MUMBAI, India, June 22 -- Intellectual Property India has published a patent application (202631067380 A) filed by Narula Institute Of Technology on May 29, 2026, for “smartphone-Based Multimodal Artificial Intelligence System For Non-Invasive Parkinson’s Disease Detection And Severity Estimation”.
Inventors include Dr. Nabanita; Dr. Chandrima Chakrabarti; Mr. Arindam Das; and Arijit Singha Roy.
The application for the patent was published on June 12, 2026, under issue no. 24/2026.
Abstract: Abstract A smartphone-based multimodal artificial intelligence system and method are disclosed for non-invasive Parkinson’s Disease screening and continuous clinical severity estimation. The system captures acoustic voice recordings, facial video sequences, and 6-axis inertial motion data via a mobile device and a low-cost, wireless BLE wrist wearable built around an ESP32 microcontroller. The heterogeneous signals are routed through specialized neural encoders—comprising a Wav2Vec2 acoustic framework, a Vision Transformer face model, and a Temporal Convolutional Network with Squeeze-Excitation channel attention—to produce high- dimensional embeddings. A Pre-LayerNorm Multimodal Transformer fuses these aligned vectors using a learnable classification (CLS) token. A dynamic modality masking component alters the self-attention graph using a boolean mask matrix to replace missing inputs with zero-valued tensors, permitting accurate execution when modalities are absent. The CLS token feeds parallel output heads that concurrently compute disease probability and a continuous UPDRS-mapped severity score within a single forward pass, delivered via a real-time cloud server to a mobile client dashboard.
Disclaimer: Curated by HT Syndication.
