مدل الگوریتم ژنتیک بر اساس شبکه عصبی مصنوعی برای پیش بینی وضعیت غدد لنفاوی زیر بغل در سرطان سینه
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|8128||2013||6 صفحه PDF||سفارش دهید||4350 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Engineering Applications of Artificial Intelligence, Volume 26, Issue 3, March 2013, Pages 945–950
Axillary Lymph Node (ALN) status is an extremely important factor to assess metastatic breast cancer. Surgical operations which may be necessary and cause some adverse effects are performed in determination ALN status. The purpose of this study is to predict ALN status by means of selecting breast cancer patient's basic clinical and histological feature(s)that can be obtained in each hospital. 270 breast cancer patients' data are collected from Ankara Numune Educational and Research Hospital and Ankara Oncology Educational and Research Hospital. These are classified using back propagation MultiLayer Perceptron (MLP), Logistic Regression (LR) and Genetic Algorithm (GA) based MLP models. Receiver Operating Characteristics (ROC) such as sensitivity, specificity, accuracy and area under of ROC (AUC) and regression are used to evaluate performances of the developed models. It is concluded from LR and GA based MLP, that menopause status and lymphatic invasion are the most significant features for determining ALN status. GA provides to select best features as MLP inputs. It also optimizes the weights of backpropagation algorithm in MLP. The values of regression and accuracy of the GA based MLP with 9features (numerical age, categorical age, menopause status, tumor size, tumor type, tumor location, T staging, tumor grade and lymphatic invasion) are found as 0.96 and 98.0% with respectively. According to results, proposed GA based MLP classifier can be used to predict the ALN status of breast cancer without surgical operations.
Breast cancer is one of the leading tumor-related causes of death in women (He et al., 2009). If cancer cells are present in Axillary Lymph Nodes (ALN), a risk for metastatic breast cancer rises. ALN may be removed by different types of lymph node dissection surgeries such as Sentinel Lymph Node Biopsy (SLNB) and Axillary Lymph Node Dissection (ALND). Research shows that more surgery operations may be necessary and cause some adverse events such as lymphoedema, restriction of arm and shoulder movement and numbness of upper arm skin (Karakis et al., 2011). If ALN status of breast cancer patients could be accurately predicted from basic clinical and histological features, surgical operations could be prevented (Patani et al., 2007 and Karakis et al., 2011). Previous researches use the statistical methods such as logistic regression, univariate and multivariate analysis in this area. It is showed that no single feature or combinations of features are sufficiently accurate to predict axillary status (Patani et al., 2007). For example, Harden et al. found that lymphvascular invasion and tumor size could be determined but tumor grade was not associated with ALN status (Harden et al., 2001). However, Farshid et al. did not find significance of tumor size and grade (Farshid et al., 2004). Hence, Artificial Neural Networks (ANN) have been proposed as a supplement or alternative to standard statistical techniques for prediction of axillary lymph node status in breast cancer patients (Tez et al., 2007 and Karakis et al., 2011). Artificial Neural Networks (ANNs) are a form of artificial intelligence that have been successfully applied to a variety of problems including pattern recognition, modeling, control and medical field (Papik et al., 1998, Haykin, 1999, Haykin, 2000, Lisboaa and Taktak, 2006, Marchevsky, 2006, Paliwal and Kumar, 2009 and Ferreira and Gil, 2012). However, there is a few artificial neural network studies in ALN status prediction (Naguib et al., 1996, Marchevsky et al., 1999, Seker et al., 2000, Seker et al., 2002, Seker et al., 2003, Mattfeldt et al., 2004, Lancashire et al., 2008 and Karakis et al., 2011). In this field, the most striking results have been obtained by Marchevsky et al. (1999) and Lancashire et al. (2008). Marchevsky et al. (1999) evaluated 19 prognostic features of 279 patients through probabilistic Neural Networks (NNs) based on genetic algorithms and logistic regression classifiers. Lancashire et al. (2008) applied the MLP to a gene microarray dataset that consists of 49 samples to identify gene signatures corresponding with estrogen receptor and ALN status in breast cancer. The high predictive accuracy values were obtained 89% and 100% respectively in these studies. However, the used gene protein data or flow-cytometric data are difficult to obtain in each hospital. Mattfeldt et al. (2004) used the histological (the age, tumor type, grade and size, skin infiltration, lymphangiosis carcinomatosa, pT4 category) and flow-cytometric (percentage of tumor cells in G2/M- and S-phases of the cell cycle, and ploidy index) features to predict ALN status. It also stated that basic features could be used as determiners. Nonetheless, the predictive accuracy of performed model was only 72%. Previous study classified axillary lymph node status of 270 breast cancer patients through pattern recognition techniques such as MultiLayer Perceptron (MLP), Support Vector Machines (SVM), Linear Discriminant Analysis (LDA) and k-Nearest Neighbor (k-NN) (Karakis et al., 2011). MLP network structure that use Levenberg–Marquardt algorithm in back propagation training was consisted of a hidden layer with 4neurons. MLP classifier found the correlation coefficient and the accuracy value of testing dataset as 0.872 and 94.4% respectively. However, none of the classifiers have the high accuracy value as features are selected randomly. Because, a general problem in MLP model selection is to obtain right parameters that fit a model with observed data. Hence, feature selection can be described as an optimization problem. The data without any feature selection might be redundant and may deteriorate. It also influences the efficiency of classification. Feature selection reduces computational cost, storage requirements, and training time. It also facilitates Back Propagation (BP) training process (Castillo et al., 2000 and Lin et al., 2008). Feature selection approaches can be categorized in two models. The filter model is based on statistical approaches and it is also quite fast technique. Nonetheless, the resulting feature subset may not be the best. The wrapper model uses some selection methods to choose feature subsets. It evaluates the result by means of the accuracy rates of classification algorithm calculates the accuracy rates. The wrapper model is widely used feature selection with Genetic Algorithm (GA) approaches in MLP (Lin et al., 2008). GA has been used with neural network to search for input features or to determine the number of nodes or connections in the network (Handels et al., 1999, Arifovica and Gencay, 2001, Palmes et al., 2005, Nicola´s et al., 2006, Ambrogi et al., 2007 and Benardos and Vosniakos, 2007). This study has used only basic clinical and histological features of breast cancer patients. The data have been obtained from Ankara Numune Educational and Research Hospital and Ankara Oncology Educational and Research Hospital. The aims of our analysis are to identify potential feature(s) of ALN status and propose a classifier model by means of MLP, LR and GA basedMLP. The rest of this study is organized as follows. After the first section, material and method are described in Section 2. Section 3 presents results of developed models. It also includes a brief discussion of the obtained results. Finally Section 4 states the conclusion and futureworks.
نتیجه گیری انگلیسی
Axillary lymph node status is very important to determine the stage, prognosis and treatment of breast cancer. Surgical operations which may be necessary and cause some adverse effects are performed in determination ALN status. For this reason, the effects of 270 breast cancer patients' basic potential features have examined to determine a classifier model for ALN status. The classifier is selected through MLP, GA based MLP and LR models. The GA analysis results show that nine input features subset such as numerical age, categorical age, menopause status, tumor size, tumor type, tumor location, T staging, tumor grade and lymphatic invasion obtain the best result in literature studies except Lancashire et al. (2008). The main aim of our study is to identify the effects of basic clinical and histological features which can be easily collected in every hospital for ALN status. In the literature, some studies about ALN status have used the features which cannot be obtained easily in each hospital (Naguib et al., 1996, Marchevsky et al., 1999, Seker et al., 2000, Seker et al., 2002, Seker et al., 2003 and Lancashire et al., 2008). The others are unable to perform a high predictive accuracy (Naguib et al., 1996, Marchevsky et al., 1999, Seker et al., 2000, Seker et al., 2002, Seker et al., 2003, Mattfeldt et al., 2004 and Karakis et al., 2011). In this study, the accuracy value of GA based MLP is found as 0.98 with using basic nine features. In addition to, menopause status, and lymphatic invasion are determined to be significant as a predictor features with respect to the result of both LR and GA as shown the details in Fig. 4 and Table 2. We believe that GA based MLP classifier can be used for making decision axillary lymph node status of breast cancer patients. Consequently, surgical operations may not require. In the future, proposed GA based MLP classifier (NeuralGenetic) should be tested with additional data of breast cancer. For this reason, software has developed for using in the Department of General Surgery of Ankara Numune Educational and Research Hospital. Thus, MLP model will be generalized with different obtaineddata.