Machine learning models for predicting transstenotic pressure gradient based on computed tomography angiography...

Area of Science:

• Medical Imaging
• Machine Learning
• Vascular Physiology

Background:

• Transstenotic pressure gradient (TPG) is crucial for diagnosing and managing conditions like idiopathic intracranial hypertension and pulsatile tinnitus.
• Current TPG assessment methods, such as digital subtraction angiography (DSA), are invasive, complex, and expensive.
• There is a need for efficient, noninvasive methods to evaluate TPG.

Purpose of the Study:

• To develop and validate a machine learning model for predicting transverse sinus TPG.
• To utilize features extracted from computed tomography angiography (CTA) for TPG prediction.
• To establish a noninvasive tool for TPG assessment.

Main Methods:

• 139 patients with DSA for TPG measurement were included.
• Six CTA-derived features were extracted: residual area ratio, stenosis length, stenosis type, Labbé vein location, drainage dominance, and contralateral stenosis.
• Six machine learning algorithms were applied to predict TPG at 4 and 8 mmHg thresholds, with logistic regression showing the best performance (AUC 0.83).

Main Results:

• Logistic regression achieved an AUC of 0.83 for both 4 mmHg and 8 mmHg thresholds.
• Shapley additive explanations identified Labbé vein location as positively correlated with TPG, while residual area ratio and stenosis type were negatively correlated.
• External validation demonstrated high accuracy, reaching 0.89 for both thresholds.

Conclusions:

• The developed machine learning model effectively predicts transverse sinus TPG using CTA-derived features.
• This model offers a promising noninvasive approach for TPG assessment.
• The findings support the potential of AI in vascular diagnostics.