Biomedical Signal Processing

Real-Time Heart Rate Monitoring Using Computer Vision & Signal Processing ❤️📊 I’ve been working on an exciting project that combines computer vision, signal processing, and real-time data analysis to estimate heart rate (BPM) from facial detection using a webcam. 🎥💡 How It Works: ✅ Face Detection: Using cvzone‘s FaceDetector, we accurately locate the user’s face in real-time. ✅ Color Magnification: A Gaussian Pyramid is applied to amplify subtle color changes caused by blood flow. ✅ Fourier Transform: We extract frequency components corresponding to pulse rate. ✅ Bandpass Filtering: Only relevant heart rate frequencies (1-2 Hz) are retained. ✅ Visualization: BPM values are plotted dynamically for real-time monitoring. Tech Stack: 🖥️ OpenCV | 🧠 cvzone | ⚡ NumPy | 🎛️ FFT | 📈 Signal Processing Key Learnings & Challenges: 🔹 Fine-tuning parameters like Gaussian levels & frequency range significantly impacts accuracy. 🔹 Efficient real-time processing is critical to avoid lag. 🔹 Signal noise handling is essential for reliable BPM estimation. 🚀 This technique has potential applications in health monitoring, fitness tracking, and remote diagnostics. Would love to hear your thoughts on its real-world applications! #MachineLearning #ComputerVision #HealthTech #SignalProcessing #OpenCV #Python #RealTimeAI #BPMDetection

Heterogeneous datasets are pervasive today, existing in various domains. Objects within these complex datasets are often represented from different perspectives, at different scales, or through multiple modalities, such as images, sensor readings, language sequences, and compact mathematical statements. Such datasets have been analyzed in the past using Multi-View Learning (MVL), Multi-Task Learning (MTL), and Tensor Learning (TL). In recent years, Multi-Modal Learning (MML) has also been employed. MML is a Machine Learning (ML) approach that integrates and processes information from multiple types of data, with different "perspectives" or "modalities" such as text, images, audio, video, or sensor data. The goal of MML is to leverage the complementary strengths of these modalities to improve model performance and enable richer understanding and predictions. Precision medicine and personalized clinical decision support systems (CDSS) tools have long aimed to leverage multimodal patient data to better capture complex, high-dimensional patient states and provider responses. This data ranges from free-form text notes and semi-structured electronic health records (EHR) to high-frequency physiological signals. While the advent of transformer architectures has enabled deeper insights from merging modalities, it has also required meticulous feature engineering and alignment. In patient monitoring, effectively analyzing diverse physiological signals within CDSS is highly challenging. #MedicalInformatics To address the challenges of analyzing multimodal patient data, the authors of [1] introduce MedTsLLM, a general multimodal large language model (LLM) framework that effectively integrates time series data and rich contextual information in the form of text. This framework performs three clinically relevant tasks (in time-series) which enable deeper analysis of physiological signals and can provide actionable insights for clinicians: • semantic segmentation • boundary detection • anomaly detection At a high level, boundary detection splits signals into periods like breaths or beats. Semantic segmentation further splits time series into distinct, meaningful segments. Anomaly detection identifies periods within the signals that deviate from normal. MedTsLLM utilizes a reprogramming layer to align embeddings of time series patches with a pretrained LLM's embedding space, making effective use of raw time series in conjunction with textual context. They additionally tailored the text prompt to include patient-specific information. Their experiments showed that MedTsLLM outperforms state-of-the-art baselines, including deep learning models, other LLMs, and clinical methods, across multiple medical domains, specifically electrocardiograms (ECG) and respiratory waveforms. Links to their preprint [1] and #Python GitHub repository [2] are shared in the comments.

Building AI on brain signals changed how I think about machine learning. On paper, it looks straightforward: Collect data -> extract features -> train a model. In reality, it is nothing like that. Here is what makes brain signal–based AI so difficult: - The data is messy. Brain signals are full of noise, artifacts, and inconsistencies. Even small movements can distort the signal. - Every human brain is different. A model that works for one person often fails for another. Generalization becomes a real challenge. - Labels are not always reliable. Emotions, cognitive states, and even disease stages are not perfectly defined. You are often learning from imperfect ground truth. - Feature engineering is not optional. Frequency patterns, temporal dynamics, spatial relationships. Ignoring domain knowledge leads to weak models. - Evaluation is tricky. Accuracy alone is misleading. You need subject-wise validation, cross-dataset testing, and robustness checks. While working on emotion recognition and Alzheimer’s classification using brain signals, I realized: 1. The challenge is not building the model. 2. The challenge is making sense of the data. That is where most AI systems fail. If you are working with real-world signals (brain signals, sensor data, etc.), you are not just doing machine learning. You are doing signal understanding. And that changes everything. #ArtificialIntelligence #AIEngineering #MachineLearning #DeepLearning #DataScience #ComputerScience #ComputerEngineering #SignalProcessing #NeuroAI #BrainSignals #HealthcareAI #MedicalAI #AIinHealthcare #RealWorldAI #AISystems #MLEngineering #AIResearch #EngineeringLife #TechCareers #BuildInPublic #WomenInTech #STEM #UETLahore #UET #PakistanTech #FutureOfAI #LearningInPublic #AICommunity #ResearchJourney #StudentEngineer

Retinal laser Doppler holography provides more than an image of vessels: the derived blood-velocity waveforms capture a beat-resolved signature of microvascular function, obtained non-invasively in a seconds-long exam. To turn this signal into clinically reliable readouts, we extract a wide panel of quantitative pulse-shape descriptors and validate them at cohort scale across clinical sites, retaining only those that are low-variability, high-significance, and interpretable. Our goal is to build the first comprehensive, transportable library of microvascular health metrics that remains stable across sites, devices, and operators. This is enabled by a standardized processing chain and explicit quality control at every step. We currently prioritize waveform-derived features such as composite morphology descriptors, timing and phase relationships, harmonic damping, and stroke-distance (velocity–time integral) partitions, chosen for their physiological interpretability and their potential robustness. We welcome interns eager to learn advanced signal processing and contribute to these tasks (https://lnkd.in/e2UWc9bt). The rationale is straightforward: pulse morphology encodes meaningful aspects of pulsatile vascular behavior—damping and spectral smoothing, wave reflection and notch dynamics, pulse sharpness, harmonic balance, and compliance-related effects—while remaining sensitive to clinically relevant modulators including intraocular pressure, intracranial pressure, and ocular perfusion pressure. Retinal Doppler holography is particularly well suited to this approach because it can resolve multiple cardiac harmonics (often ~10) in high-SNR vessel segments, enabling quality-control-gated, robust endpoints rather than noise-limited shape estimates. Icahn School of Medicine at Mount Sinai Institut Langevin - Ondes et images EPITA: Ecole d'Ingénieurs en Informatique PSL Research University Pitt Department of Ophthalmology/UPMC Vision Institute He Vision Group 何氏集团 University of Vienna Fondation Adolphe de Rothschild Hôpital national des Quinze-Vingts University of Melbourne CFIN - Center of Functionally Integrative Neuroscience Sun Yat-sen University

It was during a casual Zoom call with a former biotech CEO, now a few years into a lucrative career at a prominent hedge fund, that the thought first hit me. As he described the algorithms his team developed to detect subtle patterns in currency fluctuations, I couldn’t help but notice how similar they were to the signal processing methods used to model complex biological systems. The mathematics, the conceptual frameworks, and even the challenges of signal-to-noise optimization were identical, only applied to a different kind of dataset. "We're using third-order derivatives to catch inflection points before our competitors," he explained. "The jerks, that's what we call them, not the competitors, give us about a 200-millisecond edge." That conversation sparked a question I’ve been stewing on. If the analytical tools that quants use to predict asset’s behavior work so well, why aren’t we applying these same sophisticated methods to biological signals? After all, a muscle oximeter or continuous glucose monitor generates time-series data that is structurally similar to price movements. Both represent complex, multi-variable systems with emergent properties, feedback loops, and critical transition points. For decades there has been a one-way talent flow — scientists trained in computational biology, bioinformatics, and biomedical engineering migrate to financial institutions where their skills command premium compensation. This migration makes perfect sense— the mathematical toolkit for analyzing complex biological systems transfers seamlessly to market analysis, often with fewer regulatory hurdles and greater financial rewards. Yet rarely do we see expertise flowing in the reverse direction. The sophisticated analytical frameworks developed and refined through billions of dollars of financial market investments seldom find their way back to biomedical applications. This intellectual asymmetry represents a missed opportunity. What follows is a proposition that may seem initially seem unorthodox, which is that biosensor technology stands to benefit enormously from the analytical frameworks developed for derivatives trading. By viewing physiological parameters as "underlying assets" whose behavior can be analyzed not just through absolute values but through various derivatives, we can unlock previously invisible insights into human physiology and pathophysiology. The patterns are there in our data; we simply need more sophisticated lenses through which to view them. #compbio #biotech #wearables #systemsbio #quantitativefinace #datascience

BLUsH: bioluminescence imaging using hemodynamics A groundbreaking advancement in neuroimaging has been achieved with the development of BLUSH, a technique that translates bioluminescence into MRI-detectable hemodynamic signals. This novel method overcomes the inherent limitations of traditional optical imaging in deep tissues, offering unprecedented spatial resolution and depth penetration for in vivo studies. By converting photon emission into localized vascular responses, BLUSH enables real-time visualization of biological processes with applications spanning from neural circuit mapping to tumor tracking. This technology holds immense potential to accelerate neuroscience research and clinical translation. BLUSH, with its ability to convert bioluminescence into MRI-detectable signals, opens up a vast array of potential applications in biomedical research: 1. Tracking Cellular Dynamics in Real-Time - Cancer research: Monitoring tumor growth, metastasis, and response to therapy. - Immunology: Studying immune cell trafficking and response to pathogens or inflammation. - Stem cell research: Tracking cell differentiation and migration in vivo. 2. Neurological Studies - Neural circuit mapping: Visualizing neural activity patterns in real-time. - Neurodegenerative diseases: Monitoring disease progression and therapeutic efficacy. - Brain tumors: Tracking tumor growth and response to treatment. 3. Drug Delivery and Pharmacokinetics - Monitoring drug distribution: Tracking the movement of drug-carrying nanoparticles or cells. - Assessing drug efficacy: Evaluating the impact of therapeutics on target tissues. 4. Developmental Biology - Embryonic development: Studying cell fate determination and organogenesis. - Regenerative medicine: Monitoring tissue regeneration and repair. Technical Challenges and Future Directions While BLUSH represents a significant advancement, there are still technical challenges to overcome: - Uniform photosensitization: Achieving consistent bPAC expression in blood vessels across different tissue types remains a challenge. - Spatial resolution: Further improvements in MRI resolution and image processing techniques are needed to enhance spatial accuracy. - Quantitative analysis: Developing quantitative methods to correlate BLUSH signals with bioluminescence intensity is essential for accurate data interpretation. Future research should focus on addressing these challenges, as well as exploring the potential of BLUSH in combination with other imaging modalities for multimodal analysis. By overcoming these limitations, BLUSH has the potential to revolutionize biomedical research and drug development. #neuroscience #biomedicalengineering #imaging #bioluminescence #MRI #research #neuroscience

Which approach works better for heart rate estimation? • A deep learning model trained specifically to extract heart rate from PPG (photoplethysmogram) signals? • Or a deep learning model trained to extract both heart rate and respiration rate concurrently from the same dataset? Our study shows that the second approach performs better. Interestingly, learning from one parameter extraction can actually improve the model’s performance in extracting the other. In our paper recently published in Computers in Biology and Medicine (Elsevier), we developed multi-task learning approaches that leverage shared characteristics across PPG-related tasks to enhance the performance of PPG applications. Find the paper here: https://lnkd.in/gX5J-5Ma Why does this matter? PPG is a rich signal containing information from multiple vital signs. Leveraging this information can help deep learning models perform better for a specific task. From a signal processing perspective, the information stored in the respiration rate range might not overlap with the heart rate information. However, the shared characteristics can still help deep learning models improve performance. Our contribution in this paper: We develop MLT models for two PPG applications: 1) Heart rate and respiration rate extraction 2) Heart rate and heart rate variability quality assessment The models were evaluated using a PPG dataset collected from 46 subjects during their daily routines via smartwatches. Our results showed that the proposed multi-task learning methods outperform baseline single-task models: achieving higher accuracy in signal quality assessment tasks and lower error rates in both heart rate and respiration rate estimation. This research highlights the potential of multi-task learning in improving physiological signal analysis and how it contributes to wearable health technology. Special thanks to Mohammad Feli, Kianoosh Kazemi, Pasi Liljeberg, and Amir M. Rahmani #MultiTaskLearning #PPG #HeartRate #RespirationRate #SignalProcessing #DeepLearning

Glad to share with the community our latest scholarly contribution entitled “A Robust Feature Adaptation Approach against Variation of Muscle Contraction Forces for Myoelectric Pattern Recognition-Based Gesture Characterization,” published in the Biomedical Signal Processing and Control, Elsevier. Briefly: This study developed a novel feature adaptation scheme that partially leverages the non-Euclidean space concept based on the Riemann manifold to address the challenging issues of muscle contraction force variations (MCFV) in myoelectric pattern recognition systems. The scheme uses symmetric positive definite (SPD) matrices as features, minimizes force level discrepancies by projecting SPDs to a Riemann mean, and enhances robustness against MCFV by standardizing feature distribution. Electromyogram data of wrist and finger movements, obtained from in-house and public databases of amputees across three force levels was used to validate the performance of the scheme compared to state-of-the-art methods. The results of the evaluation revealed that the proposed method significantly addressed the issue of MCFV, with an improvement in movement decoding accuracy by more than 15.02% and an F1-score increase of 16.50% compared to other state-of-the-art techniques. Additional investigation into the suitable force level for training showed that the moderate force level would yield optimal performance compared to low or high force levels in the presence of MCFV. The study's findings revealed that the suggested control scheme could adapt to MCFV, improving the overall robustness of myoelectric systems in both commercial and clinical applications. Interestingly, the full-length article is available for 50 days of free access. Kindly share the download link with interested persons in your network (s). https://lnkd.in/eZ3Pnr5Y Big thanks to all the collaborators on this project!

#EPeeps Another example, how #AI upgrades photoplethysmography #PPG to the widest available tool for #AFib detection… Background: #PPG establishes as a widely available source for accurate #AFib detection. However: the computational requirements for analyzing raw PPG waveforms can be significant 👉🏻 PPG waveform analysis requires good internet 🛜 connectivity… The idea💡 The analysis of PPG-derived peak-to-peak intervals may offer a more feasible solution for smartphone deployment. The question: ❓Does waveform matter⁉️ 👉🏻We developed specialized neural networks for raw waveform and peak-to-peak interval analyses Training: 7,704 PPGs from #TeleCheckAF Validation: 48,912 PPGs from #VIRTUAL_SAFARI Results: Metrics were comparable between the interval and waveform models ☑️ Interval vs. waveform: - sensitivity 91.7 % vs. 81.9 % (p=0.4) - PPV 80.5 % vs. 84.5 % (p=0.3) - F1 score 85.6 % vs. 81.3 % (p=0.5) But ⚠️ With 1.6 million trainable parameters, the waveform model was more than 100 times as complex as the interval model (15,513 parameters) and required 19 times more computational power. 💪🏻 Conclusion: - PPG-derived peak-to-peak intervals and PPG waveforms were equivalent as input signals to neural networks in terms of accurate AF detection. - The reduced computational requirements of the interval model make it a more suitable option for deployment on digital end-user devices such as smartphones. 🚨Now online Computer Methods and Programs in Biomedicine @ElsevierConnect 📖 https://lnkd.in/envmxxaz Jonas Isaksen Jørgen K. Kanters Malene Nørregaard Nielsen Astrid Hermans Konstanze Betz Thomas Jespersen Kevin Vernooy Maastricht UMC+ Heart+Vascular Center/Maastricht UMC+ Department of Biomedical Sciences Københavns Universitet - University of Copenhagen Emma Svennberg Prof. Dr. David Duncker Fleur Tjong Sanjiv Narayan Tina Baykaner Natalia Trayanova Jordi H. Stefan Holzer David Albert Mintu Turakhia MD MS Lars Grieten Christian-H. Heeger Martin Manninger

🌟 Adaptive Filters: A Fascinating Application of Statistical Signal Processing 🌟 Adaptive filters are one of the most exciting applications of statistical and adaptive signal processing. Unlike normal FIR filters 🎛️—designed with fixed coefficients to remove specific unwanted frequencies like noise or interference—adaptive filters adapt dynamically to unknown signal characteristics while meeting specific performance metrics 📊. This is where statistical signal processing shines. ✨ 📌 One example is the Wiener filter (LMMSE), which I’ve discussed previously. Another is its gradient descent version, offering a recursive implementation of the Wiener filter. However, these optimal LMMSE estimators require knowledge of input and desired signal statistics, which isn’t always feasible in real-world scenarios. 🔑 Enter stochastic gradient descent (SGD): • A practical alternative that estimates the gradient, with the LMS adaptive filter algorithm being the simplest example. • The LMS algorithm provides an unbiased gradient estimator, allowing the solution to fluctuate around the optimal MMSE solution (hence the name least mean square). ✨ Key benefits of LMS: • 📡 Adaptability: Tracks variations in the channel and adjusts to changing conditions (e.g., in wireless communication). • ⚙️ Simplicity: A lightweight yet effective estimator. For scenarios involving sudden input signal variations, the normalized LMS (NLMS) is a robust choice, as it’s less sensitive to input signal statistics 📉. Simon Haykin’s book Adaptive Filter Theory presents an elegant proof where NLMS minimizes changes in filter weights, enabling superior performance under dynamic conditions. 🌍 Applications of adaptive filters span: • 📶 Telecommunications • ❤️ Biomedical signal processing • 🎧 Audio processing & echo cancellation • 🚫 Noise/interference cancellation • 🏗️ System identification The beauty of adaptive filters lies in their versatility: by redefining input, output, and desired signals, you can solve a variety of problems using the same foundational principles. This makes it one of the most fascinating topics in signal processing in my opinion!