An ontology-based fuzzy decision support system for multiple sclerosis

https://doi.org/10.1016/j.engappai.2011.02.002Get rights and content

Abstract

The use of Magnetic Resonance (MR) as a supporting tool in the diagnosis and monitoring of multiple sclerosis (MS) and in the assessment of treatment effects requires the accurate determination of cerebral white matter lesion (WML) volumes. In order to automatically support neuroradiologists in the classification of WMLs, an ontology-based fuzzy decision support system (DSS) has been devised and implemented. The DSS encodes high-level, specialized medical knowledge in terms of ontologies and fuzzy rules and applies this knowledge in conjunction with a fuzzy inference engine to classify WMLs and to obtain a measure of their volumes. The performance of the DSS has been quantitatively evaluated on 120 patients affected by MS. Specifically, binary classification results have been first obtained by applying thresholds on fuzzy outputs and then evaluated, by means of ROC curves, in terms of trade-off between sensitivity and specificity. Similarity measures of WMLs have been also computed for a further quantitative analysis. Moreover, a statistical analysis has been carried out for appraising the DSS influence on the diagnostic tasks of physicians. The evaluation has shown that the DSS offers an innovative and valuable way to perform automated WML classification in real clinical settings.

Introduction

Multiple sclerosis (MS) is an autoimmune inflammatory disease of the Central Nervous System, which causes the damage or destruction of myelin surrounding nerve fibers. In more detail, such a disease is characterized by multiple demyelinated lesions involving the brain and spinal cord, that interrupt communications between the nerves and the rest of the body (Compston and Coles, 2008).

According to the National Institute of Neurological Disorders and Stroke, about 250.000–350.000 people in the United States have been diagnosed with MS. Worldwide, the incidence of MS is approximately 0.1%. Northern Europe, the northern United States, southern Australia and New Zealand have the highest prevalence, with more than 30 cases per 100.000 people.

MS has an unpredictable clinical course and it is more common in women and in Caucasians. The average age of onset is between 20 and 40, but the disorder may develop at any age (Rosati, 2001). It produces neurological dysfunctions, such as numbness, impaired vision, loss of balance, weakness, bladder dysfunction and psychological changes. Many MS cases evolve over a long period (20–30 years) with remissions and exacerbations, but, in almost half of all cases, it relentlessly progresses to severe disability and premature death (Kidd, 2001).

Diagnosis of MS is based on the principle of dissemination in both time and space. Recent criteria state that patients should experience two attacks of such dysfunctions, occurring at different points of time and affecting different parts of the Central Nervous System. Many years may elapse between the first attack and the second one, and not all the patients who experience a first attack develop MS. Nevertheless, such attacks are extremely variable, often quite subtle; hence, they can lead to a suspicion of disease, but, in many cases, they cannot be sufficient on their own for the diagnosis. In such a sense, recently, Magnetic Resonance Imaging (MRI) has been applied as a supporting tool in MS diagnosis, enabling the visualization of cerebral MS lesions, both in clinically suspected cases and in silent ones (Miller et al., 2004).

Furthermore, the lack of laboratory markers for MS activity, progression and remission has brought much interest to the application of MRI, especially as a monitoring tool both in the course of MS and in the assessment of treatment effects (Miller, 1994, Miller et al., 1998, Filippi et al., 1995). As a matter of fact, brain Magnetic Resonance (MR) images allow to characterize MS lesions in both space and time, i.e. providing information about their number, size and spatial distribution for every single study and, moreover, highlighting changes among studies performed at different times. The use of MR images as MS marker requires the expert’s knowledge and intervention to classify MS lesions; nevertheless, manual classification is a very thorny and time-consuming task due to the huge amount of MR images to be examined and the variable number, size and spatial distribution of MS lesions per image.

Operator-assisted techniques, such as local thresholding, have been successfully employed (Filippi et al., 1996), but they are time-consuming and require operator intervention since they are monoparametric, i.e. based only on MR signal intensity data. As a consequence, they are strongly associated with some degree of variability due to the operator intervention and, in this respect, the use of an automated system for supporting neuroradiologists to detect MS lesions could be undoubtedly advantageous. Successful methods assessing multiple parameters have been developed in the last decade, in which automated classification of MS lesions is characterized by high specificity and sensitivity (Alfano et al., 2000, Akselrod-Ballin et al., 2009, Anbeek et al., 2004, Freifeld et al., 2009, Khayati et al., 2008, Sajja et al., 2006, Wei et al., 2002, Wels et al., 2008, Zijdenbos et al., 2002).

Although all these methods have a high classification accuracy, they suffer from two important limitations. First, most part of them does not reproduce the real decision-making process performed by neuroradiologists, but implements advanced algorithms that are often too complex to be understood by a non-technical audience such as the physicians. Indeed, their goal is to maximize the classification accuracy without providing comprehensible outcomes to physicians, in terms of a clear and semantic description of the generated results. As a result, they are considered by physicians as black boxes that simply generate answers with no further explanation. Differently, neuroradiologists would like to have some insight as to how a classification method works and derives its outputs.

Second, lot of the methods existing in literature rely on mathematical models based on thresholding to classify MS lesions. Hence, they neither take into account the fuzziness of input data nor reproduce the expert decision-making process applied in a vague-laden domain such as medicine. As a matter of fact, the decision-making model every trustworthy physician has in mind to perform heuristic diagnosis is often pervaded by uncertainty and vagueness. Expert knowledge therefore abounds with imprecise formulations that are not a consequence of rhetorical inability, but an intrinsic part of expert knowledge acquired through laborious experience (Steimann and Adlassnig, 2000). Any formalism disallowing uncertainty, such as crisp mathematical models, is therefore inapt to capture this knowledge (Adlassnig, 1998, Steimann and Adlassnig, 2000) and can represent an unrealistic oversimplification of reality, leading to possible wrong interpretations when compared to a direct observation.

Therefore, a system able to detect MS lesions by handling fuzziness and providing interpretation for the classification outputs would be of great clinical value.

In such a sense, an ontology-based fuzzy decision support system (DSS) has been developed in order to automatically support neuroradiologists in the classification of a type of MS lesion, i.e. white matter lesion (WML). Such a DSS encodes high-level medical knowledge elicited from experts in terms of ontologies and fuzzy rules and applies such a knowledge in conjunction with a fuzzy inference engine to classify WMLs and to obtain a measure of their volumes. Specifically, ontologies are used to represent the semantic structure of the expert's knowledge and to provide a comprehensible formulation of the generated outcomes. Fuzzy logic is used to handle fuzziness of input dataset and reproduce the expert's decision-making process to classify WMLs with further capability of attributing a confidence measure to the output results. The performance of the DSS is quantitatively evaluated on 120 patients affected by MS. Moreover, a statistical analysis is carried out to appraise to the extent the DSS has an influence on the diagnostic tasks of physicians, and whether this influence can be quantified.

Section snippets

Background

The proposed DSS relies on a knowledge-based approach integrating two knowledge-representation techniques, namely ontologies and fuzzy logic, shortly outlined in the following sub-sections.

Data description

The dataset used in this study includes 120 patients with clinically definite MS. Their age range is between 20 and 63 years. The clinical data for this study were collected, in an anonymized form, in the Department of Bio-Morphological and Functional Sciences of the University of Naples “Federico II”.

MR brain images were acquired on a 1.5 T Philips Achieva scanner. All patients had the same MR protocol consisting in two spin-echo sequences with the following parameters: (1) spin echo sequence:

Method

The methodology proposed in the work relies on the construction of a knowledge-based DSS responsible of (i) encoding the high-level, specialized medical knowledge elicited from clinical experts; (ii) making inferences through the application of this knowledge to the features described in Section 3.2; (iii) drawing conclusions with the final aim of supporting the clinicians' everyday practice.

Essentially, the methodology can be described in terms of three stages, namely Knowledge Elicitation,

Evaluation

Review papers (Trivedi et al., 2002, Kaplan, 2001, Tierney, 2001, Johnston et al., 1994) point out that the vast majority of DSSs presented in literature are evaluated in a somewhat artificial context, i.e. as stand-alone software systems designed to operate in parallel to, but not necessarily in support of the physician.

Moreover, they are predominantly evaluated depending on their diagnostic performance, in terms of accuracy, sensitivity, specificity, or complete ROC curves, but a wide array

Discussion and conclusions

The ontology-based fuzzy decision support system provides a knowledge-based method to automatically support neuroradiologists in the WML classification with high sensitivity and specificity for all the patient categories.

The DSS encodes high-level medical knowledge elicited from experts in terms of ontologies and fuzzy rules and applies such a knowledge in conjunction with a FIRE inference engine so as to classify WMLs and determine their volumes. Its output is a textual report containing a

Acknowledgments

The authors are deeply grateful to the Department of Bio-Morphological and Functional Sciences of the University of Naples “Federico II” for providing them the input dataset. They are also thankful to all the neuroradiologists involved in the study for cooperating both in the definition of the domain-knowledge and in the manual segmentation of the input dataset. Finally, a special acknowledgement is due to Dr. Bruno Alfano, director of the Institute of Biostructure and Bioimaging (IBB) of

References (51)

  • T. Lukasiewicz et al.

    Managing uncertainty and vagueness in description logics for the semantic web

    J. Web Semantics

    (2008)
  • C. Mencar et al.

    Distinguishability quantification of fuzzy sets

    Inf. Sci.

    (2007)
  • W. Tierney

    Improving clinical decisions and outcomes with information: a review

    Int. J. Med. Inf.

    (2001)
  • L. Zadeh

    Fuzzysets

    Inf. Control

    (1965)
  • K.-P. Adlassnig

    A fuzzy logical model of computer assisted medical diagnosis

    Methods Inf. Med.

    (1998)
  • A. Akselrod-Ballin et al.

    Automatic segmentation and classification of multiple sclerosis in multichannel MRI

    IEEE Trans. Biomed. Eng.

    (2009)
  • B. Alfano et al.

    Multiparametric display of spin-echo data from MR studies of brain

    J. Magn. Reson. Imaging

    (1995)
  • B. Alfano et al.

    Automated segmentation and measurement of global white matter lesion volume in patients with multiple sclerosis

    J. Magn. Reson. Imaging

    (2000)
  • J.J. Bartko

    Measurement and reliability: statistical thinking considerations

    Schizophr. Bull.

    (1991)
  • E. Berner et al.

    Effects of a decisionsupport system on physicians’ diagnostic performance

    J. Am. Med. Inf. Assoc.

    (1999)
  • M.S. Brown et al.

    Lung micronodules: automated method for detection at thin-section CT: initial experience

    Radiology

    (2003)
  • M. Filippi et al.

    Quantitative assessment of MRI lesion load in monitoring the evolution of multiple sclerosis

    Brain

    (1995)
  • M. Filippi et al.

    Semi-automated thresholding technique for measuring lesion volumes in multiple sclerosis: effects of the change of the threshold on the computed lesion loads

    Acta Neurol. Scand.

    (1996)
  • J. Fox et al.

    Knowledge acquisition for expert systems: experience in leukaemia diagnosis

    Methods Inf. Med.

    (1985)
  • Cited by (0)

    View full text