Coordinate-aware three-dimensional neural network for lower extremity arterial stenosis classification in CT...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease

Background:

• Lower Extremity Computed Tomography Angiography (CTA) is a key non-invasive method for diagnosing lower extremity artery disease (LEAD).
• Accurate assessment of arterial stenosis degree is crucial for effective patient management.
• Existing methods may require significant radiologist interpretation time.

Purpose of the Study:

• To develop and evaluate an automated classification model for assessing lower extremity arterial stenosis.
• To utilize a coordinate-aware 3D deep neural network for improved diagnostic accuracy.
• To provide a tool that assists radiologists in interpreting lower extremity CTA scans.

Main Methods:

• A retrospective study of 277 patients who underwent lower extremity CTA.
• Annotation of arterial segments by radiologists and segmentation of 12,450 3D patches.
• Implementation of a Coordinate-Aware Three-Dimensional Neural Network for stenosis classification.
• Performance evaluation using accuracy, sensitivity, specificity, F1 score, and ROC curves.

Main Results:

• The model achieved high accuracy (93.08% above-knee, 91.70% below-knee) and F1 scores (91.96% above-knee, 89.67% below-knee).
• Area Under the ROC Curve (AUC) reached 99.15% for above-knee and 98.2% for below-knee arteries.
• The proposed model outperformed 3D and 2D baseline models by a significant margin in accuracy.

Conclusions:

• A deep learning model was successfully developed for automated evaluation of lower extremity arterial stenosis on CTA.
• The coordinate-aware 3D neural network shows significant promise as an assistive tool for radiologists.
• This AI-driven approach can enhance the efficiency and accuracy of LEAD diagnosis.