A5040765329- Multimodal Machine Learning Applications
- Domain Adaptation and Few-Shot Learning
- Surgical Simulation and Training
- Anatomy and Medical Technology
- Advanced Neural Network Applications
- Soft Robotics and Applications
- Radiomics and Machine Learning in Medical Imaging
- Artificial Intelligence in Healthcare and Education
- Modular Robots and Swarm Intelligence
- Advanced Image and Video Retrieval Techniques
- Medical Image Segmentation Techniques
- Colorectal Cancer Surgical Treatments
- Colorectal Cancer Screening and Detection
- Topic Modeling
- Robotics and Sensor-Based Localization
- Robot Manipulation and Learning
- AI in cancer detection
- Bone Tumor Diagnosis and Treatments
- Non-Destructive Testing Techniques
- Image Processing Techniques and Applications
- Sarcoma Diagnosis and Treatment
- Cerebrovascular and Carotid Artery Diseases
- Micro and Nano Robotics
- Distributed systems and fault tolerance
- 3D Shape Modeling and Analysis
National University of Singapore
2017-2024
Johns Hopkins University
2024
University of York
2023
Chinese University of Hong Kong
2023
Suzhou Research Institute
2020
Global and local relational reasoning enable scene understanding models to perform human-like analysis understanding. Scene enables better semantic segmentation object-to-object interaction detection. In the medical domain, a robust surgical model allows automation of skill evaluation, real-time monitoring surgeon's performance post-surgical analysis. This paper introduces globally-reasoned multi-task capable performing instrument tool-tissue Here, we incorporate global in latent space...
Despite the availability of computer-aided simulators and recorded videos surgical procedures, junior residents still heavily rely on experts to answer their queries. However, expert surgeons are often overloaded with clinical academic workloads limit time in answering. For this purpose, we develop a question-answering system facilitate robot-assisted scene activity understanding from videos. Most existing visual question answering (VQA) methods require an object detector regions based...
Surgical data science is devoted to enhancing the quality, safety, and efficacy of interventional healthcare. While use powerful machine learning algorithms becoming standard approach for surgical science, underlying end-to-end task models directly infer high-level concepts (e.g., phase or skill) from low-level observations endoscopic video). This nature contemporary approaches makes vulnerable non-causal relationships in requires re-development all components if new tasks are be solved. The...
Despite impressive accuracy, deep neural networks are often miscalibrated and tend to overly confident predictions. Recent techniques like temperature scaling (TS) label smoothing (LS) show effectiveness in obtaining a well-calibrated model by logits hard labels with scalar factors, respectively. However, the use of uniform TS or LS factor may not be optimal for calibrating models trained on long-tailed dataset where produces probabilities high-frequency classes. In this study, we propose...
Several approaches have been introduced to understand surgical scenes through downstream tasks like captioning and scene graph generation. However, most of them heavily rely on an independent object detector region-based feature extractor. Encompassing computationally expensive detection extraction models, these multi-stage methods suffer from slow inference speed, making less suitable for real-time applications. The performance the also degrades inheriting errors earlier modules pipeline....
Origami-inspired soft and flexible robots have drawn immense attention in recent years for their wide range of medicine engineering applications. While the ability shape morphing presents a significant advantage, shape-invariant pose estimation techniques are still under-explored. Pose tracking vital to study, control automate locomotion origami robots. This paper proposes ScoopNet that performs semantic segmentation 6DOF origami-inspired worm A vision-based deep learning model can estimate...
Advances in GPT-based large language models (LLMs) are revolutionizing natural processing, exponentially increasing its use across various domains. Incorporating uni-directional attention, these autoregressive LLMs can generate long and coherent paragraphs. However, for visual question answering (VQA) tasks that require both vision with bi-directional attention or employing fusion techniques often employed to capture the context of multiple modalities all at once. As GPT does not natively...
Robotic assistance in Minimally Invasive Surgery (MIS) have extended the capabilities of surgeons via improved precision dexterity and computer assistance. By tapping on MIS, this paper aims to design a new tendon Fixation mechanism which utilizes springs actuate surgical tools for removal osseous giant cell tumor. We presents our preliminary conceptualization prototype development using spring backbone tendon-driven mechanism. investigating different routing mechanisms, first time study...
A lack of sensory feedback often hinders minimally invasive operations. Although endoscopy has addressed this limitation to an extent, endovascular procedures such as angioplasty or stenting still face significant challenges. Sensors that rely on a clear line sight cannot be used because it is unable gather in blood environments. During the stent deployment procedure, deployed stent's state critical partially open can affect flow. Despite this, no robust and noninvasive clinical solutions...
Surgical scene understanding is a key barrier for situation-aware robotic surgeries and the associated surgical training. With presence of domain shifts inclusion new instruments tissues, learning generalization (DG) plays pivotal role in expanding instrument-tissue interaction detection to domains surgery. Mimicking ability humans incrementally learn skills without forgetting their old similar domain, we employ incremental DG on graphs predict during robot-assisted To achieve DG,...
Despite the availability of computer-aided simulators and recorded videos surgical procedures, junior residents still heavily rely on experts to answer their queries. However, expert surgeons are often overloaded with clinical academic workloads limit time in answering. For this purpose, we develop a question-answering system facilitate robot-assisted scene activity understanding from videos. Most existing VQA methods require an object detector regions based feature extractor extract visual...
Robotic planning and execution in open-world environments is a complex problem due to the vast state spaces high variability of task embodiment. Recent advances perception algorithms, combined with Large Language Models (LLMs) for planning, offer promising solutions these challenges, as common sense reasoning capabilities LLMs provide strong heuristic efficiently searching action space. However, prior work fails address possibility hallucinations from LLMs, which results failures execute...
Visual question answering (VQA) in surgery is largely unexplored. Expert surgeons are scarce and often overloaded with clinical academic workloads. This overload limits their time questionnaires from patients, medical students or junior residents related to surgical procedures. At times, also refrain asking too many questions during classes reduce disruption. While computer-aided simulators recording of past procedures have been made available for them observe improve skills, they still...
The development and use of soft or flexible structural matters across various research domains have drastically increased in recent decades. Its flexible, compliant nature interactive safety made it a preferred candidate compared to its rigid bodied counterparts. However, the lack robust robot detection localization techniques has constrained feedback control system, limiting application. This paper proposes novel depth sensor-based tracking algorithm adaptive shape morphing robots. first...