Evaluation of brain nerve function in ICU patients with Delirium by deep learning algorithm-based resting state MRI

Area of Science:

• Neuroimaging
• Medical Diagnostics
• Artificial Intelligence in Medicine

Background:

• Delirium is a common condition in intensive care units (ICUs) associated with significant morbidity and mortality.
• Cranial nerve dysfunction is a recognized feature of delirium, but its neurobiological underpinnings require further investigation.
• Current neuroimaging techniques may have limitations in fully characterizing the complex brain changes associated with delirium.

Purpose of the Study:

• To evaluate the utility of resting-state magnetic resonance imaging (MRI) using the Brain Extraction Tool (BET) algorithm for assessing cranial nerve function in delirium patients.
• To compare the performance of the BET algorithm with a Convolutional Neural Network (CNN) algorithm in MRI analysis for delirium.
• To explore differences in brain nerve function between patients with delirium and healthy individuals.

Main Methods:

• 100 patients with delirium and 20 healthy volunteers underwent resting-state MRI.
• Brain images were analyzed using the Brain Extraction Tool (BET) algorithm.
• A Convolutional Neural Network (CNN) algorithm was employed for comparative analysis.
• Quantitative metrics including Root Mean Square Error, High Frequency Error Norm, and Structural Similarity were assessed.
• Regional Homogeneity (ReHo), Fractional Anisotropy (FA), and Mean Diffusivity (MD) values were analyzed to evaluate brain function and structure.

Main Results:

• The BET algorithm demonstrated significantly higher Root Mean Square Error (70.4%), High Frequency Error Norm (71.5%), and Structural Similarity (0.92) compared to the CNN algorithm (P < 0.05).
• Delirium patients exhibited significantly higher ReHo values in the pons, right hippocampus, left cerebellum, midbrain, and basal ganglia compared to controls.
• Significant decreases in ReHo were observed in the frontal gyrus, middle frontal gyrus, left inferior frontal gyrus, parietal lobe, and temporal lobe.
• Fractional Anisotropy (FA) scores decreased significantly in the left cerebellum, frontal lobe, left temporal lobe, corpus callosum, and left hippocampus.
• Mean Diffusivity (MD) values significantly increased in the medial frontal lobe, right superior temporal gyrus, anterior cingulate gyrus, bilateral insula, and left caudate nucleus (P < 0.05).

Conclusions:

• Deep learning-based MRI significantly enhances image quality, proving valuable for assessing brain nerve function in delirium patients.
• Abnormalities in brain structure and function identified through advanced MRI techniques can aid in the diagnosis of delirium.
• Resting-state MRI with advanced algorithms offers a promising approach for understanding the neurobiology of delirium and improving diagnostic accuracy.