A deep learning approach to predict collateral flow in stroke patients using radiomic features from perfusion images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Collateral circulation is crucial for stroke outcomes, but its assessment via manual imaging review is time-consuming and prone to bias.
• Current methods for grading collateral blood flow rely heavily on subjective interpretation by clinicians, leading to inconsistencies.
• Objective and efficient methods are needed to accurately assess collateral status in stroke patients.

Purpose of the Study:

• To develop and validate a multi-stage deep learning approach for automated collateral flow grading in stroke patients.
• To improve the speed, consistency, and objectivity of collateral circulation assessment using radiomic features from MR perfusion data.
• To compare the performance of the deep learning model against expert manual grading.

Main Methods:

• A deep learning network was trained using reinforcement learning for automated region of interest detection in 3D MR perfusion volumes.
• Radiomic features were extracted from the identified regions using local image descriptors and denoising auto-encoders.
• A convolutional neural network and other machine learning classifiers were employed to predict collateral flow grading into three severity classes.

Main Results:

• The deep learning model achieved an overall accuracy of 72% in the three-class collateral flow grading task.
• The automated approach demonstrated performance comparable to expert grading, with significantly higher inter-observer agreement (74% vs 16%).
• The AI-driven method offers a substantial speed improvement over manual visual inspection and eliminates grading bias.

Conclusions:

• The developed multi-stage deep learning approach provides an accurate, fast, and objective method for collateral flow grading in stroke patients.
• This AI tool has the potential to enhance clinical decision-making in stroke care by providing reliable collateral status assessment.
• Automating collateral grading can lead to more consistent patient management and potentially improve clinical outcomes in ischemic stroke.