A5103062942- Digital Imaging for Blood Diseases
- Image Retrieval and Classification Techniques
- Computational Drug Discovery Methods
- Domain Adaptation and Few-Shot Learning
- Machine Learning in Bioinformatics
- Machine Learning and Algorithms
- Protein Structure and Dynamics
- COVID-19 diagnosis using AI
- AI in cancer detection
- Rough Sets and Fuzzy Logic
- Chemical Synthesis and Analysis
- Video Analysis and Summarization
- Genomics and Phylogenetic Studies
- Data Mining Algorithms and Applications
- Advanced Database Systems and Queries
- Machine Learning in Materials Science
- Advanced Image and Video Retrieval Techniques
- Anomaly Detection Techniques and Applications
Shanghai Medical College of Fudan University
2023-2025
Fudan University
2023-2024
Shanghai Institute of Computing Technology
2024
Fujian Medical University
2023
Laboratoire d’Imagerie Biomédicale
2023
Harbin Institute of Technology
2020
Weakly supervised whole slide image classification is usually formulated as a multiple instance learning (MIL) problem, where each treated bag, and the patches cut out of it are instances. Existing methods either train an classifier through pseudo-labeling or aggregate features into bag feature attention mechanisms then classifier, scores can be used for instance-level classification. However, pseudo labels constructed by former contain lot noise, latter not accurate enough, both which...
Bag-based multiple instance learning (MIL) methods have become the mainstream for Whole Slide Image (WSI) classification. However, there are still three important issues that not been fully addressed: (1) positive bags with a low ratio prone to influence of large number negative instances; (2) correlation between local and global features pathology images has modeled; (3) is lack effective information interaction different magnifications. In this paper, we propose MILBooster, powerful...
Pathological images play a vital role in clinical cancer diagnosis. Computer-aided diagnosis utilized on digital Whole Slide Images (WSIs) has been widely studied. The major challenge of using deep learning models for WSI analysis is the huge size and existing methods struggle between end-to-end proper modeling contextual information. Most state-of-the-art utilize two-stage strategy, which they use pre-trained model to extract features small patches cut from then input these into...
Abstract Self-supervised learning plays an important role in molecular representation because labeled data are usually limited many tasks, such as chemical property prediction and virtual screening. However, most existing pre-training methods focus on one modality of data, the complementary information two modalities, SMILES graph, is not fully explored. In this study, we propose effective multi-modality self-supervised framework for graph. Specifically, graph first tokenized so that they...
As more and protein structures are discovered, blind protein-ligand docking will play an important role in drug discovery because it can predict complex conformation without pocket information on the target proteins. Recently, deep learning-based methods have made significant advancements docking, but their features suboptimal they do not fully consider difference between potential regions non-pocket feature extraction. In this work, we propose a pocket-guided strategy for guiding ligand to...
Abstract Proteins can be represented in different data forms, including sequence, structure, and surface, each of which has unique advantages certain limitations. It is promising to fuse the complementary information among them. In this work, we propose a framework called ProteinF3S for enzyme function prediction that fuses across protein surface. To achieve more effective fusion, multi-scale bidirectional fusion strategy between structure hierarchical features surface encoder interact with...
Weakly supervised whole slide image classification is usually formulated as a multiple instance learning (MIL) problem, where each treated bag, and the patches cut out of it are instances. Existing methods either train an classifier through pseudo-labeling or aggregate features into bag feature attention mechanisms then classifier, scores can be used for instance-level classification. However, pseudo labels constructed by former contain lot noise, latter not accurate enough, both which...
Active learning (AL) is an effective approach to select the most informative samples label so as reduce annotation cost. Existing AL methods typically work under closed-set assumption, i.e., all classes existing in unlabeled sample pool need be classified by target model. However, some practical clinical tasks, may contain not only that fine-grainedly classified, but also non-target are irrelevant tasks. cannot well this scenario because they tend a large number of samples. In paper, we...