A fully convolutional network (FCN) based automated ischemic stroke segment method using chemical exchange saturation...

Area of Science:

• Biomedical Imaging
• Artificial Intelligence
• Neurology

Background:

• Chemical Exchange Saturation Transfer (CEST) MRI shows promise for detecting ischemic stroke by sensing pH changes.
• Accurate segmentation of pH-altered regions in CEST MRI is challenging due to complex water signal variations.
• Manual lesion analysis is time-consuming and hinders timely therapeutic interventions.

Purpose of the Study:

• To develop an automated method for segmenting ischemic regions in CEST MRI.
• To investigate a novel segmentation framework utilizing a fully convolutional neural network.

Main Methods:

• A 1D fully convolutional neural network with bottleneck structures was developed.
• Z-spectra from rat models were manually labeled and used to train and test the segmentation model.
• Grad-CAM was employed to generate localization maps for interpreting segmentation results.

Main Results:

• The proposed network achieved high segmentation performance (SPE, SEN, ACC, DSC), outperforming conventional methods in difficult cases.
• The model demonstrated robustness to input variations, especially with data augmentation.
• Grad-CAM maps provided interpretable insights into tissue changes and correlated with quantitative methods.

Conclusions:

• The developed method effectively segments ischemic regions from CEST images.
• Grad-CAM integration offers interpretative capabilities, highlighting the method's clinical potential for stroke management.