Optimized Deconvolutional Algorithm-based CT Perfusion Imaging in Diagnosis of Acute Cerebral Infarction

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Acute cerebral infarction (ACI) diagnosis relies heavily on accurate interpretation of CT perfusion imaging.
• Existing deconvolution algorithms and image processing techniques present limitations in precision and efficiency.
• Convolutional Neural Networks (CNNs) offer potential for enhancing medical image analysis.

Purpose of the Study:

• To optimize a CNN algorithm for deconvolution in CT perfusion imaging of ACI.
• To evaluate the efficacy of an enhanced RIU-Net model with SE module for image segmentation and feature extraction.
• To compare the denoising performance of the proposed model against established algorithms.

Main Methods:

• An optimized CNN, RIU-Net with an SE module, was developed for CT image segmentation.
• The BM3D, Dn CNN, and Cascaded CNN algorithms were compared for image denoising.
• A retrospective analysis of 80 ACI patients was conducted, comparing the optimized algorithm (observation group) with ordinary methods (control group).

Main Results:

• The SE module enhanced RIU-Net improved key feature utilization in CT images.
• The observation group showed superior performance with 98.7% specificity and 93.7% accuracy, detecting (1.6 ± 0.2) lesions.
• The control group achieved 93.2% specificity and 87.6% accuracy, detecting (1.3 ± 0.4) lesions (P < 0.05).

Conclusions:

• The optimized CNN model demonstrates significant capabilities in image denoising and segmentation for ACI diagnosis.
• The enhanced RIU-Net model accurately extracts critical information, improving diagnostic precision in CT perfusion imaging.
• The proposed algorithm is clinically valuable for the accurate and efficient diagnosis of acute cerebral infarction.