Publications by authors named "Max Meng"
Article Synopsis
- The letter introduces a new type of probe designed for photoacoustic tomography using a continuum robotic-steered system for endoluminal imaging.
- A compact coaxial optical and acoustic imaging system has been created, featuring a custom piezoelectric micromachined ultrasonic transducer, all contained within a 12 mm by 30 mm cylindrical housing.
- Testing has shown that this robotic-assisted system is effective for early detection of gastrointestinal cancers, successfully locating lesions and characterizing deeper abnormalities, indicating strong potential for future clinical use.
Renal cancer, although still rare among individuals under 45 years of age, is on the rise in the general population. The risk and timing of subsequent renal cancer in survivors of childhood cancer is not well established. Using the SEER registry, we reported the incidence of subsequent malignant renal neoplasms after treatment for primary malignancy diagnosed under 20 years of age.
View Article and Find Full Text PDFArticle Synopsis
- AI-assisted polyp segmentation is essential for early colorectal cancer detection but faces challenges due to insufficient annotated data and class imbalance in polyp segmentation.
- The study introduces PolypMixNet, a semi-supervised learning framework that leverages novel augmentation techniques and a Mean Teacher architecture to enhance model performance while addressing limited annotated data and class imbalance.
- Evaluated on four public colonoscopy datasets, PolypMixNet demonstrated impressive results, achieving 88.97% Dice and 88.85% mIoU scores, significantly outperforming existing semi-supervised methods even with just 15% labeled training data.
Article Synopsis
- Automatic segmentation of retinal OCT images is key for diagnosing and treating eye diseases but remains challenging due to blurry boundaries and insufficient labeled data.
- The proposed Boundary-Enhanced Semi-supervised Network (BE-SemiNet) uses an auxiliary distance map regression task to improve segmentation accuracy while working with limited labeled data and ample unlabeled data.
- Experiments show that BE-SemiNet dramatically enhances performance compared to existing methods, making it a promising approach for practical applications in clinical settings where obtaining labeled data is difficult.
A photoacoustics-enhanced drilling probe for radiation-free pedicle screw implantation in spinal surgery.
Front Bioeng Biotechnol
September 2022
Article Synopsis
- The article introduces a new system for spinal surgery that uses 3D ultrasound imaging and photoacoustic sensing to improve the accuracy of pedicle screw placement.
- The 3D ultrasound tracks patient movement during surgery to guide entry points and drilling paths, while the photoacoustic sensing differentiates between types of bone to prevent damage to surrounding tissues.
- A pilot study on a bovine spine shows that this technology could enhance the safety and effectiveness of spinal stabilization procedures in the future.
Design and Control of a Magnetically-Actuated Capsule Robot With Biopsy Function.
IEEE Trans Biomed Eng
September 2022
Article Synopsis
- Developed a magnetically-actuated biopsy capsule (MABC) robot for gastrointestinal diagnosis that can not only capture images but also collect tissue samples.
- The robot moves using external electromagnetic actuation, with capabilities for both plane motion on the GI tract's surface and 3D motion in space.
- Prototype tests showed precise movement and sampling accuracy, highlighting the potential for enhanced diagnostic techniques in clinical settings without using internal power sources.
Article Synopsis
- The paper introduces a new photoacoustic computed tomography (PACT) system that uses dual ultrasonic transducers with varying frequencies to improve imaging of complex biological tissues.
- This system addresses the problem of low imaging quality due to missing PA signals that occur when using a single-frequency transducer.
- Experimentation shows that the new system achieves high resolution and ideal brightness/contrast, making it promising for clinical use where frequency limitations exist.
A Virtual Scanning Framework for Robotic Spinal Sonography with Automatic Real-time Recognition of Standard Views.
Annu Int Conf IEEE Eng Med Biol Soc
November 2021
Article Synopsis
- Ultrasound imaging is often used for spinal diagnosis, but traditional manual scanning can be physically and mentally taxing for sonographers.
- Robotic ultrasound systems offer a way to reduce operator strain by automating the scanning process, though they typically lack real-time image interpretation capabilities.
- This study proposes a robotic system that automates spinal scans and includes advanced image recognition using deep learning techniques, achieving a 96.71% accuracy in identifying standard spinal views.
APRNet: Alternative Prediction Refinement Network for Polyp Segmentation.
Annu Int Conf IEEE Eng Med Biol Soc
November 2021
Article Synopsis
- Colorectal cancer is now the second leading cause of cancer-related deaths, highlighting the need for effective automatic polyp segmentation in screening systems.
- The proposed model, APRNet, builds upon UNet architecture to improve the segmentation accuracy by refining predictions at various resolutions using prediction residual refinement modules.
- APRNet has demonstrated impressive results on benchmark datasets, achieving new state-of-the-art performance with high accuracy and dice scores, making it a promising tool for assisting in polyp detection during colonoscopy procedures.
Tip estimation approach for concentric tube robots using 2D ultrasound images and kinematic model.
Med Biol Eng Comput
August 2021
Article Synopsis
- Concentric tube robots (CTR) are designed for minimally invasive surgery, offering high dexterity and compact size, but require accurate tip position tracking for safe operation.
- The proposed method utilizes 2D ultrasound images along with a forward kinematic model to optimize the ultrasound scanning process, needing only three scan positions for each tube while reconstructing its shape.
- The technique demonstrates a tip estimation accuracy of 0.59 mm and can be applied to existing robotic systems without structural modifications.
Purpose: The COVID-19 pandemic has led to the cancellation or deferment of many elective cancer surgeries. We performed a systematic review on the oncological effects of delayed surgery for patients with localised or metastatic renal cell carcinoma (RCC) in the targeted therapy (TT) era.
Method: The protocol of this review is registered on PROSPERO(CRD42020190882).
Article Synopsis
- COVID-19 often requires patients to have severe respiratory support, sometimes leading to extended use of mechanical ventilation in ICU settings.
- Open Tracheostomy (OT) is preferred for its clear visibility during surgery, but Percutaneous Tracheostomy (PT) is a less invasive option that has limitations due to its reliance on surface landmarks for placement.
- The proposed solution is a flexible mini-robotic system that utilizes robotic needling technology to perform a more precise and controlled tracheal puncture from an "inside-out" approach, potentially minimizing complications associated with PT.
Article Synopsis
- Point-based rigid registration (PBRR) techniques are essential in image-guided surgery (IGS) for accurately tracking surgical tools and estimating target registration error (TRE), which helps surgeons make better decisions during procedures.
- The paper introduces an improved TRE estimation algorithm that accounts for heterogeneous fiducial localization errors (FLE), allowing for differences in error between fiducials and in various directions.
- Monte Carlo simulation results show that this new algorithm performs better than traditional methods, especially in cases with heterogeneous FLE, making it useful for online tracking of surgical tool errors during IGS.
A Novel Approach for Automatic State Detection of A Magnetically Actuated Capsule.
Annu Int Conf IEEE Eng Med Biol Soc
July 2020
Article Synopsis
- Recent advancements in Simultaneous Magnetic Actuation and Localization (SMAL) technology aim to enhance the functionality of wireless capsule endoscopy (WCE) in the intestine.
- The study introduces a new approach to assess the capsule's state, which helps improve localization accuracy by fitting a relationship between theoretical magnetic field values and actual measurements.
- Experimental tests on models show that the proposed method successfully detects various states of the capsule during its magnetic actuation.
Zoom in Lesions for Better Diagnosis: Attention Guided Deformation Network for WCE Image Classification.
IEEE Trans Med Imaging
December 2020
Article Synopsis
- Wireless capsule endoscopy (WCE) is a noninvasive imaging technique for visualizing the gastrointestinal tract, but current image classification methods like CNNs struggle with small lesions and background noise.
- A new two-branch Attention Guided Deformation Network (AGDN) is proposed to enhance WCE image classification by using attention maps to focus on and amplify lesion areas, along with incorporating Third-order Long-range Feature Aggregation (TLFA) modules for better context and feature representation.
- Experimental results demonstrate that AGDN achieves superior performance with a 91.29% classification accuracy on public WCE datasets, outperforming existing methods.
Design of a multi-arm concentric-tube robot system for transnasal surgery.
Med Biol Eng Comput
March 2020
Article Synopsis
- The concentric tube robot (CTR) is gaining interest for its compact design and ability to control curved movements, but current prototypes are mainly single-arm, limiting their functionality to simple tasks like drug delivery.
- This paper introduces a three-arm CTR system that includes a four-degree-of-freedom (DOF) vision arm and two six-DOF manipulator arms, equipped with specialized end effectors for various surgical procedures.
- Experimental results show a mean motion accuracy of 0.33 mm, and a tissue excision experiment in a skull model demonstrates the system's potential in treating nasopharyngeal carcinoma.
Article Synopsis
- Adequate bladder tumor detection is crucial during transurethral resection (TURBT) to minimize cancer recurrence, but standard white light cystoscopy misses about 20% of tumors.
- Researchers developed a deep learning algorithm called CystoNet to enhance tumor detection during cystoscopy by using video frames of confirmed bladder cancers for training and testing the model.
- CystoNet demonstrated impressive results in a validation study, showing 90.9% sensitivity and 98.6% specificity, indicating it could significantly improve bladder cancer detection and surgical outcomes.
Article Synopsis
- Autonomous mobile robots face challenges in navigating complex 3D environments, particularly with features like staircases and slopes.
- The study presents an integrated navigation system that uses a novel SLAM framework to create a 3D OctoMap, allowing the robot to differentiate between slopes and staircases.
- The system enhances navigation efficiency by employing a variable step size Rapidly-exploring Random Tree (RRT) method and includes a camera re-localization technique for improved stability in 3D localization, demonstrating success in various environments.
Article Synopsis
- The paper presents a novel path planning method for robotic luggage trolleys at airports, focusing on efficient collection while being considerate of pedestrians.
- It formulates the problem as a Traveling Salesman Problem (TSP), but instead of using standard distance metrics, it employs a complex metric that accounts for pedestrian emotions and interactions.
- Utilizing a potential function and the Social Force Model (SFM), the method generates a path that minimizes conflict with pedestrians and uses a Self-Organizing Map (SOM) to determine the optimal route for the robot to collect all luggage trolleys effectively.
Article Synopsis
- Recent advancements in positioning and tracking for capsule robots in the gastrointestinal tract aim to provide real-time spatial data, but current methods struggle with the GI tract's unstructured environment.
- A new relative position estimation method uses the robot's movement distance to give a more accurate position by first obtaining absolute positions through magnetic tracking and then fitting the movement trajectory with a Bézier curve.
- This method improves upon existing techniques by enabling more precise guidance for medical instruments, requires no external reference objects on the patient, and demonstrates effective distance estimation in experiments.
Article Synopsis
- - The paper introduces an enhanced calibration method for a rotating two-dimensional light detection and ranging (R2D-LIDAR) system, crucial for creating accurate 3D environmental maps.
- - The R2D-LIDAR, which integrates a 2D LIDAR with a rotating unit, is valued in robotics for its affordability and detailed data, but requires careful calibration to address device misalignment and wear.
- - This work uses the Levenberg-Marquardt algorithm to effectively correct biases in the 2D LIDAR, with experimental results showing a reliable error range in sensor offset estimation of -15 mm to 15 mm, ensuring better scan accuracy.
Article Synopsis
- Understanding surgical tool-tip tracking error is crucial for effective image-guided surgery decision-making.
- A new error metric called total target registration error (TTRE) is introduced, which considers target localization error across two registration spaces.
- The authors validate their error model through Monte Carlo simulations, achieving over 90% accuracy in matching simulated and theoretical tool-tip tracking error statistics.
Gastrointestinal bleeding detection in wireless capsule endoscopy images using handcrafted and CNN features.
Annu Int Conf IEEE Eng Med Biol Soc
July 2017
Article Synopsis
- The paper discusses a new method for detecting gastrointestinal bleeding using wireless capsule endoscopy (WCE) that integrates both handcrafted features and features from convolutional neural networks (CNNs).
- The authors have developed a smaller CNN architecture to reduce computational costs while maintaining effectiveness.
- Experimental results indicate that this new approach performs well even with limited training data, achieving results on par with or superior to current state-of-the-art methods.
Continuous systolic and diastolic blood pressure estimation utilizing long short-term memory network.
Annu Int Conf IEEE Eng Med Biol Soc
July 2017
Article Synopsis
- A new method for estimating blood pressure using a long short-term memory (LSTM) neural network is introduced to enhance chronic disease monitoring.
- The paper also presents a novel ambulatory blood pressure processing technique called the Two-stage Zero-order Holding (TZH) algorithm, which improves upon traditional Pulse Transit Time (PTT) methods.
- Results show that the LSTM-based approach achieves low Root-Mean-Squared Errors (RMSE) of 2.751 mmHg for systolic and 1.604 mmHg for diastolic pressure, indicating strong accuracy and potential for integration into health monitoring systems.
Motion artifact reduction in photoplethysmogram signals based on Periodic Component Factorization.
Annu Int Conf IEEE Eng Med Biol Soc
July 2017
Article Synopsis
- Continuous blood pressure measurement using pulse transit time (PTT) has been extensively researched, but its accuracy is significantly compromised by hand movements during exercise.
- A new algorithm, Periodic Component Factorization (PCF), has been developed to effectively remove motion artifacts from photoplethysmography (PPG) signals, improving the accuracy of blood pressure estimations.
- PCF outperforms traditional methods like FastICA by successfully extracting dependent source components from noisy PPG signals, particularly in scenarios where the signals exhibit periodic patterns.