Predicting Delayed Neurocognitive Recovery After Non-cardiac Surgery Using Resting-State Brain Network Patterns...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Delayed neurocognitive recovery (DNR) is a common postoperative complication.
• Objective methods for identifying high-risk individuals are currently lacking.
• Resting-state functional magnetic resonance imaging (rs-fMRI) can reveal brain network alterations.

Purpose of the Study:

• To develop a machine learning model for predicting DNR.
• To utilize pre-operative resting-state functional connectivity (FC) features.
• To identify cognitive-related brain network features associated with DNR.

Main Methods:

• rs-fMRI data from 74 elderly patients undergoing non-cardiac surgery were analyzed.
• Seed-based whole-brain FC was computed for default mode network (DMN), limbic network, salience network (SN), and central executive network (CEN).
• Machine learning models (SVM, decision tree, random forest) were trained and validated to predict DNR.

Main Results:

• The DNR group showed aberrant FC in specific regions within DMN, SN, CEN, and limbic networks.
• A random forest model using DMN and CEN FC features demonstrated superior predictive performance.
• The best model achieved an AUC of 0.958, with high accuracy (0.935) and recall (0.900) on the test set.

Conclusions:

• Machine learning, particularly the random forest model, can effectively predict DNR using pre-operative brain network features.
• FC patterns in the DMN and CEN are significant predictors of DNR after non-cardiac surgery.
• This predictive model offers a potential tool for early identification and prevention of DNR.