MRI-Based Radiomics in Distinguishing Kaposiform Hemangioendothelioma (KHE) and Fibro-Adipose Vascular Anomaly (FAVA) in Extremities: A Preliminary Retrospective Study
jpedsurg.org shows its articles on its own site.
Read the article on jpedsurg.org ↗Article · Mar 2022 · 1 min read
In brief
In brief
Machine learning analysis of MRI scans successfully distinguishes between two rare pediatric vascular tumors—KHE and FAVA—that appear similar on conventional imaging. The radiomic model achieved 85% accuracy in testing, offering a potential diagnostic tool to guide treatment decisions for these challenging extremity lesions.
Written by the GCMD Library team from the article.
Objective
To investigate the pretreatment differentiation between Kaposiform hemangioendothelioma (KHE) and fibro-adipose vascular anomaly (FAVA) in extremities of pediatric patients. To build and validate an MRI-based radiomic model.
Method
In this retrospective study, we obtained imaging data from 43 patients. We collected and compared clinical information, sketched region of interest (ROI), and extracted radiomic features from fat-suppressed T2-weighted (T2FS) images of the two cohorts of 30 and 13 patients respectively (training versus testing cohort 7:3). To select features, we used two sample t-test and the least absolute shrinkage and selection operator (LASSO) regression. The support vector machine (SVM) classification was constructed and evaluated by receiver operating characteristic (ROC) analysis.
Results
Thirty patients with KHE and 13 patients with FAVA in the extremities were included. Most lesions demonstrated low to intermediate signal intensity on T1-weighted images and hyperintense signals on T2-weighted ones. They also showed similar traits pathologically. Initially, 107 radiomic features were acquired and then three were finally selected. The support vector machine (SVM) model was able to differentiate the two anomalies from each other with an area under the curve (AUC) of 0.807 (95%CI 0.602-1.000) and 0.846 (95%CI 0.659-1.000) in training and testing cohort, respectively.
Conclusion
The derived radiomic features were helpful in differentiating KHE from FAVA. A model which contained these features might further improve the performance and hopefully could serve as a potential tool for identification.