Machine Learning Based on Multiparametric Features from Dual-Layer Detector Spectral CT for Identifying Ulcer-Like...

Area of Science:

• Radiology
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Aortic intramural hematoma (IMH) is a dangerous form of acute aortic syndrome (AAS).
• Ulcer-like projections (ULPs) in IMH increase the risk of dissection, aneurysm, or rupture.
• Conventional CT angiography struggles to reliably differentiate small ULPs from surrounding hematoma or plaques.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models integrating spectral CT quantitative parameters for classifying ULPs in IMH.
• To improve the diagnostic accuracy of ULP detection in IMH patients.

Main Methods:

• Retrospective analysis of 95 IMH patients.
• Extraction of spectral CT quantitative features (VMI, IC, NIC, Zeff, spectral curve slope).
• Feature selection using LASSO regression and construction of six ML models, including random forest.

Main Results:

• The random forest model achieved an AUC of 0.889, with 93.7% sensitivity and 82.8% accuracy in the internal test set.
• SHAP analysis identified effective atomic number (Zeff) and spectral curve slope as key predictive features.
• The models effectively utilized material composition and iodine attenuation characteristics for classification.

Conclusions:

• Multiparameter spectral CT-based ML models show exploratory internal feasibility for ROI-based ULP classification in IMH.
• This approach complements single-parameter analysis and aids in risk stratification for AAS.
• Further validation is needed for clinical application.