NeuroSketch combines three AI-powered diagnostic tests — spiral drawing, voice analysis, and motor tracking — to detect Parkinson's disease before it's too late.
Parkinson's affects over 7 lakh Indians, yet the country has fewer than 1,200 neurologists. Early-onset cases, misdiagnosis, and rural healthcare gaps are leaving millions without answers.
A significant number of Indian patients develop PD symptoms between ages 22 and 49 — far earlier than commonly assumed.
Symptoms like tremor and rigidity are often dismissed as normal aging, causing critical delays in diagnosis and treatment.
Diagnosis relies entirely on clinical evaluation — subjective, inconsistent, and prone to misdiagnosis without specialist access.
Inadequate healthcare facilities and financial constraints prevent timely access to specialist care, especially in rural India.
NeuroSketch is an AI-powered web application that runs three parallel diagnostic tests using nothing more than a browser and a camera — no specialist required.
Patients draw a spiral on screen. The AI analyzes kinematic features — stroke velocity, acceleration, and radial velocity — to detect the micro-tremors invisible to the naked eye.
A 5-second voice recording is analyzed for jitter, shimmer, and frequency variations. Parkinson's subtly alters vocal quality long before other symptoms appear.
Computer vision tracks finger-tapping and hand movements via webcam. Speed, amplitude, frequency, and period are extracted to detect motor impairments in real time.
Patient draws a spiral on a digital canvas — a standard clinical Archimedes spiral test, now digitized.
The drawing is converted to binary (black/white) and normalized to remove scale/position bias before feature extraction.
Kinematic features are extracted: stroke velocity, acceleration, radial velocity, and geometric regularity measures.
Logistic Regression, SVC, KNN, and Random Forest classifiers vote on the binary Healthy/Parkinson's classification.
Patient records a sustained vowel sound ("ahhh"). The audio is captured directly in the browser.
Audio waveform is converted to spectrogram. MDVP (Multi-Dimensional Voice Program) features are extracted.
Jitter (%), Shimmer, HNR, RPDE, DFA, PPE and more — capturing subtle vocal cord irregularities.
Support Vector Machine trained on the Oxford Parkinson's Dataset classifies with ~87% accuracy on test data.
Patient performs finger-tapping or hand movement tasks in front of the webcam for ~60 seconds.
MediaPipe Hands tracks 21 3D landmarks per hand frame-by-frame, computing inter-finger distances.
Speed, amplitude, frequency, and period of tapping are calculated from the distance-over-time signal.
A LightGBM regressor predicts PD symptom severity score based on extracted video motion features.
Our diagnostic approach is grounded in the same 15 clinical assessment methods used by neurologists — just made accessible via AI.
A structured plan to bring all three diagnostic modules together into a polished, unified MVP.
Streamlit interface + model integration
CV tremor detection & data collection
Feature extraction, model training, frontend
Combine all tools, internal testing & evaluation
Documentation, demo video, submission
We're a team of young builders with strong foundations — but we're actively seeking mentorship to push NeuroSketch to the next level.
Need guidance on advanced model optimization and hyperparameter tuning for our classifiers
Limited experience building video-based motion detection models at scale
Early-stage user interface is functional but not yet user-friendly for elderly patients
Need help scaling our prototype tools into a polished, deployable MVP
Technical Mentorship: Support on CV model architecture, training pipelines, and feature engineering for medical data
Dataset Guidance: Help sourcing and handling appropriate Parkinson's motor/video datasets
UI/UX Mentorship: Creating a clean, intuitive diagnostic flow with accessibility best practices for elderly users
Clinical Validation: Connecting with neurologists to validate our approach and refine diagnostic accuracy
Real feedback from evaluators, educators, and healthcare professionals who've seen NeuroSketch in action.
Explore the live app, watch our demo, or follow our journey as we build toward a world where Parkinson's is caught early — for everyone.