8576

Chest CT automatic analysis for lung nodules detection implemented on a GPU computing system

Niccolo Camarlinghi, Francesco Bagagli, Piergiorgio Cerello, Alessandra Retico, Maria Evelina Fantacci
Dipartimento di Fisica dell’univerisit’a di Pisa
IEEE NSS, 2012

@article{camarlinghi2012chest,

   title={Chest CT automatic analysis for lung nodules detection implemented on a GPU computing system},

   author={Camarlinghi, N. and Bagagli, F. and Cerello, P. and Retico, A. and Fantacci, M.E.},

   year={2012}

}

Download Download (PDF)   View View   Source Source   

2163

views

The aim of this work is the efficient implementation of the Hessian based filters. These filters are commonly used in medical image analysis and are employed in the Voxel Based Neural Approach (VBNA) lung CAD (Computer Aided Detection) system for lung nodule detection. This work mainly focuses on the optimization of the filter devoted to the detection of internal nodule candidates, called Multi Scale Dot Enhancement (MSDE) algorithm. Two fast variants of the MSDE algorithm are here proposed and compared: the first relies on an analytical optimization of the algorithm and it is implemented on a standard CPU, whereas the second consists in implementing the filter in the CUDA Graphical Processing Unit (GPU) framework. The algorithms were tested with computed tomography images belonging to the Lung Image Database Consortium (LIDC) public research database using an Intel Core i7 950 @ 3.07GHz and a NVIDIA GeForce GTX 580. Both the approaches lead to an improvement in the algorithm performance with respect to the original implementation, without any loss of precision. The initial implementation, realized in the Insight ToolKit open source image analysis framework (ITK), had an average execution time of 69 sec per CT using five scales of enhancement. The analytically-optimized CPU algorithm leads to a computational speed gain of 2.5x (28 sec per CT), whereas the parallel CUDA implementation leads to a speed-up of 38x (1.8 sec per CT) with respect to the original implementation, and 15x with respect to the analytical approach. This work has been developed in the framework of the INFN-funded MAGIC-5 project.
Rating: 2.5/5. From 2 votes.
Please wait...

* * *

* * *

HGPU group © 2010-2024 hgpu.org

All rights belong to the respective authors

Contact us: