Deep Learning and Structured Prediction for the Segmentation of Mass in Mammograms
In this paper, we explore the use of deep convolution and deep belief networks as potential functions in structured prediction models for the segmentation of breast masses from mammograms. In particular, the structured prediction models are estimated with
- PDF / 1,747,935 Bytes
- 8 Pages / 439.363 x 666.131 pts Page_size
- 36 Downloads / 174 Views
ACVT, School of Computer Science, The University of Adelaide 2 School of ITEE, The University of Queensland
Abstract. In this paper, we explore the use of deep convolution and deep belief networks as potential functions in structured prediction models for the segmentation of breast masses from mammograms. In particular, the structured prediction models are estimated with loss minimization parameter learning algorithms, representing: a) conditional random field (CRF), and b) structured support vector machine (SSVM). For the CRF model, we use the inference algorithm based on tree re-weighted belief propagation with truncated fitting training, and for the SSVM model the inference is based on graph cuts with maximum margin training. We show empirically the importance of deep learning methods in producing state-of-the-art results for both structured prediction models. In addition, we show that our methods produce results that can be considered the best results to date on DDSM-BCRP and INbreast databases. Finally, we show that the CRF model is significantly faster than SSVM, both in terms of inference and training time, which suggests an advantage of CRF models when combined with deep learning potential functions. Keywords: Deep learning, Structured output learning, Mammogram segmentation.
1
Introduction
Screening mammogram is one of the most effective imaging modalitites to detect breast cancer, and it is used for the segmentation of breast masses (among other tasks), which is a challenging task due to the variable shape/size of masses [1] and their low signal-to-noise ratio (see Fig. 1). In clinical practice, lesion segmentation is usually a manual process, and so its efficacy is associated with the radiologist’s expertise and workload [2], where a clear trade-off can be noted between sensitivity (Se) and specificity (Sp) in manual interpretation, with a median Se of 83.8% and Sp of 91.1% [2]. The main goal of this paper is to introduce and evaluate a new methodology for segmenting masses from mammograms based on structured prediction models
This work was partially supported by the Australian Research Council’s Discovery Projects funding scheme (project DP140102794). Prof. Bradley is the recipient of an Australian Research Council Future Fellowship(FT110100623).
c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part I, LNCS 9349, pp. 605–612, 2015. DOI: 10.1007/978-3-319-24553-9_74
606
N. Dhungel, G. Carneiro, and A.P. Bradley
Fig. 1. Structured prediction model with a list of potential functions that include two deep learning methods and two structured prediction models
that use deep learning as their potential functions (Fig. 1). Our main contribution is the introduction of powerful deep learning appearance models, based on CNN [3,4] and DBN [5], into the following recently proposed structured output models: a) a conditional random field (CRF), and b) structured support vector machines (SSVM). The CRF model performs inference with tree re-weighted belief propagation [6
Data Loading...
