Breast Lesion Segmentation Software for DCE-MRI - Open Source GPGPU Based Optimization

Abstract

tion process. The proposed framework is based on level sets, machine learning and parallel programming and can be used for the assessment of volume changes in breast lesions, especially for the evaluation of new cancer therapies like neoadjuvant therapy. The proposed method has two main steps: a pre-processing module and a breast lesion segmentation step. In the pre-processing step, dynamic contrast-enhanced MRI (DCE-MRI) images in the dynamic sequence are registered using an intensity-based rigid transformation algorithm. After the registration, voxels that correspond to breast lesions are enhanced using a machine learning classifier. Finally, during the segmentation step, the post-processed images are used for volume segmentation of the breast lesions using a level set method based on the active contours without edges (ACWE) algorithm by Chan and Vese [4]. The segmentation algorithm is implemented in OpenCL for the general-purpose computing on GPUs (GPGPUs) to accelerate the original model by parallelizing two main steps of the segmentation process: the computation of the signed distance function (SDF) and the evolution of the segmented curve. A non optimized 2D version of the ACWE algorithm was previously implemented in OpenCL by Zavala et al. in 2012 [5]. Efficient algorithms for segmentation are a key step in medical imaging and of fundamental importance in computer aided diagnosis of breast cancer for: diagnostics, evaluation of neoadjuvant therapy, or surgery. With the advance of high resolution images, as 3D dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) images, the computational cost of segmentation methods has become more expensive as the amount of data has grown. In this work, a segmentation method for breast cancer lesions in DCE-MRI images based on the active contour without edges (ACWE) algorithm and using parallel programming with general purpose computing on graphics processing units (GPGPUs) is presented. The performance of the segmentation algorithm is evaluated on a set of 32 breast DCE-MRI cases in terms of speedup, and compared to non-GPU based approaches. A high speedup (40 or more) is obtained for high resolution images, providing real-time outputs.

Cite

@inproceedings{zavalaromero2018breast,
  title = {Breast Lesion Segmentation Software for DCE-MRI - Open Source GPGPU Based Optimization},
  author = {Zavala-Romero, Olmo and Meyer-Baese, Anke and Lobbes, Marc B.I.},
  year = {2018},
  url = {/media/papers/Proceedings/2018 - Zavala-Romero et al. - Breast Lesion Segmentation Software for DCE-MRI - Open Source GPGPU Based Optimization_proceedings.pdf},
}

Topics

• Computer Vision
• Deep Learning
• Health
• SciML