W.N. Street. Cancer diagnosis and prognosis via linear-programming-based machine learning. Technical Report 94-14, Computer Sciences Department, University of Wisconsin, Madison, Wisconsin, August 1994. Ph.D. thesis.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....minimization is a linear program. This robust linear program (RLP) 4] has been used for decision tree construction [1] RLP combined with the greedy sequential backward elimination method for feature minimization, a simplified version of SBE, forms the basis of a breast cancer diagnosis system [28, 27]. The second error function is a slight modification of the first. In addition to decreasing the average magnitude of misclassified points, it also decreases the maximum classification error. This problem can also be written as a linear program [3] We will refer to it as the Perturbed Robust ....

....satisfying a specific misclassification error bound. In our e#ort to achieve this goal, we propose removing # from the problem by bounding the error function in a constraint. Problem (7) removes features while maintaining accuracy within some tolerance, #. A similar concept was used by [7] and 6 [27] in their feature elimination methods. In [7] feature elimination was allowed to continue as long as a specific error tolerance was maintained. Street [27] computed planes for all feature counts and then used a tuning set to determine the best plane. We call this problem feature minimization with ....

[Article contains additional citation context not shown here]

W.N. Street. Cancer diagnosis and prognosis via linear-programming-based machine learning. Technical Report 94-14, Computer Sciences Department, University of Wisconsin, Madison, Wisconsin, August 1994. Ph.D. thesis.

.... cancer has recurred (TTR = time to recur) In cases in which cancer has not recurred [outcome = non recur] the third feature is disease free time in months (DFS) Features 4 33 are the mean, standard error and worst values of ten real valued features computed for each cell nucleus by the Xcyt [154] program at time of diagnosis. Feature 34 is the diameter of the excised tumor and feature 35 is the number of positive axillary lymph nodes observed at time of surgery. Four of the 198 instances are missing the value for feature 35. These instances were discarded. We created the sets A and B to ....

....32 features. Feature 1 is an identification and was discarded. Feature 2 is diagnosis (M = malignant, B = benign) and was used to assign instances to the two datasets A and B. Features 3 32 are the mean, standard error, and worst measurements of ten real valued features computed by the Xcyt [154] program at time of diagnosis. When referring to the WDBC dataset, we refer to the point set Aae R 30 consisting of features 3 32 for 357 instances which have a benign diagnosis and the set Bae R 30 consisting of the same features for 212 instances with a malignant diagnosis. Each feature has ....

W. N. Street. Cancer diagnosis and prognosis via linear-programming-based machine learning. Computer Sciences Department, Mathematical Programming Technical Report 94-14, University of Wisconsin, Madison, Wisconsin, August 1994.

....problem without feature minimization is a linear program. This robust linear program (RLP) 3] has been used for decision tree construction [1] RLP combined with the greedy sequential backward elimination method for feature minimization forms the basis of a breast cancer diagnosis system [17, 16]. Our feature minimization method could also be applied to algorithms that minimize the number of points misclassified such as [2, 11] or to other successful linear programming approaches [10, 15] but we leave these extensions for future work. 3 The following robust linear programming problem, ....

....satisfying a specific misclassification error bound. In our effort to achieve this goal, we propose removing from the problem by bounding the error function in a constraint. Problem (7) removes features while maintaining accuracy within some tolerance, ffi. A similar concept was used by [6] and [16] in their feature elimination methods. In [6] feature elimination was allowed to continue as long as a specific error tolerance was maintained. Street [16] computed planes for all feature counts and then used a tuning set to determine the best plane. We call this problem feature minimization with ....

[Article contains additional citation context not shown here]

W.N. Street. Cancer diagnosis and prognosis via linear-programming-based machine learning. Technical Report 94-14, Computer Sciences Department, University of Wisconsin, Madison, Wisconsin, August 1994. Ph.D. thesis.

....from fine needle aspirates [Mangasarian] Their goal was to achieve the upper bound of the reported accuracy of the method of visual inspection of fine needle aspirates. 2.2.1 Features The system that they used is called Xcyt, which was written by one of the coauthors in his Ph.D. dissertation [Street, 1994]. A fine needle aspirate is taken directly from a lump in a patient s breast. The extracted fluid is then stained to emphasize the nuclei of the cells in the fluid. Then, a digital image of the fluid is taken. In the previous two papers that used mammograms for diagnosis, the authors simply ....

....also apt to generate disease free survival curves for individual breast cancer patients [Street, 1998] 3.2.1 Data The network was tested on two different sets of data. The first data set is the Wisconsin Prognostic Breast Cancer Data (WPBCD) from which the author extracts 32 features using Xcyt [Street, 1994] (which was described in section 2.2) These features include: area, radius, perimeter, symmetry, number and size of concavities, fractal dimension (of the boundary) compactness, smoothness (local variation of radial segments) and texture (variance of gray levels inside the boundary) ....

W. N. Street, Cancer Diagnosis and Prognosis via Linear-Programming-Based Machine Learning, Ph.D. dissertation, University of Wisconsin-Madison, 1994.

....can result in excessively large trees that do not reflect the underlying structure of the data. Pruning may not be sufficient to compensate for overfitting. This problem is readily shown in multivariate decision trees. The pruning process frequently produces a tree consisting of a single decision [2, 5, 33]. Univariate algorithms appear less susceptible to this problem. Murthy and Salzburg found that greedy heuristics worked well and lookahead algorithms offered little improvement [25, 26] We believe that this is because univariate decision trees have only one degree of freedom at each decision. ....

W.N. Street. Cancer diagnosis and prognosis via linear-programming-based machine learning. Technical Report 94-14, Computer Sciences Department, University of Wisconsin, Madison, Wisconsin, August 1994. Ph.D. thesis.

....applications of linear programming in the field of breast cancer research, one in diagnosis and one in prognosis. Both applications, currently in clinical practice, depend on the analysis of cellular images, accomplished with a computer program called Xcyt, written by one of the authors [31], that we describe in Section 1. The first application to breast cancer diagnosis has been described earlier [17, 18, 32, 34, 35] and is outlined in Section 2. The details of the diagnosis process in a clinical setting are described in Section 3. The second application to breast cancer prognosis ....

....that we describe in Section 1. The first application to breast cancer diagnosis has been described earlier [17, 18, 32, 34, 35] and is outlined in Section 2. The details of the diagnosis process in a clinical setting are described in Section 3. The second application to breast cancer prognosis [31] has not been published in the open literature and is described in some detail in Section 4. Computational results concerning the expected accuracy of the prognostic system are contained in Section 5. Section 6 shows the clinical importance of recurrence prediction, and Section 7 describes some ....

W. N. Street. Cancer Diagnosis and Prognosis via Linear-Programming-Based Machine Learning. PhD thesis, University of Wisconsin-Madison, August 1994. Available as University of Wisconsin Mathematical Programming TR 94--14.

....manner in which inequalities are handled in linear programming, we are able to include all available cases to build a more accurate, robust predictive model. Our solution to this estimation problem is termed the recurrence surface approximation (RSA) technique (Mangasarian et al. 1994; Street, 1994; Wolberg et al. 1995) RSA uses linear programming to determine a linear combination of the input features that accurately predicts TTR. The intuitive motivation for the RSA approach is that: ffl Recurrences actually take place at some point in time prior to their detection. However, the ....

Street, W. N. (1994). Cancer Diagnosis and Prognosis via Linear-Programming-Based Machine Learning.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC