Deep learning-driven multi-class classification of brain strokes using computed tomography: A step towards enhanced...

Area of Science:

• Artificial Intelligence in Medical Imaging
• Neurology and Stroke Diagnostics
• Deep Learning for Healthcare

Background:

• Brain stroke diagnosis relies heavily on timely and accurate interpretation of medical imaging.
• Computed Tomography (CT) is a primary imaging modality for acute stroke assessment.
• Enhancing diagnostic accuracy and speed is crucial for effective stroke management and patient outcomes.

Purpose of the Study:

• To develop and validate deep learning models for predicting and classifying brain stroke conditions using CT imaging.
• To assess the potential of these AI models in improving diagnostic accuracy and supporting clinical decision-making.
• To evaluate model performance through rigorous internal and external validation methods.

Main Methods:

• A retrospective study utilized 8186 CT images from 250 patients (2017-2022) across two centers.
• Two deep learning models were developed using the Expanded ResNet101 framework.
• Model performance was evaluated using confusion matrices, accuracy, precision, F1 scores, and external validation by expert reviewers.

Main Results:

• Internal validation showed high performance: Model 01 (99.6% accuracy, 99.4% precision, 99.6% F1 score) and Model 02 (99.2% accuracy, 98.8% precision, 99.1% F1 score).
• External validation yielded accuracies of 78.6% for Model 01 and 60.2% for Model 02.
• Statistical significance was confirmed for external validation results (P < 0.001).

Conclusions:

• Deep learning models demonstrate significant potential in accurately predicting brain stroke outcomes from CT images.
• The developed models achieved high internal performance metrics, indicating strong predictive capabilities.
• Further development with larger, diverse datasets could enhance these models for robust clinical support in stroke prognosis and decision-making.