A5049297116- COVID-19 diagnosis using AI
- Anomaly Detection Techniques and Applications
- Advanced Neural Network Applications
- Machine Learning and Data Classification
- Radiomics and Machine Learning in Medical Imaging
- Domain Adaptation and Few-Shot Learning
- Human Pose and Action Recognition
- Infrastructure Maintenance and Monitoring
- Music and Audio Processing
- Anatomy and Medical Technology
- Digital Media Forensic Detection
- AI in cancer detection
- Diabetic Foot Ulcer Assessment and Management
- Brain Tumor Detection and Classification
- Medical Imaging and Analysis
- Respiratory viral infections research
- Data-Driven Disease Surveillance
- Lung Cancer Diagnosis and Treatment
- Biomedical Text Mining and Ontologies
- Digital Imaging for Blood Diseases
- Speech and Audio Processing
- Geophysical Methods and Applications
- Advanced X-ray and CT Imaging
- Artificial Intelligence in Healthcare
- Text and Document Classification Technologies
Cornell University
2025
Australian Centre for Robotic Vision
2020-2024
The University of Adelaide
2020-2024
Consistency learning using input image, feature, or network perturbations has shown remarkable results in semi-supervised semantic segmentation, but this approach can be seriously affected by inaccurate predictions of unlabelled training images. There are two consequences these predictions: 1) the based on "strict" cross-entropy (CE) loss easily overfit prediction mistakes, leading to confirmation bias; and 2) applied will use potentially erroneous as signals, degrading consistency learning....
Effective semi-supervised learning (SSL) in medical image analysis (MIA) must address two challenges: 1) work effectively on both multi-class (e.g., lesion classification) and multi-label multiple-disease diagnosis) problems, 2) handle imbalanced (because of the high variance disease prevalence). One strategy to explore SSL MIA is based pseudo labelling strategy, but it has a few shortcomings. Pseudo-labelling general lower accuracy than consistency learning, not specifically design for can...
Recent advances in prototypical learning have shown remarkable potential to provide useful decision interpretations associating activation maps and predictions with class-specific training prototypes. Such has been well-studied for various single-label diseases, but quite relevant more challenging multi-label diagnosis, where multiple diseases are often concurrent within an image, existing models struggle obtain meaningful effective class prototypes due the entanglement of diseases. In this...
The deployment of automated deep-learning classifiers in clinical practice has the potential to streamline diagnosis process and improve accuracy, but acceptance those relies on both their accuracy interpretability. In general, accurate provide little model interpretability, while interpretable models do not have competitive classification accuracy. this paper, we introduce a new framework, called InterNRL, that is designed be highly interpretable. InterNRL consists student-teacher where...
Minimally invasive surgery (MIS) is among the preferred procedures for treating a number of ailments as patients benefit from fast recovery and reduced blood loss. The trade-off that surgeons lose direct visual contact with surgical site have limited intra-operative imaging techniques real-time feedback. Computer vision methods well segmentation tracking tissues tools in video frames, are increasingly being adopted to MIS alleviate such limitations. So far, most advances been focused on...
<title>Abstract</title> Heart failure (HF), a major global health challenge, affects millions worldwide and poses substantial healthcare economic burdens. The left ventricular ejection fraction (LVEF) is critical dynamic parameter used to characterize HF guide treatment. In this study, we developed validated an artificial intelligence (AI) model capable of predicting abnormal LVEF directly from static, non-gated, non-contrast chest computed tomography (CT) scans, novel application for...
Constructing effective representation of lesions is essential for disease classification and localization in medical image analysis. Prototype-based models address this by leveraging visual prototypes to capture representative lesion patterns, yet effectively handling the complexity diverse characteristics remains a critical challenge, as they typically rely on single-level, fixedsize suffer from prototype redundancy. In paper, we present HierProtoPNet, new prototypebased framework designed...
Prototypical-part methods, e.g., ProtoPNet, enhance interpretability in image recognition by linking predictions to training prototypes, thereby offering intuitive insights into their decision-making. Existing which rely on a point-based learning of typically face two critical issues: 1) the learned prototypes have limited representation power and are not suitable detect Out-of-Distribution (OoD) inputs, reducing decision trustworthiness; 2) necessary projection back space images causes...
Deep learning models can extract predictive and actionable information from complex inputs. The richer the inputs, better these usually perform. However, that leverage rich inputs (e.g., multi-modality) be difficult to deploy widely, because some may missing at inference. Current popular solutions this problem include marginalization, imputation, training multiple models. Marginalization obtain calibrated predictions but it is computationally costly therefore only feasible for low...
3D medical image segmentation methods have been successful, but their dependence on large amounts of voxel-level annotated data is a disadvantage that needs to be addressed given the high cost obtain such annotation. Semi-supervised learning (SSL) solve this issue by training models with unlabelled and small labelled dataset. The most successful SSL approaches are based consistency minimises distance between model responses obtained from perturbed views data. These perturbations usually keep...
The training of deep learning models generally requires a large amount annotated data for effective convergence and generalisation. However, obtaining high-quality annotations is laboursome expensive process due to the need expert radiologists labelling task. study semi-supervised in medical image analysis then crucial importance given that it much less obtain unlabelled images than acquire labelled by radiologists. Essentially, methods leverage sets enable better generalisation using only...
Segmentation is a crucial task in the medical imaging field and often an important primary step or even prerequisite to analysis of volumes. Yet treatments such as surgery complicate accurate delineation regions interest. The BraTS Post-Treatment 2024 Challenge published first public dataset for post-surgery glioma segmentation addresses aforementioned issue by fostering development automated tools MRI data. In this effort, we propose two straightforward approaches enhance performances deep...
Recent advances in prototypical learning have shown remarkable potential to provide useful decision interpretations associating activation maps and predictions with class-specific training prototypes. Such has been well-studied for various single-label diseases, but quite relevant more challenging multi-label diagnosis, where multiple diseases are often concurrent within an image, existing models struggle obtain meaningful effective class prototypes due the entanglement of diseases. In this...
Intra-operative automatic semantic segmentation of knee joint structures can assist surgeons during arthroscopy in terms situational awareness. However, due to poor imaging conditions (e.g., low texture, overexposure, etc.), is a challenging scenario, which justifies the scarce literature on this topic. In paper, we propose novel self-supervised monocular depth estimation regularise training arthroscopy. To further estimation, use clean images captured by stereo arthroscope routine objects...