An efficient ANN SoC for detecting Alzheimer's disease based on recurrent computing.

Area of Science:

• Biomedical Engineering
• Computer Science
• Neuroscience

Background:

• Alzheimer's Disease (AD) detection faces challenges with high misdiagnosis rates and costly hardware.
• Existing neural network methods for AD detection often lack high-performance, on-chip implementable solutions.
• Complex neural networks hinder clinical applications and hardware integration.

Purpose of the Study:

• To develop a novel, efficient Alzheimer's Disease detection framework.
• To address the limitations of traditional AD detection techniques, including high costs and misdiagnosis rates.
• To create a high-performance, low-cost AD detection chip utilizing Artificial Neural Networks (ANNs).

Main Methods:

• A recurrent computational strategy was employed, embedding an ANN within a System on Chip (SoC) for Electroencephalogram (EEG) analysis.
• Hardware encoding of preprocessed EEG data using reduced IEEE754 single-precision encoding minimized memory footprint.
• Data remapping, hierarchical, and Processing Element (PE) reuse technologies optimized ANN computations and memory access.

Main Results:

• The optimized SoC demonstrated a 70% reduction in area and a 50% reduction in power consumption compared to traditional designs.
• The proposed detection model exhibited 3 to 4 times faster training speeds and incurred less overhead than traditional models.
• Achieved a high accuracy rate of 98.53% for Alzheimer's Disease detection.

Conclusions:

• The novel ANN-based SoC framework offers an efficient and cost-effective solution for Alzheimer's Disease detection.
• This approach overcomes the limitations of complex neural networks for on-chip implementation and clinical use.
• The developed system provides a promising advancement in early and accurate Alzheimer's Disease diagnosis.