Review
Using support vector machines in diagnoses of urological dysfunctions

https://doi.org/10.1016/j.eswa.2009.12.055Get rights and content

Abstract

Urinary incontinence is one of the largest diseases affecting between 10% and 30% of the adult population and an increase is expected in the next decade with rising treatment costs as a consequence. There are many types of urological dysfunctions causing urinary incontinence, which makes cheap and accurate diagnosing an important issue. This paper proposes a support vector machine (SVM) based method for diagnosing urological dysfunctions. 381 registers collected from patients suffering from a variety of urological dysfunctions have been used to ensure the (generalization) performance of the decision support system. Moreover, the robustness of the proposed system is examined by fivefold cross-validation and the results show that the SVM-based method can achieve an average classification accuracy at 84.25%.

Introduction

Presently urinary incontinence affects between 10% and 30% of the adult population and it is expected to increase in the next decade with accelerating treatment costs as a consequence (Cortes and Kelleher, 2005, Wein, 2004). This rise in incidence is similar for the male and the female parts of the adult population (Irwin et al., 2006) (see Table 1).

The use of classifier systems in medical diagnosis is increasing gradually. There is no doubt that evaluation of data taken from patients and decisions of experts are the most important factors in diagnosis. However, expert systems and different artificial intelligence techniques for classification have the potential of being good supportive tools for the expert. Classification systems can help in increasing accuracy and reliability of diagnoses and minimizing possible errors, as well as making the diagnoses more time efficient (Akay, 2008).

Some of the related work in the field of the urological diagnosis has been developed basically by means of artificial neural networks (ANNs) (Gil et al., 2007, Gil et al., 2008). To increase the accuracy and the generalization ability we propose the use of a Support Vector Machine (SVM) based system combined with techniques for dimensionality reduction. In addition to ANNs, the SVM (Cortes & Vapnik, 1995) has also emerged as a powerful tool for classification. SVMs were proposed by Vapnik (1995) and is based on the structured risk minimization (SRM) principle. Hence it tries to minimize an upper bound of the generalization error instead of the empirical error as in the artificial neural networks. Therefore a particular advantage of SVMs over other classifiers is that they can achieve better performance when applied to real world problems (He, Hu, Harrison, Tai, & Pan, 2006). Some classifiers, such as ANNs suffer from the overfitting problem. In the case of the SVM overfitting is unlikely to occur. Overfitting is caused by too much flexibility in the decision boundary.

SVMs are global representatives of the whole set of training points, and there are usually few of them, which gives little flexibility. Thus overfitting is unlikely to occur (Witten & Frank, 2005). SVMs have been successfully applied to a wide variety of applications, e.g. including pattern recognition, biology and financial domains (Hearst et al., 1998, Hua and Sun, 2001, Huang and Wu, 2006, Shin et al., 2005, Wu et al., 2008, Yan et al., 2008).

The remaining part of the paper is organized as follows: first, we give a brief description of some basic SVM concepts. Next we describe the design of our proposal of the SVM-based decision support system with dimensionality reduction and the training of the SVM by the available data. Then we describe our testing of the system and analyze the results. Finally we draw relevant conclusions and suggest future lines of research.

Section snippets

Support vector machines

In this section, the basic concept of SVM will be briefly described. More thorough descriptions can be found in Burges, 1998, Theodoridis and Koutroumbas, 2003, Hsu et al., 2003. A typical two class problem as Fig. 1 shows is similar to the problem of diagnosing urological patients as either ill or healthy.

For a classification problem, it is necessary to first try to estimate a function f:RN→{±1} using training data, which are l N-dimensional patterns xi and class labels yi, where(x1,y1),…,(xl,y

Urological data

The input data in the system starts when a patient reports to a physician. Then, a large number of information to be considered during the diagnosis will be saved in a database. In this study, an exhaustive urological exploration with 20 different measurements has been carried out by using 381 patients with dysfunctions in the lower urinary tract (LUT). The 20 input variables (Table 2) that are essential to the diagnosis of the LUT diseases of interest are extracted from the urological

Conclusions and future work

In this paper we have evaluated the performance of a classifier constructed by means of the SVM method when applied to the diagnosis of urological dysfunctions. The SVM were trained with data from a database with registers of patients with urological dysfunctions. The experiment starts with a preprocessing of the urodynamical measures from every patient. This preprocessing includes missing data treatment and normalization process. After that, data are provided to the SVM which determines

Acknowledgments

We want to express our acknowledgements to the urologists of the Hospital of San Juan (Alicante-Spain), who have made it possible to reach a better understanding of the different types of urological dysfunctions. Moreover, the data used in the development of this system is the result of several years of this collaboration.

References (40)

  • Akay, M. F. (2008). Support vector machines combined with feature selection for breast cancer diagnosis. Expert systems...
  • K.P. Bennett et al.

    Robust linear programming discrimination of two linearly inseparable sets

    Optimization Methods and Software

    (1992)
  • C.J.C. Burges

    A tutorial on support vector machines for pattern recognition

    Data Mining and Knowledge Discovery

    (1998)
  • C. Cortes et al.

    Support-vector networks

    Machine Learning

    (1995)
  • R. Courant et al.
    (1953)
  • Das, S. (2001). Filters, wrappers and a boosting-based hybrid for feature selection. In Machine learning-international...
  • W. Duch

    Filter methods

    Studies in fuzziness and soft computing

    (2006)
  • X. Fu et al.

    Data dimensionality reduction with application to simplifying RBF network structure and improving classification performance

    IEEE Transactions on Systems, Man and Cybernetics, Part B

    (2003)
  • Gil, D., Soriano, A., Ruiz, D., & Montejo, C. A. (2007). Embedded system for diagnosing dysfunctions in the lower...
  • D. Gil et al.

    Application of artificial neural networks in the diagnosis of urological dysfunctions

    Expert Systems with Applications

    (2008)

    • Cited by (21)

    • Machine learning reveals salivary glycopatterns as potential biomarkers for the diagnosis and prognosis of papillary thyroid cancer

      2022, International Journal of Biological Macromolecules
      Citation Excerpt :

      SVM diagnosis was performed on 381 patients with urinary system diseases. The results showed that the method based on Support Vector Machine can reach 84.25 % classification accuracy [42]. If the SVM data mining method is combined with the expertise of various research fields, SVM can be extended to different fields.

    • Application of total X-Ray fluorescence to gunshot residue determination

      2019, Applied Radiation and Isotopes
      Citation Excerpt :

      Quadratic discriminant analysis (QDA) is the same as the former method but employs a quadratic equation which is usually more efficient (Esteki et al., 2018; Rodionova et al., 2016). Supporting vector machines (SVM) finds the best hyper-plane that discriminate two different classes of samples (Zendehboudi et al., 2018; Gil and Johnsson, 2010). In K-nearest neighbors (KNN) algorithm, the distance from the unknown sample to the samples in the training data set is calculated, then the class of the unknown sample is decided according to the class of majority of the k (k is a natural number) closest samples (Chirici et al., 2016; Devroye and Wagner, 1982).

    • Identifying central and peripheral nerve fibres with an artificial intelligence approach

      2018, Applied Soft Computing Journal
      Citation Excerpt :

      Then, it could be identified in the most suitable group, either in the optic or the cochlear nerve. The authors have experience in the medical field by applying several AI techniques in classification and prediction tasks with very good results [50,51,48]. Additionally, our architecture, characterized by complexity, generalization, and flexibility, can be extended to other biological control systems.

    • A novel acoustic emission detection module for leakage recognition in a gas pipeline valve

      2017, Process Safety and Environmental Protection
      Citation Excerpt :

      As a pattern recognition method, the support vector machine (SVM) has been extensively employed to solve classification and regression problems (Cortes and Vapnik, 1995). The SVM has been successfully used in many fields, including valve leakage (Yang et al., 2005), mechanical fault diagnosis (Oh and Sohn, 2009), image and video processing (Ding et al., 2008), medical engineering (Gil and Johnsson, 2010), and chemical engineering (Kulkarni et al., 2005). To improve the accuracy of classification, several feature parameters were calculated in the time and frequency domains.

    • Predicting seminal quality with artificial intelligence methods

      2012, Expert Systems with Applications
      Citation Excerpt :

      These AI methods can also help to make decision in the next steps of the infertility treatment, focusing or not on the male partner and avoiding painful and expensive examinations on the female. The authors have experience in the field of urology by applying several AI techniques in classification tasks with very good results (Gil & Johnsson, 2010b; Gil et al., 2009, Gil, Johnsson, Garcia Chamizo, Paya, & Fernandez, 2011). This makes it possible to adapt the artificial intelligence to meet the needs of the data analyzed in this study.

    • Hybrid principal component analysis and support vector machine model for predicting the cost performance of commercial building projects using pre-project planning variables

      2012, Automation in Construction
      Citation Excerpt :

      Details of the two stages of the proposed hybrid model are provided in the following subsections. Input feature vectors include irrelevant dimensions and redundant dimensions, both of which decrease the prediction accuracy of the resulting model while increasing the computational cost of employing the data-mining techniques needed to construct the prediction model [30–32]. To minimize such costs, in this study the PCA approach proposed by Dunteman [33] was adopted to allow for more efficient construction of the prediction model by reducing the dimensionality of the problem before the data mining took place.

    View all citing articles on Scopus
    View full text
    Copyright © 2009 Elsevier Ltd. All rights reserved.