Automatic Assessment of ASPECTS Using Diffusion-Weighted Imaging in Acute Ischemic Stroke Using Recurrent Residual...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Early detection and quantification of acute ischemic lesions are critical for effective stroke management.
• Diffusion-weighted imaging (DWI) is essential for assessing acute stroke severity.
• The Alberta Stroke Program Early Computed Tomographic Score (ASPECTS) is a key metric for evaluating stroke extent.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for automatic binary classification of DWI-ASPECTS in acute stroke patients.
• To compare the performance of a recurrent residual convolutional neural network (RRCNN) against other deep learning models.
• To assess the potential of the algorithm as an ancillary tool for clinical decision-making.

Main Methods:

• Development of a recurrent residual convolutional neural network (RRCNN) classifier.
• Training and validation using 390 DWI datasets from acute anterior circulation stroke patients.
• Comparison of RRCNN performance against VGG16, Inception V3, and 3D convolutional neural network (3DCNN) models.

Main Results:

• The RRCNN model achieved an accuracy of 87.3%, an AUC of 0.941, and an F1-score of 0.888 for classifying low (1-6) versus high (7-10) DWI-ASPECTS.
• The proposed RRCNN model outperformed the pre-trained VGG16, Inception V3, and 3DCNN models.
• The algorithm demonstrated robust performance in differentiating between low and high DWI-ASPECTS groups.

Conclusions:

• The developed deep learning algorithm provides a rapid and accurate assessment of DWI-ASPECTS.
• This AI tool can serve as a valuable ancillary resource for physicians managing acute stroke.
• The findings support the integration of AI in improving stroke diagnosis and treatment decisions.