Melissa Boswell’s Post

🔍 Investigating Preprocessing Techniques for Enhanced EEG Classification 🧠 By Mohamed Rasmy Helwan University Cairo This study examines the impact of various preprocessing techniques on EEG signal classification using a CNN-BiLSTM model. Utilizing the inherently noisy BCI Competition 2008 – Graz dataset, Independent Component Analysis (ICA) was applied to all files for artifact removal, with Artifact Subspace Reconstruction (ASR) employed selectively: once across all files and another time on only the noisiest subjects. The analysis also evaluated the effect of epoching before applying ICA and ASR. While applying ASR to only the noisiest subjects showed limited impact, applying it across the entire dataset led to slight performance improvements. A refined CNN-BiLSTM architecture was developed to better handle non-stationary signals, and bias initialization in BiLSTM units was also explored, though it was not the primary focus of this study. The findings highlight the importance of optimal preprocessing, careful epoching, and model design in improving EEG classification performance for Brain-Computer Interface (BCI) applications. #Neurotechnology #EEG #BCI #ML #quantum #medtech #cerebralink #MachineLearning #BrainComputerInterface #SignalProcessing #AI #HealthcareInnovation José Manuel Muñoz Harry Lambert Allan McCay Vasyl Mykytiuk Helwan University Faculty of Medicine, Helwan University Blackrock Neurotech Neuralink 🔗 👇 👇👇👇👇👇 https://lnkd.in/eMjCdc7Y

【Medicine/Health】 Meta-Learning of Motor Skills in the Dorsal Premotor Cortex of the Brain Researchers at University of Tsukuba have discovered that the dorsal premotor cortex serves a "meta-learning" function, overseeing and regulating physical movements. Once believed to be limited to movement planning, this region has now been shown, through computational modeling and brain stimulation, to also facilitate the retention and forgetting of motor memories. Read more details here; https://lnkd.in/gnhx3y8V Original Paper; https://lnkd.in/gyBuY2Bm

AR meets biomechanics! 🎉 Imagine seeing real-time muscle simulations projected directly onto the body. A team of researchers - Xingyao Yu, Michael Sedlmair, #DavidRosin, #OliverRöhrle, Johannes Kässinger, #FrankDürr, #ChristianBecker, and #BenjaminLee - made this a reality with their latest work "PerSiVal: On-Body AR Visualization of Biomechanical Arm Simulations". 💪✨ Using AR, motion capture, and neural networks, they developed prototypes to visualize upper-arm muscles dynamically, making complex data accessible and engaging. This innovation promises better tools for physiotherapists, educators, and researchers alike - bridging the gap between data and understanding. 🌍💡 See https://shorturl.at/pUevD for more and check out the full paper in IEEE CG&A via DOI: 10.1109/MCG.2024.3494598. 📚🔗

Gait phase detection by using a portable system and artificial neural network https://lnkd.in/gGbux8Ng Gait phases are important to evaluate the walking function and to identify the characteristics of pathological gaits. However, it is difficult to differentiate gait phases outside gait laboratories, thus, this study aimed to develop a method to detect 8 gait sub-phases using a wearable multiple sensor system and artificial neural network (ANN). Motion sensors were used to acquire the acceleration of lower limbs, and force sensitive resistors were used to detect contact state and force between the foot and the ground. Walking was recorded using a high-speed camera. Two feed forward back-propagation (BP) neural networks were developed. The resilient BP algorithm was used to train ANN. A total of 66 volunteers participated in this study. For the stance and swing phase detection, simulation of the training data showed an accuracy of 98.0 %. The data from the test set showed a recognition accuracy of 97.75 %. Because the ending point of the last phase ‘Terminal Swing’ is always 100 % GC, we only listed seven phases. The prediction accuracy of seven phases were: 35.9 %, 63.8 %, 93.6 %, 94.9 %, 94.8 %, 97.9 % and 98 % using the limb acceleration data only. The average accuracy for seven phases were 68 %, 91.3 %, 97.8 %, 98.9 %, 98.8 %, 99.1 %, and 99.5 % using the limb acceleration and foot pressure data for fast, normal, and slow gait speeds. This study provides a new method for eight gait sub-phases detection with high accuracy combining a wearable system and ANN, which may make gait phase analysis possible under free-living conditions. #Gait phases #Wearable system #Artificial neural network #Acceleration #Force sensitive resistor #technology #research #innovation

DeepLabCut is an open-source toolbox used for 3D markerless pose estimation in animals and humans. It allows researchers to estimate body poses across different species and behaviors without needing physical markers, making it a versatile tool for behavioral and movement analysis. By using deep learning, specifically convolutional neural networks (CNNs), DeepLabCut can track body parts with high precision, even when there are complex movements or occlusions. This tool is widely adopted in neuroscience, biomechanics, and ethology because it provides robust pose estimation in various environments and for different animal models, from mice to primates. It supports multi-camera setups for 3D pose reconstruction, enabling researchers to obtain spatially accurate body part coordinates from multiple angles, making it ideal for studying behaviors in naturalistic settings. The system also adapts well to new species through transfer learning, requiring minimal labeled data for training.

At the dawn of generative AI, it's crucial not to overlook the foundations of machine learning. Our advances in this field continue to bring significant value, particularly in specific applications. I'm delighted to share our latest paper, which explores the use of radiomics to characterize parotid tumors. This approach combines Feel free to check it out and share your thoughts !. Ammari, S., Quillent, A., Elvira, V. et al. Using Machine Learning on MRI Radiomics to Diagnose Parotid Tumours Before Comparing Performance with Radiologists: A Pilot Study. J Digit Imaging. Inform. med. (2024). https://lnkd.in/eReWzmuU Thanks to : Arnaud Quillent & Émilie Chouzenoux Centre de Vision Numérique, OPIS, CentraleSupélec, Inria Víctor Elvira School of Mathematics, University of Edinburgh Garcia Gabriel ,François Bidault, Corinne Balleyguier & Nathalie Lassau Biomaps and Department of Imaging, Gustave Roussy #Radiomics #Machine learning #CVN #Gustaveroussy #INRIA

DeepLabCut is an open-source toolbox used for 3D markerless pose estimation in animals and humans. It allows researchers to estimate body poses across different species and behaviors without needing physical markers, making it a versatile tool for behavioral and movement analysis. By using deep learning, specifically convolutional neural networks (CNNs), DeepLabCut can track body parts with high precision, even when there are complex movements or occlusions. This tool is widely adopted in neuroscience, biomechanics, and ethology because it provides robust pose estimation in various environments and for different animal models, from mice to primates. It supports multi-camera setups for 3D pose reconstruction, enabling researchers to obtain spatially accurate body part coordinates from multiple angles, making it ideal for studying behaviors in naturalistic settings. The system also adapts well to new species through transfer learning, requiring minimal labeled data for training.

The hippocampus (HPC) plays a critical role in spatial navigation, learning, and memory. Neuronal activity in the HPC shows that neurons fire when an animal is located at regions of different space sizes in the world. In addition, when memory consolidation-processes in the HPC are inactivated, animals can navigate in familiar environments but not in new ones. How can these cognitive tasks related in the brain? In “A model of how hierarchical representations constructed in the hippocampus are used to navigate through space” Chalmers et al., 2024 present a computational model in a reinforcement learning setup that shows insights that might help us to understand how these findings work in the brain. Using this model, the authors reproduce results in rodent HPC inactivation experiments where the proposed hierarchical spatial representation shows how spatial navigation, learning and memory are affected when the HPC is impaired. I like this paper not only because shows a nice intersection between neuroscience and AI research that I am really interested in but also because it is a collaboration with my friend and colleague Eric Chalmers at Mount Royal University. Many of these ideas came from chats we had when we both were postdocs at the Canadian Centre for Behavioural Neuroscience at The University of Lethbridge. Also we got to work with Matthieu B. and Rob McDonald in this project. Great group of people! Paper: https://lnkd.in/g8T6GExs #NeuroAI #MachineLearning #ArtificialIntelligence #Hippocampus #SpatialNavigation

Alert! New #PhD position immediately available at NeuroSpin in the Inria- Commissariat a l'Energie Atomique et aux Energies Alternatives #MIND team in the context of the [ AIDAS] German-French lab on #AI, #DataAnalytics and #Simulation. #Topic: Transfer & unsupervised learning for high-resolution fMRI image reconstruction at ultra-high magnetic field (11.7 Tesla). #Description and #application procedure available here: https://lnkd.in/eJ79QaEt #Deadline: December, 31 2024. Keywords: #brain #neuroimaging #fMRI #deeplearning #transferlearning #Simulation. CEA Joliot. Ulugbek Kamilov, Tolga Cukur, Patrícia Figueiredo,

For anyone interested in watching the talks at the #NIH #BRAINInitiative #NeuroAI workshop I was at last week, the full videos are available. Day 1 is available at: https://lnkd.in/gTkJ4-rY Day 2 is available at: https://lnkd.in/g3hpd8Rg My talk is min 54:25:00 to 1:04:20 on Day 1. There are several extremely interesting talks and panels on both days. In particular, there is considerable focus on using AI to build a foundation model for the human brain that can be used as the basis of virtual experiments and diagnosis. The idea seems far too ambitious at this point, but the effort is producing truly amazing results on specific topics such as a new understanding of what features vision may be using beyond the ones traditionally understood. It’s really exciting. Day 2 focuses more on neuromorphic hardware and robotics, which is also exciting and critical. I think that neuromorphic hardware will be crucial for scaling AI to where it needs to go.