- COVID-19 diagnosis using AI
- Medical Imaging Techniques and Applications
- Radiomics and Machine Learning in Medical Imaging
- Image and Signal Denoising Methods
- Medical Image Segmentation Techniques
- Lung Cancer Diagnosis and Treatment
- Dental Radiography and Imaging
- Advanced X-ray and CT Imaging
- AI in cancer detection
- Advanced MRI Techniques and Applications
- Advanced Image Processing Techniques
- Optical measurement and interference techniques
- 3D Surveying and Cultural Heritage
- Advanced Neural Network Applications
- Image Processing Techniques and Applications
- Neural Networks and Applications
- Image Enhancement Techniques
- Image and Object Detection Techniques
- Breast Cancer Treatment Studies
- Medical Imaging and Analysis
- Advanced Image Fusion Techniques
- Remote Sensing and LiDAR Applications
- Stock Market Forecasting Methods
- Time Series Analysis and Forecasting
- Domain Adaptation and Few-Shot Learning
University of Shanghai for Science and Technology
2013-2025
Powerchina Huadong Engineering Corporation (China)
2024
PowerChina (China)
2024
National Clinical Research Center for Digestive Diseases
2024
Shanghai Ninth People's Hospital
2024
Shanghai Jiao Tong University
2024
University of Virginia
2024
Shandong Institute of Automation
2021
Chinese Academy of Sciences
2021
Fudan University Shanghai Cancer Center
2021
Stock prediction is a very hot topic in our life. However, the early time, because of some reasons and limitation device, only few people had access to study. Thanks rapid development science technology, recent years more are devoted study it becomes easier for us make stock by using different ways now, including machine learning, deep learning so on. In this paper, we proposed method based on Convolutional Neural Network predict price movement Chinese market. We set opening price, high low...
The lesion regions of a medical image account for only small part the image, and critical imbalance exists in distribution positive negative samples, which affects segmentation performance regions. Dice loss is beneficial involving an extreme samples but it ignores background regions, also contain large amount information. In this work, we propose improved dice that can mine information areas modify network architecture to improve performance. called weighted soft (WSDice loss). Our function...
Purpose: To develop a computer‐aided detection (CADe) scheme for nodules in chest radiographs (CXRs) with high sensitivity and low false‐positive (FP) rate. Methods: The authors developed CADe consisting of five major steps, which were improving the overall performance schemes. First, to segment lung fields accurately, multisegment active shape model. Then, two‐stage nodule‐enhancement technique was conspicuity nodules. Initial nodule candidates detected segmented by using clustering...
Major challenges in current computer-aided detection (CADe) schemes for nodule chest radiographs (CXRs) are to detect nodules that overlap with ribs and/or clavicles and reduce the frequent false positives (FPs) caused by ribs. Detection of such a CADe scheme is very important, because radiologists likely miss subtle nodules. Our purpose this study was develop improved sensitivity specificity use "virtual dual-energy" (VDE) CXRs where suppressed massive-training artificial neural networks...
Most lung nodules that are missed by radiologists as well computer-aided detection (CADe) schemes overlap with ribs or clavicles in chest radiographs (CXRs). The purpose of this study was to separate bony structures such and from soft tissue CXRs. To achieve this, we developed anatomically specific multiple massive-training artificial neural networks (MTANNs) combined total variation (TV) minimization smoothing a histogram-matching-based consistency improvement method. MTANNs were designed...
Over the past few years, researchers have demonstrated possibilities to use Computer-Aided Diagnosis (CAD) provide a preliminary diagnosis. Recently, it is also becoming increasingly common for doctors and computer practitioners collaborate on developing CAD. Since early diagnosis of breast cancer most critical step, precise segmentation tumor with accurate edge shape vital diagnoses reduction in patients’ pain. In view deficient accuracy existing method, we proposed novel method based U-Net...
Intraoral 3D measurement can obtain digital impressions in real time. However, owing to saliva, enamel, metallic denture, etc., the quality of captured 2D image using intraoral equipment, which is used for feature and reconstruction, usually degraded. In this paper, we propose a equipment with high performance measurement. First, group-multi-line coding algorithm instead sinusoidal or gray code fringe projection profilometry (FPP) was applied stripe pattern decoding. The distance between...
Abstract Background This paper presents a computer‐aided method for automatic detection of the positioning endotracheal, feeding and nasogastric tubes, identification tube types in radiography intensive care unit (ICU) patients. Application this may allow clinicians to detect tips more easily accurately, thus improve quality patient ICU. Methods One‐hundred‐and‐seven portable X‐ray images were collected from 20 patients, using Kodak computed system. It was determined whether each image did...
The accurate prediction of the flow field characteristics complex mountains is great practical significance for development and construction wind farms, but it not yet fully understood. main purpose this study to propose a method under mountain conditions, which can optimize boundary conditions required numerical simulation through acceleration ratio and, at same time, couple measurement data reflect real distribution. results show that proposed has good applicability in reproduce...
To assist physicists in developing radiation therapy treatment plans and evaluating the effects of radiotherapy, an accurate automatic tumor segmentation approach positron emission tomography (PET) images is highly demanded clinical practice. In present paper we investigate construct a neural network architecture for auto-segmenting tumors by leveraging 14-layer U-Net model with two blocks VGG19 encoder pre-trained ImageNet. For pursuing efficient learning, series training strategies are...
The enhancement of lung nodules in chest radiographs plays an important role computer-aided diagnosis, and is more useful for doctor observing analyzing. In this paper, we introduce a parameterized logarithmic image processing (PLIP) method based on Laplacian Gaussian (LoG) filtering to enhance radiographs. This combines the advantages both algorithms which can (CXRs) with better contrast edge information. By means measure by entropy evaluation (EMEE) objectively, experimental results show...
In intensive care units (ICUs), supporting devices play an important role, and the placement of these must be accurate, such as catheters tubes. Taking portable chest radiograph (CXRs) for patients in ICU is a standard procedure. However, non-optimized exposure settings misaligned body positions usually mean that CXRs are not acceptable working condition. The purpose this study was to enhance assist radiologists positioning endotracheal, feeding, nasogastric tubes patients. unsharp masking...
Direct digital impressions can be obtained using intraoral three-dimensional (3D) scanners. Among them, the fringe projection profilometry (FPP)-based scanner serves as a micro-3D imaging system. Here, proper calibration is critical for highly accurate impressions. However, due to its miniaturized structure, some of designed geometric parameters are difficult attained in factory assembly model-based reconstruction. Moreover, angle structured light typically small avoid shadows dental sulcus....
Its crux lies in the optimization of a tradeoff between accuracy and fairness resultant models on selected feature subset. The technical challenge our setting is twofold: 1) streaming inputs, such that an informative may become obsolete or redundant for prediction if its information has been covered by other similar features arrived prior to it, 2) non-associational correlation, bias be leaked from those seemingly admissible, non-protected features. To overcome this, we propose Streaming...