Dempster-Shafer Theory Based Feature Selection with Sparse Constraint for Outcome Prediction in Cancer Therapy
As a pivotal task in cancer therapy, outcome prediction is the foundation for tailoring and adapting a treatment planning. In this paper, we propose to use image features extracted from PET and clinical characteristics. Considering that both information s
- PDF / 320,719 Bytes
- 8 Pages / 439.363 x 666.131 pts Page_size
- 94 Downloads / 196 Views
Sorbonne Universit´es, Universit´e de Technologie de Compi`egne, CNRS, UMR 7253 Heudiasyc, 60205 Compi`egne, France 2 Universit´e de Rouen, QuantIF - EA 4108 LITIS, 76000 Rouen, France 3 Centre Henri-Becquerel, Department of Nuclear Medicine, 76038 Rouen, France 4 Washington University School of Medicine, Department of Radiation Oncology, Saint Louis, 63110 MO, USA
Abstract. As a pivotal task in cancer therapy, outcome prediction is the foundation for tailoring and adapting a treatment planning. In this paper, we propose to use image features extracted from PET and clinical characteristics. Considering that both information sources are imprecise or noisy, a novel prediction model based on Dempster-Shafer theory is developed. Firstly, a specific loss function with sparse regularization is designed for learning an adaptive dissimilarity metric between feature vectors of labeled patients. Through minimizing this loss function, a linear low-dimensional transformation of the input features is then achieved; meanwhile, thanks to the sparse penalty, the influence of imprecise input features can also be reduced via feature selection. Finally, the learnt dissimilarity metric is used with the Evidential K-Nearest-Neighbor (EKNN) classifier to predict the outcome. We evaluated the proposed method on two clinical data sets concerning to lung and esophageal tumors, showing good performance. Keywords: Outcome Prediction, PET, Feature Selection, Sparse Constraint, Dempster-Shafer Theory.
1
Introduction
Accurately predicting the treatment outcome prior to or even during cancer therapy is of great clinical value. It facilitates the adaptation of a treatment planning for individual patient. Medical imaging plays a fundamental role in assessing the response of a treatment, as it can monitor and follow-up the evolution of tumor lesions non-invasively [8]. Up to now, the metabolic uptake information provided by fluoro-2-deoxy-D-glucose (FDG) positron emission tomography (PET) has been proven to be predictable for pathologic response of a treatment in several cancers, e.g., lung tumor [6,11] and esophageal tumor [15]. Abounding image features can be extracted from FDG-PET, such as standardized uptake values (SUVs), like SUVmax , SUVpeak and SUVmean , that describe metabolic uptake c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part III, LNCS 9351, pp. 695–702, 2015. DOI: 10.1007/978-3-319-24574-4_83
696
C. Lian et al.
in a region of interest, total lesion glycolysis (TLG) and metabolic tumor volume (MTV) [15]. In addition, some complementary analysis of PET images, e.g., image texture analysis [16], can also provide supplementary evidence for outcome evaluation. The quantification of these features before and during cancer therapy has been claimed to be predictable for treatment response [8]. Nevertheless, their further application is still hampered by some practical difficulties. First, compared to a relatively large amount of interesting features from the point of view of clinicians, we oft
Data Loading...
