Large vessel occlusion detection by non-contrast CT using artificial ıntelligence.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Computer vision models are increasingly used for diagnosing disorders via CT and MR imaging.
• Acute ischemic stroke diagnosis requires timely detection of brain vessel occlusions.

Purpose of the Study:

• To develop and evaluate a deep feature engineering model for detecting large and small brain vessel occlusions in acute ischemic stroke.
• To improve early diagnosis and recanalization chances for stroke patients.

Main Methods:

• Utilized a dataset of 324 patient CT images, divided into horizontal and vertical patches.
• Extracted deep features using a pretrained AlexNet model (fc6 and fc7 layers).
• Applied iterative neighborhood component analysis (INCA) for feature selection, identifying 43 optimal features, followed by kNN classification with tenfold cross-validation.

Main Results:

• The AlexNet-based deep feature engineering model achieved 100% classification accuracy.
• The model successfully differentiated between large and small brain vessel occlusions.

Conclusions:

• The proposed model demonstrates high efficacy in detecting brain vessel occlusions in non-contrast CT images.
• This approach can significantly assist neurology experts in identifying early recanalization opportunities in acute ischemic stroke.