hybrid systems, challenging as magnetic resonance (MR) images do not reflect coefficients directly. address this issue, we present deep neural network methods to derive the continuous for brain MR images. With only Dixon input, existing U-net structure was...
Position emission tomography (PET) is widely used in clinics and research due to its quantitative merits high sensitivity, but suffers from low signal-to-noise ratio (SNR). Recently convolutional neural networks (CNNs) have been improve PET image quality. Though successful efficient local feature extraction, CNN cannot capture long-range dependencies well limited receptive field. Global multi-head self-attention (MSA) a popular approach information. However, the calculation of global MSA for...
Positron emission tomography (PET) is a functional imaging modality widely used in oncology, cardiology, and neuroscience. It highly sensitive, but suffers from relatively poor spatial resolution, as compared with anatomical modalities, such magnetic resonance (MRI). With the recent development of combined PET/MR systems, we can improve PET image quality by incorporating MR information into reconstruction. Previously, kernel learning has been successfully embedded static dynamic...
Spatial resolution is an important metric for performance characterization in PET systems. Measuring spatial straightforward with a linear reconstruction algorithm, such as filtered backprojection, and can be performed by reconstructing point source scan calculating the full-width-at-half-maximum (FWHM) along principal directions. With widespread adoption of iterative methods, it desirable to quantify using algorithm. However, task difficult because algorithms are nonlinear non-negativity...
The desire to understand normal and disordered human brain function of upright, moving persons in natural environments motivates the development ambulatory micro-dose PET imager (AMPET). An ideal system would be light weight but with high sensitivity spatial resolution, although these requirements are often conflict each other. One potential approach meet design goals is a compact brain-only imaging device head-sized aperture. However, geometry increases parallax error peripheral lines...
Positron emission tomography (PET) is an ill-posed inverse problem and suffers high noise due to limited number of detected events. Prior information can be used improve the quality reconstructed PET images. Deep neural networks have also been applied regularized image reconstruction. One method use a pretrained denoising network represent perform constrained maximum likelihood estimation. In this work, we propose generative adversarial (GAN) further performance. We modify objective function...
As of August 30th, there were in total 25.1 million confirmed cases and 845 thousand deaths caused by coronavirus disease 2019 (COVID-19) worldwide. With overwhelming demands on medical resources, patient stratification based their risks is essential. In this multi-center study, we built prognosis models to predict severity outcomes, combining patients' electronic health records (EHR), which included vital signs laboratory data, with deep learning- CT-based prediction.
Direct reconstruction methods have been developed to estimate parametric images directly from the measured PET sinograms by combining imaging model and tracer kinetics in an integrated framework. Due limited counts received, signal-to-noise-ratio (SNR) resolution of produced direct frameworks are still limited. Recently supervised deep learning successfully applied medical denoising/reconstruction when large number high-quality training labels available. For static imaging, can be acquired...
Spinal cord injuries (SCIs) often lead to lifelong disability. Among the various types of injuries, incomplete and discomplete where some axons remain intact, offer potential for recovery. However, demyelination these spared can worsen Demyelination is a reversible phenomenon, drugs such as 4-aminopyridine (4AP), which target K<sup>+</sup> channels in demyelinated axons, show that conduction be restored. Yet, accurately assessing monitoring after SCI remains challenging because lack suitable...
Vision-language models (VLMs) align visual and textual representations, enabling high-performance zero-shot classification image-text retrieval in 2D medical imaging. However, extending VLMs to 3D imaging remains computationally challenging. Existing rely on Vision Transformers (ViTs), which are expensive due self-attention's quadratic complexity, or convolutions, demand excessive parameters FLOPs as kernel size increases. We introduce DCFormer, an efficient image encoder that factorizes...
Vision-language models (VLMs) have shown promise in 2D medical image analysis, but extending them to 3D remains challenging due the high computational demands of volumetric data and difficulty aligning spatial features with clinical text. We present Med3DVLM, a VLM designed address these challenges through three key innovations: (1) DCFormer, an efficient encoder that uses decomposed convolutions capture fine-grained at scale; (2) SigLIP, contrastive learning strategy pairwise sigmoid loss...