Multichannel Neuromodulation System with Analog Front-End and Neural Classifier
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing closed-loop neuromodulation devices, such as system-on-chip (SoC) devices with integrated machine learning, are limited by low channel count and poor generalizability, which restricts their therapeutic efficacy and application scope.
Innovation Solution
The proposed solution involves an analog front-end device with a switch matrix and dual multiplexing choppers for dynamic channel selection, along with a filter and feature extraction engine that includes a time-division multiplexed (TDM) finite impulse response (FIR) filter and a feature extraction engine (FEE) capable of extracting phase synchrony, frequency, and temporal features. These features are then processed by a tree-structured hierarchical neural network classifier for disease symptom detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If existing SoC devices with integrated machine learning are used, then device complexity is reduced, but channel count is limited to 8-32 channels
Solution Approach 1:
The system segments the processing architecture into distinct functional blocks: a high-channel-count analog front-end (supporting 128-256 channels) and a separate machine learning processing unit. This segmentation allows the front-end to handle more channels while the ML unit processes features independently, resolving the contradiction between device complexity and channel count.
Solution Approach 2:
The patent transitions from processing raw multi-channel signals to extracting and processing neural biomarkers as features. This dimensional transformation from time-domain signals to feature-space representations enables handling of higher channel counts by reducing the data dimensionality that requires complex processing.
2Adaptability or versatility
If existing SoC devices with integrated machine learning are used, then integration is improved, but generalizability is poor
Solution Approach 1:
The system extracts multiple types of neural biomarkers (spectral, temporal, spatial, and cross-frequency coupling features) from the same multi-channel neural signals, enabling the device to address various neurological conditions and applications. This multi-functionality approach improves generalizability without proportionally increasing integration complexity.
3Measurement precision
If high channel count is implemented, then measurement precision is improved, but power consumption increases
Solution Approach 1:
The system extracts and processes only the most relevant neural biomarkers and features from the multi-channel signals using machine learning algorithms. By taking out and processing only essential features rather than all raw signal data, the system maintains measurement precision while significantly reducing the computational power required.
Solution Approach 2:
The system implements partial processing by focusing computational resources on extracting key neural biomarkers rather than processing all signal parameters in full detail. This partial action approach maintains sufficient measurement precision for clinical applications while reducing overall power consumption.
4Adaptability or versatility
If high channel count is implemented, then application scope is expanded, but chip surface area increases
Solution Approach 1:
The patent merges multiple signal processing functions (filtering, feature extraction, biomarker computation) into a integrated machine learning processing unit that handles data from all channels. This consolidation allows high channel count implementation without proportionally increasing chip surface area, as shared processing resources serve multiple functions.
Data Source
AI summary
A closed-loop neuromodulation system, including an electrode array that is implantable to a brain of a subject, analog front-end device (AFD) for selectively selecting and reading a plurality of channels from electrode array, a finite impulse response (FIR) filter for selectively filtering signals from the AFD, a feature extraction engine (FEE) operatively connected to the FIR filter, configured to selectively extract features from signals provided by the FIR filter, a tree-structured hierarchical neural network classifier for detecting disease symptoms, and a multi-channel stimulator having high-voltage (HV) drivers operatively connectable to the electrode array.