TFThomas Fevens

Thomas Fevens

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Thomas Fevens has a total of 42 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Enhancing Memory Recall in LLMs with Gauss-Tin: A Hybrid Instructional and Gaussian Replay Approach

    by Iing Muttakhiroh, Thomas Fevens

    Despite the significant advancements in Large Language Models (LLMs), catastrophic forgetting remains a substantial challenge, where models lose previously acquired knowledge upon learning new information. Continual learning (CL) strategies have emerged as a potential solution to this problem, with replay-based techniques demonstrating superior performance in preserving learned knowledge. In this context, we introduce Gauss-Tin, a novel approach that integrates the replay strategy with a Gaussian mixture model to enhance the quality of sample selection during training, supplemented by instructional guidance to facilitate the generation of past learning. This method aims to improve LLMs' retention capabilities by strategically reinforcing important past learnings while accommodating new information. Our experimental results indicate a promising 6% improvement in retention metrics over traditional methods, suggesting that Gauss-Tin is an effective strategy for mitigating catastrophic forgetting in LLMs. This study underscores the potential of hybrid models in enhancing the robustness and adaptability of LLMs in dynamic learning environments.

  • Enhanced U-Net Architecture for Brain Tumor Localization and Segmentation in T1-Weighted MRI

    by Somayeh Davar, Thomas Fevens

    Detection and segmentation of Magnetic Resonance Imaging (MRI) scans is a critical task in medical imaging, where achieving high segmentation precision and reliability remains challenging due to variations in tumor shape, size, intensity, and boundary definition across different MRI modalities. Recently, deep learning techniques have significantly improved the efficiency oflocalization and segmentation of various medical fields, including brain tumoursanalysis. This paper presents a novel two-stage approach for brain tumour segmentation in T1-weighted contrast-enhanced MRI (CE-MRI) scans, leveraging both YOLO(you only look once) and Modified U-Net. In the first stage, YOLO is employed to quickly and accurately localize regions of interest (ROIs) where brain tumours are present. To accelerate this step, YOLOv3 are incorporated, improving the computation speed and efficiency of the model. In the second stage, a modified U-NET model, enhanced with spatial and channel attention modules, is utilized to perform precise segmentation of the tumours within the identified ROIs. The performance of the proposed framework is evaluated using metrics including precision, recall, Jaccard index, and Dice similarity coefficient (DSC). Our approach demonstrates promising results compared to previous methods using the same database.


  • Open Access

    Comparing the Drop Vertical Jump Tracking Performance of the Azure Kinect to the Kinect V2

    by Patrik Abdelnour, Kevin Y. Zhao, Athanasios Babouras, Jason Philip Aaron Hiro Corban, Nicolaos Karatzas, Thomas Fevens, Paul Andre Martineau

    Traditional motion analysis systems are impractical for widespread screening of non-contact anterior cruciate ligament (ACL) injury risk. The Kinect V2 has been identified as a portable and reliable alternative but was replaced by the Azure Kinect. We hypothesize that the Azure Kinect will assess drop vertical jump (DVJ) parameters associated with ACL injury risk with similar accuracy to its predecessor, the Kinect V2. Sixty-nine participants performed DVJs while being recorded by both the Azure Kinect and the Kinect V2 simultaneously. Our software analyzed the data to identify initial coronal, peak coronal, and peak sagittal knee angles. Agreement between the two systems was evaluated using the intraclass correlation coefficient (ICC). There was poor agreement between the Azure Kinect and the Kinect V2 for initial and peak coronal angles (ICC values ranging from 0.135 to 0.446), and moderate agreement for peak sagittal angles (ICC = 0.608, 0.655 for left and right knees, respectively). At this point in time, the Azure Kinect system is not a reliable successor to the Kinect V2 system for assessment of initial coronal, peak coronal, and peak sagittal angles during a DVJ, despite demonstrating superior tracking of continuous knee angles. Alternative motion analysis systems should be explored.


  • Comparing novel smartphone pose estimation frameworks with the Kinect V2 for knee tracking during athletic stress tests

    by Athanasios Babouras, Patrik Abdelnour, Thomas Fevens, Paul A. Martineau

    Purpose To compare the accuracy of the Microsoft Kinect V2 with novel pose estimation frameworks, in assessing knee kinematics during athletic stress tests, for fast and portable risk assessment of anterior cruciate ligament (ACL) injury. Methods We captured 254 varsity athletes, using the Kinect V2 and a smartphone application utilizing Google’s MediaPipe framework. The devices were placed as close as possible and used to capture a person, facing the cameras, performing one of three athletic stress tests at a distance of 2.5 ms. Custom software translated the results from both frameworks to the same format. We then extracted relevant knee angles at key moments of the jump and compared them, using the Kinect V2 as the ground truth. Results The results show relatively small angle differences between the two solutions in the coronal plane and moderate angle differences on the sagittal plane. Overall, the MediaPipe framework results seem to underestimate both knee valgus angles and knee sagittal angles compared to the Kinect V2. Conclusion This preliminary study demonstrates the potential for Google’s MediaPipe framework to be used for calculating lower limb kinematics during athletic stress test motions, which can run on most modern smartphones, as it produces similar results to the Kinect V2. A smartphone application similar to the one developed could potentially be used for low cost and widespread ACL injury prevention.


  • Open Access

    Comparing a Portable Motion Analysis System against the Gold Standard for Potential Anterior Cruciate Ligament Injury Prevention and Screening

    by Nicolaos Karatzas, Patrik Abdelnour, Jason Philip Aaron Hiro Corban, Kevin Y. Zhao, Louis-Nicolas Veilleux, Stephane G. Bergeron, Thomas Fevens, Hassan Rivaz, Athanasios Babouras, Paul A. Martineau

    Knee kinematics during a drop vertical jump, measured by the Kinect V2 (Microsoft, Redmond, WA, USA), have been shown to be associated with an increased risk of non-contact anterior cruciate ligament injury. The accuracy and reliability of the Microsoft Kinect V2 has yet to be assessed specifically for tracking the coronal and sagittal knee angles of the drop vertical jump. Eleven participants performed three drop vertical jumps that were recorded using both the Kinect V2 and a gold standard motion analysis system (Vicon, Los Angeles, CA, USA). The initial coronal, peak coronal, and peak sagittal angles of the left and right knees were measured by both systems simultaneously. Analysis of the data obtained by the Kinect V2 was performed by our software. The differences in the mean knee angles measured by the Kinect V2 and the Vicon system were non-significant for all parameters except for the peak sagittal angle of the right leg with a difference of 7.74 degrees and a p-value of 0.008. There was excellent agreement between the Kinect V2 and the Vicon system, with intraclass correlation coefficients consistently over 0.75 for all knee angles measured. Visual analysis revealed a moderate frame-to-frame variability for coronal angles measured by the Kinect V2. The Kinect V2 can be used to capture knee coronal and sagittal angles with sufficient accuracy during a drop vertical jump, suggesting that a Kinect-based portable motion analysis system is suitable to screen individuals for the risk of non-contact anterior cruciate ligament injury.


  • Open Access

    Source-free Domain Adaptation Requires Penalized Diversity

    by Laya Rafiee Sevyeri, Ivaxi Sheth, Farhood Farahnak, Alexandre See, Samira Ebrahimi Kahou, Thomas Fevens, Mohammad Havaei

    While neural networks are capable of achieving human-like performance in many tasks such as image classification, the impressive performance of each model is limited to its own dataset. Source-free domain adaptation (SFDA) was introduced to address knowledge transfer between different domains in the absence of source data, thus, increasing data privacy. Diversity in representation space can be vital to a model`s adaptability in varied and difficult domains. In unsupervised SFDA, the diversity is limited to learning a single hypothesis on the source or learning multiple hypotheses with a shared feature extractor. Motivated by the improved predictive performance of ensembles, we propose a novel unsupervised SFDA algorithm that promotes representational diversity through the use of separate feature extractors with Distinct Backbone Architectures (DBA). Although diversity in feature space is increased, the unconstrained mutual information (MI) maximization may potentially introduce amplification of weak hypotheses. Thus we introduce the Weak Hypothesis Penalization (WHP) regularizer as a mitigation strategy. Our work proposes Penalized Diversity (PD) where the synergy of DBA and WHP is applied to unsupervised source-free domain adaptation for covariate shift. In addition, PD is augmented with a weighted MI maximization objective for label distribution shift. Empirical results on natural, synthetic, and medical domains demonstrate the effectiveness of PD under different distributional shifts.

  • Open Access

    A Novel Approach for Selecting Effective Threshold Values in Ternary State Estimation Using Particle Swarm Optimization

    by Somayeh Davar, Thomas Fevens

    Inspired by recent breakthroughs in cyber-physical systems (CPSs) and their applications, in this paper, we propose a novel multi-objective method to optimize the threshold values within the ternary event-based framework. To reduce communication overhead, the particle swarm optimization (PSO) approach is applied as an optimizer to identify Pareto optimal set values of the threshold. The proposed optimization technique is subject to constraints to ensure its feasibility. The simulation results confirm the efficiency of the recommended method. Furthermore, the simulation results demonstrate that the proposed framework is comprehensive and capable of finding a wide variety of Pareto optimal ternary event-based state estimations for each predefined threshold.


  • Open Access

    Boosting Segmentation Performance across Datasets using Histogram Specification with Application to Pelvic Bone Segmentation

    by Prabhakara Subramanya Jois, Aniketh Manjunath, Thomas Fevens

    Accurate segmentation of the pelvic computerized tomography (CT) imaging is crucial for the clinical diagnosis of pelvic bone diseases and for planning patient-specific hip surgeries. Several methodologies have been proposed for such segmentation tasks with the emergence and advancements of deep learning for digital healthcare. But in a low data scenario, the lack of abundant data needed to train a Deep Neural Network is a significant bottleneck. In this work, we propose a methodology based on modulation of image tonal distributions and deep learning to boost the performance of networks trained on limited data. The strategy involves pre-processing of test data through histogram specification. This simple yet effective approach can be viewed as a style transfer methodology. The segmentation task uses a U-Net configuration with an EfficientNet-B0 backbone, optimized using an augmented BCE-IoU loss function. This configuration is validated on a total of 284 images taken from two publicly available CT datasets, The Cancer Imaging Archive (TCIA) and the Visible Human Project (VHBD). The average performance measures for the Dice coefficient and Intersection over Union are 95.7% and 91.9%, respectively, which give strong evidence for the effectiveness of the approach, which is highly competitive with state-of-the-art methodologies.