We present the first attempt to perform short glass fiber segmentation from x-ray computed tomography (CT) volumetric datasets using deep learning architecture. We are able to increase the accuracy of segmentation up to ~ 0.9 Dice score voxel-wise, compared to ~ 0.7 Dice score achieved by standard 3D Frangi veselness technique [1] on our dataset. The experiments have been performed on synthetic and real CT scans acquired at 3.9 μm isotropic resolution. The artificial neural networks show robustness in segmenting fiber regions that are close to each other, which Hessian based techniques tend to misclassify.
Fully Convolutional Deep Network Architectures for Automatic Short Glass Fiber Segmentation from CT scans
Jitendra Rathore;Simone Carmignato;
2018
Abstract
We present the first attempt to perform short glass fiber segmentation from x-ray computed tomography (CT) volumetric datasets using deep learning architecture. We are able to increase the accuracy of segmentation up to ~ 0.9 Dice score voxel-wise, compared to ~ 0.7 Dice score achieved by standard 3D Frangi veselness technique [1] on our dataset. The experiments have been performed on synthetic and real CT scans acquired at 3.9 μm isotropic resolution. The artificial neural networks show robustness in segmenting fiber regions that are close to each other, which Hessian based techniques tend to misclassify.
File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
