Osteosarcoma is primary malignant neoplasms derived from cells of mesenchymal origin, and often has distinct phenotypes at different stages. The location of tumor and reaction zone can be identified by an expert in ma...Osteosarcoma is primary malignant neoplasms derived from cells of mesenchymal origin, and often has distinct phenotypes at different stages. The location of tumor and reaction zone can be identified by an expert in magnetic resonance imaging (MRI), with MRI being one of the choices for evaluating the extent of osteosarcoma. However, it is still a challenge to automatically extract tumor from its surrounding tissues because of their low intensity differences in MRI. We investigated an approach based on Zernike moment and support vector machine (SVM) for osteosarcoma segmentation in T1-weighted image (TIWI). Firstly, the different order moments around each pixel are calculated in small windows. Secondly, the grayscale and the module values of different order moments are used as a texture feature vector which is then used as the training set for SVM. Finally, an SVM classifier is trained based on this set of features to identify the osteosarcoma, and the segmented tumor tissue is rendered in 3D by the ray casting algorithm based on graphics processing unit (GPU). The performance of the method is validated on T1WI, showing that the segmentation method has a high similarity index with the expert's manual segmentation.展开更多
Dropout and other feature noising schemes have shown promise in controlling over-fitting by artificially corrupting the training data. Though extensive studies have been performed for generalized linear models, little...Dropout and other feature noising schemes have shown promise in controlling over-fitting by artificially corrupting the training data. Though extensive studies have been performed for generalized linear models, little has been done for support vector machines （SVMs）, one of the most successful approaches for supervised learning. This paper presents dropout training for both linear SVMs and the nonlinear extension with latent representation learning. For linear SVMs, to deal with the intractable expectation of the non-smooth hinge loss under corrupting distributions, we develop an iteratively re-weighted least square （IRLS） algorithm by exploring data augmentation techniques. Our algorithm iteratively minimizes the expectation of a re- weighted least square problem, where the re-weights are analytically updated. For nonlinear latent SVMs, we con- sider learning one layer of latent representations in SVMs and extend the data augmentation technique in conjunction with first-order Taylor-expansion to deal with the intractable expected hinge loss and the nonlinearity of latent representa- tions. Finally, we apply the similar data augmentation ideas to develop a new IRLS algorithm for the expected logistic loss under corrupting distributions, and we further develop a non-linear extension of logistic regression by incorporating one layer of latent representations. Our algorithms offer insights on the connection and difference between the hinge loss and logistic loss in dropout training. Empirical results on several real datasets demonstrate the effectiveness of dropout training on significantly boosting the classification accuracy of both linear and nonlinear SVMs.展开更多
文摘Osteosarcoma is primary malignant neoplasms derived from cells of mesenchymal origin, and often has distinct phenotypes at different stages. The location of tumor and reaction zone can be identified by an expert in magnetic resonance imaging (MRI), with MRI being one of the choices for evaluating the extent of osteosarcoma. However, it is still a challenge to automatically extract tumor from its surrounding tissues because of their low intensity differences in MRI. We investigated an approach based on Zernike moment and support vector machine (SVM) for osteosarcoma segmentation in T1-weighted image (TIWI). Firstly, the different order moments around each pixel are calculated in small windows. Secondly, the grayscale and the module values of different order moments are used as a texture feature vector which is then used as the training set for SVM. Finally, an SVM classifier is trained based on this set of features to identify the osteosarcoma, and the segmented tumor tissue is rendered in 3D by the ray casting algorithm based on graphics processing unit (GPU). The performance of the method is validated on T1WI, showing that the segmentation method has a high similarity index with the expert's manual segmentation.
文摘Dropout and other feature noising schemes have shown promise in controlling over-fitting by artificially corrupting the training data. Though extensive studies have been performed for generalized linear models, little has been done for support vector machines （SVMs）, one of the most successful approaches for supervised learning. This paper presents dropout training for both linear SVMs and the nonlinear extension with latent representation learning. For linear SVMs, to deal with the intractable expectation of the non-smooth hinge loss under corrupting distributions, we develop an iteratively re-weighted least square （IRLS） algorithm by exploring data augmentation techniques. Our algorithm iteratively minimizes the expectation of a re- weighted least square problem, where the re-weights are analytically updated. For nonlinear latent SVMs, we con- sider learning one layer of latent representations in SVMs and extend the data augmentation technique in conjunction with first-order Taylor-expansion to deal with the intractable expected hinge loss and the nonlinearity of latent representa- tions. Finally, we apply the similar data augmentation ideas to develop a new IRLS algorithm for the expected logistic loss under corrupting distributions, and we further develop a non-linear extension of logistic regression by incorporating one layer of latent representations. Our algorithms offer insights on the connection and difference between the hinge loss and logistic loss in dropout training. Empirical results on several real datasets demonstrate the effectiveness of dropout training on significantly boosting the classification accuracy of both linear and nonlinear SVMs.
