Radiomics-based machine learning approach in differentiating fibro-adipose vascular anomaly from venous malformation

Area of Science:

• Radiology
• Medical Imaging
• Machine Learning

Background:

• Fibro-adipose vascular anomaly (FAVA) is a complex vascular malformation.
• FAVA shares imaging and clinical features with other vascular malformations, leading to misdiagnosis.
• Accurate differentiation is crucial for proper patient management.

Purpose of the Study:

• To develop a radiomics-based machine learning model.
• The model aims to assist radiologists in distinguishing FAVA from common venous malformations (VMs).

Main Methods:

• Retrospective analysis of 178 patients with vascular malformations (41 FAVA, 137 VM).
• Radiomics features extracted from MRI (T1-weighted and fat-saturated T2-weighted images).
• Feature selection using LASSO and Boruta methods; eight logistic regression models developed and evaluated using cross-validation.

Main Results:

• Multi-sequence models demonstrated strong performance.
• LASSO-based model: AUC 97%±3.8, sensitivity 94%±12.4, specificity 89%±9.0.
• Boruta-based model: AUC 97%±3.7, sensitivity 95%±10.5, specificity 87%±9.0.

Conclusions:

• Radiomics-based machine learning models show promise.
• These models can aid in differentiating FAVA from common VMs.
• Potential for improved diagnostic accuracy and patient care.