A cascaded convolutional neural networks for stroke detection imaging.

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Biomedical Engineering

Background:

• Electrical impedance tomography (EIT) is increasingly utilized for stroke detection.
• Improving EIT's prediction accuracy and noise resilience is crucial for clinical application.
• Solving the inverse problem in EIT is key to accurate stroke detection.

Purpose of the Study:

• To develop an advanced EIT system for stroke detection with enhanced accuracy and anti-noise capabilities.
• To address the inverse problem in EIT using a novel cascade convolutional neural network architecture.

Main Methods:

• A cascade convolutional neural network (CNN) was designed, comprising two distinct parts.
• The first part utilizes an optimized encoding-decoding network for high-resolution imaging.
• The second part employs a residual module to extract voltage information characteristics and prevent data loss.

Main Results:

• The proposed cascade CNN demonstrates superior anti-noise performance compared to existing networks.
• Physical experiments validated the algorithm's ability to approximate object locations within the field.
• The network effectively solves the inverse problem for improved EIT accuracy.

Conclusions:

• Cascade CNNs offer a significant advancement in EIT for stroke detection.
• The developed method provides improved imaging resolution and robustness against noise.
• This approach holds promise for more reliable and accurate stroke diagnosis using EIT.