Best imaging signs identified by radiomics could outperform the model: application to differentiating lung carcinoid tumors from atypical hamartomas.

Paul HabertAntoine DecouxLilia ChermatiLaure GibaultPascal ThomasArthur VaroquauxFrançoise Le Pimpec-BarthesArmelle ArnouxLoïc JuquelKathia ChaumoitreStéphane GarciaJean-Yves GaubertLoïc DuronLaure Fournier

Published in: Insights into imaging (2023)

• 3D-'Median' was the best feature to differentiate carcinoids from atypical hamartomas (AUC = 0.85). • 3D-'Median' feature is reproducible (ICC = 0.97) and was generalized to an external dataset. • Radiomics signature from 3D-segmentations differentiated carcinoids from atypical hamartomas with an AUC = 0.76. • 2D-ROI value reached similar performance to 3D-'median' but was less reproducible (ICC = 0.90).

Keyphrases

machine learning
contrast enhanced
lymph node metastasis
deep learning
high resolution
computed tomography
neural network
mass spectrometry
photodynamic therapy