Using Deep-Learning-Based Artificial Intelligence Technique to Automatically Evaluate the Collateral Status of...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Collateral status is a key predictor for outcomes in acute ischemic stroke (AIS) with large vessel occlusion (LVO).
• Multiphase computed-tomography angiography (mCTA) is crucial for assessing collateral status but visual evaluation is labor-intensive.
• Developing automated methods for collateral status assessment is essential for efficient clinical workflows.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) prediction model for automatic collateral status evaluation using mCTA.
• To investigate the feasibility of using convolutional neural network (CNN) techniques for this task.

Main Methods:

• A retrospective study of 82 AIS patients who underwent endovascular thrombectomy was conducted.
• mCTA images at basal ganglion and supraganglion levels were used to train a CNN model.
• The model was trained on 57 cases and validated on 25 cases, with collateral status determined by visual assessment.

Main Results:

• The AI model achieved high accuracy (0.999 ± 0.015) on the training set.
• The model demonstrated a validation accuracy of 0.746 ± 0.008, with an area under the ROC curve of 0.7.
• These results indicate the potential for AI in automating collateral status assessment.

Conclusions:

• The AI model derived from mCTA images shows feasibility for automatic collateral status evaluation.
• This AI approach could streamline the assessment process in clinical practice.
• Further validation and refinement of the AI model are warranted.