A5029722566- Advanced Neural Network Applications
- Medical Image Segmentation Techniques
- Radiomics and Machine Learning in Medical Imaging
- AI in cancer detection
- Medical Imaging and Analysis
- Domain Adaptation and Few-Shot Learning
- COVID-19 diagnosis using AI
- Fetal and Pediatric Neurological Disorders
- Brain Tumor Detection and Classification
- Head and Neck Cancer Studies
- Advanced Radiotherapy Techniques
- Lung Cancer Diagnosis and Treatment
- Digital Imaging for Blood Diseases
- Artificial Intelligence in Healthcare and Education
- Neonatal and fetal brain pathology
- Medical Imaging Techniques and Applications
- Meningioma and schwannoma management
- Industrial Vision Systems and Defect Detection
- Innovative Energy Harvesting Technologies
- Cardiovascular Health and Disease Prevention
- Retinal Imaging and Analysis
- Pregnancy and preeclampsia studies
- Advanced Sensor and Energy Harvesting Materials
- Image Retrieval and Classification Techniques
- Maternal and fetal healthcare
University of Electronic Science and Technology of China
2019-2025
Shanghai Artificial Intelligence Laboratory
2022-2025
Sichuan Cancer Hospital
2024
First Affiliated Hospital of Xi'an Jiaotong University
2022-2024
Affiliated Hospital of Shaanxi University of Chinese Medicine
2017-2024
Shaanxi University of Chinese Medicine
2024
Beijing Academy of Artificial Intelligence
2022-2023
Institute of Microelectronics
2022-2023
Chinese Academy of Sciences
2022-2023
University of Chinese Academy of Sciences
2022-2023
Convolutional neural networks (CNNs) have achieved state-of-the-art performance for automatic medical image segmentation. However, they not demonstrated sufficiently accurate and robust results clinical use. In addition, are limited by the lack of image-specific adaptation generalizability to previously unseen object classes (a.k.a. zero-shot learning). To address these problems, we propose a novel deep learning-based interactive segmentation framework incorporating CNNs into bounding box...
Medical image analysis and computer-assisted intervention problems are increasingly being addressed with deep-learning-based solutions. Established deep-learning platforms flexible but do not provide specific functionality for medical adapting them this domain of application requires substantial implementation effort. Consequently, there has been duplication effort incompatible infrastructure developed across many research groups. This work presents the open-source NiftyNet platform deep...
Despite the state-of-the-art performance for medical image segmentation, deep convolutional neural networks (CNNs) have rarely provided uncertainty estimations regarding their segmentation outputs, e.g., model (epistemic) and image-based (aleatoric) uncertainties. In this work, we analyze these different types of uncertainties CNN-based 2D 3D tasks. We additionally propose a test-time augmentation-based aleatoric to effect transformations input on output. Test-time augmentation has been...
Accurate medical image segmentation is essential for diagnosis and treatment planning of diseases. Convolutional Neural Networks (CNNs) have achieved state-of-the-art performance automatic segmentation. However, they are still challenged by complicated conditions where the target has large variations position, shape scale, existing CNNs a poor explainability that limits their application to clinical decisions. In this work, we make extensive use multiple attentions in CNN architecture...
Segmentation of pneumonia lesions from CT scans COVID-19 patients is important for accurate diagnosis and follow-up. Deep learning has a potential to automate this task but requires large set high-quality annotations that are difficult collect. Learning noisy training labels easier obtain alleviate problem. To end, we propose novel noise-robust framework learn the segmentation task. We first introduce Dice loss generalization Mean Absolute Error (MAE) robustness against noise, then Pneumonia...
Accurate medical image segmentation is essential for diagnosis, surgical planning and many other applications. Convolutional Neural Networks (CNNs) have become the state-of-the-art automatic methods. However, fully results may still need to be refined accurate robust enough clinical use. We propose a deep learning-based interactive method improve obtained by an CNN reduce user interactions during refinement higher accuracy. use one obtain initial segmentation, on which are added indicate...
One of the fundamental challenges in supervised learning for multimodal image registration is lack ground-truth voxel-level spatial correspondence. This work describes a method to infer transformation from higher-level correspondence information contained anatomical labels. We argue that such labels are more reliable and practical obtain reference sets pairs than Typical interest may include solid organs, vessels, ducts, structure boundaries other subject-specific ad hoc landmarks. The...
Artificial Intelligence (AI) is having a tremendous impact across most areas of science. Applications AI in healthcare have the potential to improve our ability detect, diagnose, prognose, and intervene on human disease. For models be used clinically, they need made safe, reproducible robust, underlying software framework must aware particularities (e.g. geometry, physiology, physics) medical data being processed. This work introduces MONAI, freely available, community-supported,...
High-resolution volume reconstruction from multiple motion-corrupted stacks of 2D slices plays an increasing role for fetal brain Magnetic Resonance Imaging (MRI) studies. Currently existing methods are time-consuming and often require user interactions to localize extract the several slices. We propose a fully automatic framework that consists four stages: 1) localization based on coarse segmentation by Convolutional Neural Network (CNN), 2) fine another CNN trained with multi-scale loss...
Automatic segmentation of brain tumors from medical images is important for clinical assessment and treatment planning tumors. Recent years have seen an increasing use convolutional neural networks (CNNs) this task, but most them either 2D with relatively low memory requirement while ignoring 3D context, or exploiting features large consumption. In addition, existing methods rarely provide uncertainty information associated the result. We propose a cascade CNNs to segment hierarchical...
Automatic segmentation of vestibular schwannomas (VSs) from MRI could significantly improve clinical workflow and assist in patient management. Accurate tumor volumetric measurements provide the best indicators to detect subtle VS growth, but current techniques are labor intensive dedicated software is not readily available within setting. The authors aim develop a novel artificial intelligence (AI) framework be embedded routine for automatic delineation volumetry VS.
Recently, deep learning with Convolutional Neural Networks (CNNs) and Transformers has shown encouraging results in fully supervised medical image segmentation. However, it is still challenging for them to achieve good performance limited annotations training. In this work, we present a very simple yet efficient framework semi-supervised segmentation by introducing the cross teaching between CNN Transformer. Specifically, simplify classical co-training from consistency regularization...
Automatic segmentation of vestibular schwannomas (VS) from magnetic resonance imaging (MRI) could significantly improve clinical workflow and assist patient management. We have previously developed a novel artificial intelligence framework based on 2.5D convolutional neural network achieving excellent results equivalent to those achieved by an independent human annotator. Here, we provide the first publicly-available annotated dataset VS releasing data annotations used in our prior work....
Domain Adaptation (DA) has recently raised strong interests in the medical imaging community. While a large variety of DA techniques been proposed for image segmentation, most these have validated either on private datasets or small publicly available datasets. Moreover, mostly addressed single-class problems. To tackle limitations, Cross-Modality (crossMoDA) challenge was organised conjunction with 24th International Conference Medical Image Computing and Computer Assisted Intervention...
Computed Tomography (CT) plays an important role in monitoring radiation-induced Pulmonary Fibrosis (PF), where accurate segmentation of the PF lesions is highly desired for diagnosis and treatment follow-up. However, task challenged by ambiguous boundary, irregular shape, various position size lesions, as well difficulty acquiring a large set annotated volumetric images training. To overcome these problems, we propose novel convolutional neural network called PF-Net incorporate it into...