Elsevier

Computers in Biology and Medicine

Volume 64, 1 September 2015, Pages 334-346
Computers in Biology and Medicine

Computer-aided diagnosis system based on fuzzy logic for breast cancer categorization

https://doi.org/10.1016/j.compbiomed.2014.10.006Get rights and content

Highlights

  • We develop a new method for categorizing findings in mammograms according to BIRADS.

  • We use fuzzy logic to model a computer-aided diagnosis tool.

  • We construct a fuzzy inference system.

  • The system shows a degree of pertinence of a finding to each BIRADS category.

  • The system achieved an accuracy of 76.7% for nodules and 83.3% for calcifications.

Abstract

Background

Fuzzy logic can help reduce the difficulties faced by computational systems to represent and simulate the reasoning and the style adopted by radiologists in the process of medical image analysis. The study described in this paper consists of a new method that applies fuzzy logic concepts to improve the representation of features related to image description in order to make it semantically more consistent. Specifically, we have developed a computer-aided diagnosis tool for automatic BI-RADS categorization of breast lesions. The user provides parameters such as contour, shape and density and the system gives a suggestion about the BI-RADS classification.

Methods

Initially, values of malignancy were defined for each image descriptor, according to the BI-RADS standard. When analyzing contour, for example, our method considers the matching of features and linguistic variables. Next, we created the fuzzy inference system. The generation of membership functions was carried out by the Fuzzy Omega algorithm, which is based on the statistical analysis of the dataset. This algorithm maps the distribution of different classes in a set.

Results

Images were analyzed by a group of physicians and the resulting evaluations were submitted to the Fuzzy Omega algorithm. The results were compared, achieving an accuracy of 76.67% for nodules and 83.34% for calcifications.

Conclusions

The fit of definitions and linguistic rules to numerical models provided by our method can lead to a tighter connection between the specialist and the computer system, yielding more effective and reliable results.

Introduction

Due to the increasing of the amount of digital medical images produced recently, the activities related to acquisition, management and availability of these images have required efforts of researchers and professionals involved in these processes to study and propose computational methods that provide effective support to daily tasks related to image-based diagnosis. Databases containing patient information as well as clinical data have become usual in healthcare environments. Machine learning and data representation techniques allow us to develop tools for computer-aided diagnosis (CAD) [1], [2], [3], which have been shown to improve the quality of the tasks performed by radiologists [4] and thereby the performance of radiological diagnosis [5].

Methods for CAD usually face a significant obstacle: the so-called semantic gap, which is characterized by the divergence between the results of automated methods and the results expected by users, due to the difficulties of these methods to represent real situations [6], [7]. However, regardless of dealing with a specific context, a purely numeric representation of an image attribute is itself a semantic gap, since it is relatively distant from how the human analyst interprets the image. When analyzing the shape of a breast nodule, the diagnosis is generated in natural language, for example slightly irregular contour. In this case, it is very difficult to set a numeric descriptor for the contour that fits the user׳s semantics. So, we decided to use fuzzy concepts in this context. The concepts of fuzzy logic are designed to handle this kind of situation and can be used to simulate human reasoning in decision support systems [8]. Fuzzy sets handle the separation between classes, commonly limited by ranges of numeric values, by membership grades, replacing abrupt transitions. This approach is suitable to situations in which the boundary between the sets is not well delineated, e.g., the assessment performed by a human being. Fuzzy methods can approach objective knowledge (represented by numerical data) and subjective knowledge (represented by linguistic terms).

The Breast Imaging Reporting and Data System (BI-RADS) standard was developed in collaboration with members and committees of the American College of Radiology (ACR) to guide radiologists and physicians in the evaluation of breast nodules and calcifications [9]. This system aims to unify the terminology used in mammography exams, the categories of findings and their appropriate treatment [10]. The use of the BI-RADS lexicon is highly recommended in the decision-making process as it follows a natural language structure: “if the nodule is round, has well-defined margins and low density, it is more likely to be classified as benign”. This kind of structure can be properly modeled through fuzzy sets. Fuzzy logic is also suitable to represent the divergence between radiologists. In clinical practice, the same image can be analyzed in different ways by each expert. Fuzzy inference models can encompass inter- and intra-observer variations.

In related works, some approaches have been proposed using fuzzy logic to assist the diagnosis of breast lesions. In Ref. [11] the authors present a method to distinguish lobulated contours from microlobulated ones. They created membership functions to represent features, such as the number of contour undulations and its size, and in Ref. [12] the authors implemented a fuzzy rule-based method aiming to characterize the shape of nodules in mammograms. In Ref. [13], three distinct regions in mammograms are compared: regions containing tumors; regions containing calcifications and regions containing no injuries. The authors proposed a fuzzy inference system based on the intensity of the pixels of these three regions, inferring the type of lesion (tumor or calcification) or the absence of lesion. In contrast, the methodology used in our project, in addition to treating attributes as fuzzy sets, also provides a fuzzy output, generating a quantitative response which is standardized by the BI-RADS system.

Fuzzy logic is also applied in Ref. [14], which presents an iterative segmentation algorithm to automatically identify regions of interest (ROIs) from nodules considering the breast tissue density and applying a fuzzy classifier for diagnosis. Concerning breast cancer treatment, Garibaldi et al. [15] present a method based on non-stationary fuzzy sets to model the need of recommendation of post-operative treatment considering clinical variables as attributes. This approach yielded better results when compared to standard fuzzy inference systems.

Another applications to support medical diagnosis can be found in Ref. [16] to help the analysis of malformations of the cerebral cortex. The input variables are based on the expert׳s knowledge. In Refs. [17], [18], fuzzy cognitive maps are used to support diagnosis. In computer vision, fuzzy systems can also be used to describe intrinsic attributes of images [19], [20]. Moreover, different studies have applied fuzzy-based techniques to medical applications [21], [22], [23], [24], [25], [26].

This project aims to investigate the effectiveness of fuzzy representation of the evaluation parameters of breast lesions. For this purpose, we have developed a fuzzy system to provide an automated quantitative analysis of breast nodules and calcifications, based on image visual assessment parameters supplied by the users. This system is based on a new method proposed in Ref. [27] for designing of membership functions in fuzzy sets. This algorithm uses training sets to determine the data distribution for each class and establishes the design of the membership functions to be used by the fuzzy inference model. The system provides a second opinion to aid the radiologist (or other clinicians) to define the diagnosis and the patient׳s mammography classification. For this computerized analysis, we have used the main attributes related to mammographic findings in the reports of the nodules and calcifications. We have also used the BI-RADS evaluation categories and their corresponding prognosis. We have designed and implemented a CAD tool for generating BI-RADS diagnosis. We have evaluated our proposed methodology by the effectiveness in generating results more closer to the ones expected by users. Furthermore, the use of parameters and rules defined in terms of natural language should contribute to an effective adoption of CAD systems by the radiologists.

This paper is organized as follows: Section 2 presents the technical background related to fuzzy logic and the BI-RADS standard; Section 3 describes the proposed methodology for the representation of the BI-RADS attributes as fuzzy sets as well as their evaluation categories; Section 4 presents the experiments and the consequent results; Section 5 presents discussions and Section 6 presents conclusions.

Section snippets

Fuzzy inference model

Fuzzy set theory distinguishes from the classical set theory since it concerns vague aspects related to information. This theory is less restrictive and can be more suitable to handle information given by humans. It extends the classical binary logic (0 or 1) to the continuous domain (0 to 1), through a gradual transition between pertinence and non-pertinence of an element within a set [8], [28]. There are several fuzzy inference models. The most known is the Mamdani model, where the semantic

Material and methods

The proposed methodology is described in the following sections.

Design of membership functions

In the evaluation of nodules, we obtained 126 values of malignancy degrees for the shape coefficient and for the density coefficient. For the margin coefficient we obtained 180 values of malignancy degrees due to the fact that the radiologist could select more than one characteristic to this attribute during his/her analysis. After applying the Fuzzy-Omega algorithm, two or three membership functions were generated for each attribute. However, subsequent tests indicated that the use of only two

Discussion

The implemented system and the results presented in the previous section lead us to carry out some comments on certain issues.

Conclusion

This paper presents a method for the fuzzy characterization of the main linguistic terminology descriptors used in the evaluation of breast nodules and calcifications, as well as the BI-RADS categories associated with them, for mammographic images. The proposed system was developed in collaboration with domain specialists. Computational methods supported the steps of the system modeling and implementation. The result is a CAD system that suggests to the user a BI-RADS category following the

Summary

CAD systems based on medical imaging aim to assist and improve accuracy of diagnosis through the use of a computer, indicating suspicious areas in the image as well as abnormalities. This assistance can be very important since the radiologist׳s diagnosis is based on subjective analysis and personal variations, such as visual fatigue or distraction, image quality and overlay structures, among others. Throughout its evolution, research projects addressing methods for CAD encountered a significant

Conflicts of interest statement

None declared.

Acknowledgments

The authors thank the radiologists of School Hospital of Medicine of University of São Paulo at Ribeirão Preto, Brazil, for the analysis of the selected images.

This research was supported by the São Paulo Research Foundation (FAPESP) (Grant no. 07/59862-0).

Gisele H.B. Miranda received the B.Sc. degree in Biomedical Informatics from the University of São Paulo, Brazil, in 2009 and the M.Sc. in Bioinformatics also from the University of São Paulo, Brazil, in 2011. Presently she holds a fellowship in a medical systems company. Her research interests include image processing and analysis, complex networks and computer-aided diagnosis.

References (42)

  • W.K. Moon

    Computer-aided diagnosis of breast masses using quantified BI-RADS findings

    Comput. Methods Progr. Biomed.

    (2013)
  • W.C. Shen

    Breast ultrasound computer-aided diagnosis using BI-RADS features

    Acad. Radiol.

    (2007)
  • G. Giannakopoulou

    Downgrading BIRADS 3 to BIRADS 2 category using a computer-aided microcalcification analysis and risk assessment system for early breast cancer

    Comput. Biol. Med.

    (2010)
  • K. Doi

    Current status and future potential of computer-aided diagnosis in medical imaging

    Br. J. Radiol.

    (2005)
  • K. Doi

    Diagnostic Imaging over the last 50 years: research and development in medical imaging science and technology

    Phys. Med. Biol.

    (2006)
  • J. Tang

    Computer-aided detection and diagnosis of breast cancer with mammography: recent advances

    IEEE Trans. Inf. Technol. Biomed.

    (2009)
  • T.K. Yin et al.

    A computer-aided diagnosis for locating abnormalities in bone scintigraphy by a fuzzy system with a three-step minimization approach

    IEEE Trans. Med. Imaging

    (2004)
  • J.G. Cooper

    Does computer-aided clinical decision support improve the management of acute abdominal pain? A systematic review

    Emerg. Med. J.

    (2011)
  • M. Zolnoori et al.

    Application of intelligent systems in asthma disease: designing a fuzzy rule-based system for evaluating level of asthma exacerbation

    J. Med. Syst.

    (2012)
  • P. Welter et al.

    Bridging the integration gap between imaging and information systems: a uniform data concept for content-based image retrieval in computer-aided diagnosis

    J. Am. Med. Inf. Assoc.

    (2011)
  • L.A. Zadeh

    Fuzzy logic and approximate reasoning

    Synthese

    (1975)
  • Cited by (91)

    • Optimum-path forest and active learning approaches for content-based medical image retrieval

      2022, Optimum-Path Forest: Theory, Algorithms, and Applications
    • Use of deep learning approaches for the prediction of diseases from medical images

      2024, Medical Imaging Informatics: Machine learning, deep learning and big data analytics
    View all citing articles on Scopus

    Gisele H.B. Miranda received the B.Sc. degree in Biomedical Informatics from the University of São Paulo, Brazil, in 2009 and the M.Sc. in Bioinformatics also from the University of São Paulo, Brazil, in 2011. Presently she holds a fellowship in a medical systems company. Her research interests include image processing and analysis, complex networks and computer-aided diagnosis.

    Joaquim C. Felipe received the B.Sc. degree in Mechanical Engineering from the University of São Paulo, Brazil, in 1986, the M.Sc. degree in Computer Science from the Federal University of São Carlos, Brazil, in 2000 and the Ph.D. degree in Computer Science from the University of Sao Paulo, Brazil, in 2005. He is currently a full Professor in the Computing and Mathematics Department of University of São Paulo at Ribeirão Preto, Brazil. His research interests include medical imaging, image processing and analysis, computer-aided diagnosis and bioinformatics.

    1

    Tel.: +55 16 3602 4359; fax: +55 16 3633 9949.

    View full text