Herman Verrelst

Prognostic importance of degree of differentiation and cyst rupture in stage I invasive epithelial ovarian carcinoma

BACKGROUND Previous studies on prognostic factors in stage I invasive epithelial ovarian carcinoma have been too small for robust conclusions to be reached. We undertook a retrospective study in a large international database to identify the most important prognostic variables. METHODS 1545 patients with invasive epithelial ovarian cancer (International Federation of Gynaecology and Obstetrics [FIGO] stage I) were included. The records of these patients were examined and data extracted for univariate and multivariate analysis of disease-free survival in relation to various clinical and pathological variables. FINDINGS The multivariate analyses identified degree of differentiation as the most powerful prognostic indicator of disease-free survival (moderately vs well differentiated hazard ratio 3.13 [95% CI 1.68-5.85], poorly vs well differentiated 8.89 [4.96-15.9]), followed by rupture before surgery (2.65 [1.53-4.56]), rupture during surgery (1.64 [1.07-2.51]), FIGO 1973 stage Ib vs Ia 1.70 [1.01-2.85]) and age (per year 1.02 [1.00-1.03]). When the effects of these factors were accounted for, none of the following were of prognostic value: histological type, dense adhesions, extracapsular growth, ascites, FIGO stage 1988, and size of tumour. INTERPRETATION Degree of differentiation, the most powerful prognostic indicator in stage I ovarian cancer, should be used in decisions on therapy in clinical practice and in the FIGO classification of stage I ovarian cancer. Rupture should be avoided during primary surgery of malignant ovarian tumours confined to the ovaries.

A comparison of methods for preoperative discrimination between malignant and benign adnexal masses: The development of a new logistic regression model☆☆☆★

OBJECTIVE The aim of this study was to assess the complementary use of ultrasonographic end points with the level of circulating CA 125 antigen by multivariate logistic regression analysis algorithms to distinguish malignant from benign adnexal masses before operation. STUDY DESIGN One hundred ninety-one patients aged 18 to 93 years with overt adnexal masses were examined by transvaginal ultrasonography with color Doppler imaging and 31 variables were recorded. The end points were the histologic classification of the tumor and the areas under the receiver-operator characteristic curves of alternative algorithms. RESULTS One hundred forty patients had benign tumors and 51 (26.7%) had malignant tumors: 31 primary invasive tumors (37% International Federation of Gynecology and Obstetrics stage I), 5 tumors of borderline malignancy (100% International Federation of Gynecology and Obstetrics stage I), and 15 tumors were metastatic and invasive. The most useful variables for the logistic regression analysis were the menopausal status, the serum CA 125 level, the presence of >/=1 papillary growth (>3 mm in length), and a color score indicative of tumor vascularity and blood flow. The optimized procedure had a sensitivity of 95.9% and a specificity of 87.1%. The area under the receiver-operator characteristic curve was significantly higher (P <.01) than the corresponding values from the independent use of serum CA 125 levels or indexes of tumor form or vascularity. CONCLUSION Regression analysis of a few complementary variables can be used to accurately discriminate between malignant and benign adnexal masses before operation.

Detection of Mobile Phone Fraud Using Supervised Neural Networks: A First Prototype

We present the first prototype of a tool based on a supervised neural network for the detection of fraud in mobile communications. This prototype is being developed in the framework of a project of the European Commission on Advanced Security for Personal Communications (ASPeCT), together with two other prototypes, based on unsupervised neural networks and knowledge-based systems.

Bayesian networks in ovarian cancer diagnosis: potentials and limitations

The pre-operative discrimination between malignant and benign masses is a crucial issue in gynaecology. Next to the large amount of background knowledge, there is a growing amount of collected patient data that can be used in inductive techniques. These two sources of information result in two different modelling strategies. Based on the background knowledge, various discrimination models have been constructed by leading experts in the field, tuned and tested by observations. Based on the patient observations, various statistical models have been developed, such as logistic regression models and artificial neural network models. For the efficient combination of prior background knowledge and observations, Bayesian network models are suggested. We summarize the applicability of this technique, report the performance of such models in ovarian cancer diagnosis and outline a possible hybrid usage of this technique.

On-Line Learning Fokker-Planck Machine

In this letter we present an on-line learning version of the Fokker-Planck machine. The method makes use of a regularized constrained normalized LMS algorithm in order to estimate the time-derivative of the parameter vector of a radial basis function network. The RBF network parametrizes a transition density which satisfies a Fokker-Planck equation, associated to continuous simulated annealing. On-line learning using the constrained normalized LMS method is necessary in order to make the Fokker-Planck machine applicable to large scale nonlinear optimization problems.

The Effect of Antibody Morphology on Non-self Detection

Anomaly detection algorithms inspired by the natural immune system often use the negative selection metaphor to implement non-self detection. Much research has gone into ways of generating good sets of non-self detectors or antibodies and these methods’ time and space complexities. In this paper, the antibody morphology is defined as the collection of properties defining the shape, data-representation and data-ordering of an antibody. The effect these properties can have on self/non-self classification capabilities is investigated. First, a data-representation using fuzzy set theory is introduced. A comparison is made between the classification performance using fuzzy and m-ary data-representations using some benchmark machine learning data-sets from the UCI archive. The effects of an antigen data reordering mechanism based on Major Histocompatibility Complex (MHC) molecules is investigated. The population level effect this mechanism can have by reducing the number of holes in the antigen space is discussed and the importance of data order in the r-contiguous symbol match-rule is highlighted. Both are analysed quantitatively using some UCI data-sets.

Application of NLq Neural Control Theory to a Ball and Beam System

This paper is a straightforward application of NL q stability criteria to neural model-based controller design. We discuss the design of a linear dynamic output feedback controller for a ball and beam system for which a neural state space model is identified. This is done by applying dynamic backpropagation, constrained by internal or I/O stability conditions for NL q systems. The performance of the controller has been tested both by computer simulations and on a real ball and beam set-up.