NeuroPulse AI

NeuroPulse AI uses interactive cognitive assessments and machine learning to track neurological biomarkers, providing objective, real-time insights for concussion recovery and mental wellness.

Comment

Inspiration

Millions of individuals suffer from concussions, traumatic brain injuries (TBIs), and cognitive fatigue every year—from student-athletes to military veterans. Currently, post-concussion recovery evaluations rely heavily on subjective self-reporting ("How does your head feel today?"), which leads to premature return to sports or work and increases the risk of secondary brain injury.

We were inspired to build NeuroPulse AI to bridge the gap between subjective symptoms and objective clinical biomarkers using real-time machine learning, interactive neurological tests, and accessible web technologies.

What it does

NeuroPulse AI is an intelligent, real-time cognitive biomarker tracking and mental wellness platform. It provides:

  1. Real-Time Reaction & Motor Stability Assessment: Measures processing latency in milliseconds ($ms$) and motor micro-jitter ($\sigma_{RT}$) across trials.
  2. Stroop Interference Cognitive Challenge: Evaluates selective attention, visual processing error rates ($\text{Error Rate} = \frac{\text{Incorrect}}{\text{Total}} \times 100\%$), and executive response inhibition.
  3. Machine Learning Recovery Evaluator: Evaluates multi-modal biomarker inputs using predictive ML models to classify recovery state (from Optimal to High Deficit).
  4. Mental Health AI Journal: Uses Natural Language Processing (NLP) to analyze daily sentiment, emotional valence, and psychological stress indices.
  5. Live Model Telemetry: Displays transparent feature weightings, normalization parameters, and diagnostic telemetry for clinical visibility.

How we built it

We engineered NeuroPulse AI using a modern Python data science and web stack:

  • Frontend & Dashboard: Built with Streamlit and styled with modern CSS glassmorphism UI components.
  • Data Visualizations: Built interactive time-series plots and feature importance graphs using Plotly.
  • Machine Learning Engine: Built using Scikit-Learn pipelines. Features are standard-scaled via Z-score normalization: $$Z = \frac{X - \mu}{\sigma}$$ Predictions are computed through ensemble classifiers evaluated on feature vectors combining reaction speed, motor variance, error rates, sleep hours, and symptom severity: $$\vec{X} = [\text{Mean RT}, \text{RT Std}, \text{Error Rate}, \text{Sleep}, \text{Symptom Score}]$$
  • NLP Sentiment Engine: Formulated custom lexicon-driven sentiment scoring: $$\text{Sentiment Score} = \frac{N_{\text{positive}} - N_{\text{anxiety}}}{\max(N_{\text{total}}, 1)}$$

Challenges we ran into

  1. Sub-millisecond Timing Accuracy in Web Interfaces: Measuring precise reaction time latency over web browsers introduces event loop delays. We designed state-driven timer callbacks to capture accurate timestamps.
  2. Multi-Modal Data Integration: Normalizing heterogeneous data streams—combining quantitative reaction times ($ms$) with qualitative self-reported symptoms ($0-10$ scale)—required mathematical scaling to prevent feature dominance.
  3. Balancing Clinical Accuracy with UX: Designing a high-tech UI that remains accessible and low-stimulation for patients experiencing light sensitivity or brain fatigue.

Accomplishments that we're proud of

  • Successfully built an end-to-end working AI platform in less than 24 hours.
  • Created interactive, real-time cognitive tests (Reaction latency & Stroop challenge) directly inside the web environment.
  • Developed an ensemble ML pipeline with dynamic clinical confidence breakdowns and feature importance visualizations.
  • Integrated an NLP sentiment engine that translates unstructured daily journaling into quantitative psychological stress metrics.

What we learned

  • How to extract meaningful neurological biomarkers (reaction speed, motor jitter, attention error rate) from simple web interactions.
  • The importance of model transparency and explainability in digital healthcare tools.
  • Effective strategies for combining quantitative physical metrics with qualitative mental health sentiment data.

What's next for NeuroPulse AI

  • Computer Vision Eye Tracking: Incorporating webcam-based saccadic eye movement tracking to assess pupil response and visual fixation stability.
  • Wearable Device Integration: Syncing with smartwatches (Apple Health, Fitbit) to continuously pull HRV (heart rate variability) and sleep telemetry.
  • Longitudinal Clinical Dashboards: Exporting HIPAA-compliant PDF reports for neurologists, physical therapists, and athletic trainers.

Built With

Share this project:

Updates