Elsevier

Pattern Recognition

Volume 48, Issue 7, July 2015, Pages 2304-2317
Pattern Recognition

Computer aided erosions and osteophytes detection based on hand radiographs

https://doi.org/10.1016/j.patcog.2015.01.018Get rights and content

Highlights

  • We introduce a new algorithm of local segmentation based on a region growing.

  • We define a new shape description language for syntactic pattern recognition.

  • We present a multi-stage algorithm, which expands on those created so far.

  • We designed, implemented and tested an automatic computer system which detects erosions and osteophytes (the system operator has to only load the image and mark the area of the analysis).

Abstract

In this paper we present a computer system to detect erosions and osteophytes from hand radiographs, the most common symptoms of rheumatic diseases. The designed, implemented and verified algorithm uses techniques of image processing, image analysis and pattern recognition. In the stages of image processing and image analysis, the locations of metacarpal bones, the outlines of finger bones, the locations and outlines of joints and finally the borders of joint surfaces are identified. In the pattern recognition stage, a shape description language is used for each border of the joint surface to detect the locations of erosions and osteophytes on hand radiographs. The presented algorithm expands on those known from the literature, because besides erosions it also detects osteophytes. Moreover, in contrast to previous systems, it analyses proximal interphalangeal joints and distal interphalangeal joints. The obtained results are satisfactory and very promising. The joints are successfully located in 98.3% of cases. The average mean distance between the borders pointed out by radiologists and obtained from the system varies between 0.094 mm and 0.157 mm, while the sensitivity and the specificity equal around 70% in most of the cases. Therefore, it can become a basis for the diagnosis of certain diseases.

Introduction

Within the scope of rheumatology and diagnostic radiology, it is essential to distinguish between inflammatory and non-inflammatory diseases. To give a diagnosis at an early stage of a disease, radiographs are taken of the patient׳s hands and the symmetric joints are analysed. The analysis is conducted in order to detect the lesions, which are taken into consideration during diagnosis, together with other tests (e.g. blood tests). However, due to the number of hand joints, such a standard analysis is exceedingly complicated and time consuming. To minimize the time spent on this analysis and to make a radiograph examination more frequent and precise, this process should be automated.

In this paper we present a computer system to detect erosions and osteophytes from hand radiographs, the most common symptoms of rheumatic diseases. The designed, implemented and verified algorithm uses the techniques of image processing, image analysis and pattern recognition. In the stages of image processing and image analysis, the locations of metacarpal bones, the outlines of finger bones, the locations and outlines of joints and finally the borders of joint surfaces are identified. In the pattern recognition stage, a shape description language is used for each border of the joint surface to detect the locations of erosions and osteophytes on hand radiographs.

The main achievements described in this paper:

  • introducing a new algorithm of local segmentation based on a region growing (see Section 4.1.2),

  • defining a new shape description language for syntactic pattern recognition (see Section 4.3.2),

  • presenting a multi-stage algorithm, which expands on those created so far (see Sections 4 and 5),

  • designing, implementing and testing an automatic computer system which detects erosions and osteophytes (the system operator has to only load the image and mark the area of the analysis – see Section 4).

Detecting osteophytes is innovatory due to the fact that in the research results published so far, the borders of joint surfaces were used only for detecting joint space narrowing and erosions.

The test set consists of the left and right hand radiographs of 60 patients, taken in the anterior posterior position. Among those, 20 are healthy, 20 suffer from degenerative diseases and 20 suffer from inflammatory diseases. Such a test set was used to conduct numerous analyses whose results were investigated.

This paper is organized in the following manner. In Section 2, the medical basis of the topic is outlined. In Section 3, attempts to detecting joint space narrowing and erosions using the existing methods are very briefly presented. Detecting erosions and osteophytes using a special algorithm is presented in Section 4. The obtained results and the discussion are presented in 5 Experiments, 6 Discussion, respectively.

Section snippets

Some data concerning the medical aspects of the paper

Despite the fact that magnetic resonance imaging (MRI) shows the greatest sensitivity for detecting and monitoring bone erosions [16], conventional or digital radiography of the hand is the most commonly used imaging method in diagnosis, as well as monitoring disease progression and the treatment response in case of the patients with rheumatic musculoskeletal diseases [3]. This is due to the fact that radiography is widely available, inexpensive and easy to perform, as well as valuable in

Detecting joint space narrowing and erosions using the existing methods

There are a number of papers concerning the topic of computer aided rheumatoid diagnosis based on hand radiographs. Some of them focus on segmentation of the hand radiographs – see [18], [23], [24], [28], [30], [36], [41]. Others relate to identifying the joint surface borders and detecting joint space narrowing – see [10], [11], [34], [35], [36], [38], [39], [25]. However, according to Peloschek et al. [32], only a few of them relate to detecting erosions – see [26], [27].

The segmentation of

Detecting erosions and osteophytes using a special algorithm

The presented algorithm can be employed to detect erosions and osteophytes on hand radiographs. It is an automatic technique; however, the examined hand has to satisfy the assumptions about the hand anatomy. These assumptions relate to the average length and width of the bones [21], the average width of the soft tissue in the finger regions [14], as well as the proportions of the bones [17]. Obviously, in some cases, such assumptions are not met. For example, the patient may have a lack of the

Validation set

In total, 1440 joint surfaces were analysed in the left and the right hand radiographs of 60 patients (60 patients×2 hands×2 fingers of interest×3 joints×2 joint surfaces), acquired through the offices of the University Hospital in Kraków, Poland. Among all the patients, 20 were healthy, 20 suffered from degenerative diseases and 20 suffered from inflammatory diseases. They were selected by a rheumatologist who accepted the radiographs of the whole hand taken in the anterior posterior position

Discussion

There are many algorithms determining joint space narrowing and a few locating erosions (see Section 3). Until now however, there has been no automatic algorithm for determining both erosions and osteophytes in MCP, PIP and DIP joints. Such a system for aiding rheumatoid diagnosis is presented in this paper. It starts with locating all the joints of interest and identifying the borders of their surfaces. These borders are then used to determine the locations of erosions and osteophytes.

Conflict of interest

None declared.

Bartosz Zieliński received the master degree in computer science from the Jagiellonian University in 2007. He received Ph.D. in 2012 in computer science from the Institute of Fundamental Technological Research, Polish Academy of Sciences. He works at the Computer Science and Computer Mathematics Institute of the Jagiellonian University. His scientific interests are artificial intelligence, computer vision and grammatical inference.

References (41)

  • D. Bachmann et al.

    Radiological Atlas of Rheumatoid Diseases

    (1994)
  • M. Bielecka et al.

    A shape description language for osteophytes detection in upper surfaces of the metacarpophalangeal joints

    Adv. Intell. Soft Comput.

    (2011)
  • M. Bielecka, A. Bielecki, M. Korkosz, M. Skomorowski, W. Wojciechowski, B. Zieliński, Application of shape description...
  • M. Bielecka, A. Bielecki, M. Korkosz, M. Skomorowski, W. Wojciechowski, B. Zieliński, Modified Jakubowski shape...
  • M. Bielecka, M. Skomorowski, B. Zieliński, A fuzzy shape descriptor and inference by fuzzy relaxation with application...
  • A. Bielecki, M. Korkosz, W. Wojciechowski, B. Zieliński, Identifying the borders of the upper and lower...
  • J. Böttcher et al.

    Digital X-ray radiogrammetry combined with semiautomated analysis of joint space widths as a new diagnostic approach in rheumatoid arthritisa cross-sectional and longitudinal study

    Arthritis Rheum.

    (2005)
  • J. Böttcher et al.

    Computerized digital imaging techniques provided by digital X-ray radiogrammetry as new diagnostic tool in rheumatoid arthritis

    J. Digit. Imaging

    (2006)
  • O. Bretscher

    Linear Algebra with Applications

    (1995)
  • A. Buryanov et al.

    Proportions of hand segments

    Int. J. Morphol.

    (2010)
  • Cited by (0)

    Bartosz Zieliński received the master degree in computer science from the Jagiellonian University in 2007. He received Ph.D. in 2012 in computer science from the Institute of Fundamental Technological Research, Polish Academy of Sciences. He works at the Computer Science and Computer Mathematics Institute of the Jagiellonian University. His scientific interests are artificial intelligence, computer vision and grammatical inference.

    Marek Skomorowski received his Ph.D. in 1984 in computer science from the AGH University of Science and Technology, Kraków, Poland. He works at the Computer Science and Computer Mathematics Institute of the Jagiellonian University, Kraków, Poland. His scientific interests include pattern recognition, modelling and computer simulation.

    Wadim Wojciechowski is graduated from the Medical Academy in Karaganda (Kazakhstan) with a Diploma of a Physician. He finished his complete training scheme in Radiology included practical and theoretical exams in 2002. Ph.D. Thesis title of qualification awarded: The value of computed tomography virtual bronchoscopy in evaluation tumours of bronchial tree. He is the President of Polish Medical Society of Radiology in Krakow.

    Mariusz Korkosz is graduated in medicine from the Jagiellonian University Medical College in 1991 and finished internal medicine (1996) and rheumatology (1999) training. He completed his Ph.D. thesis on the osteoporosis evaluation in ankylosing spondylitis in 2001 from the Jagiellonian University. Since 2007 he is the head of Division of Rheumatology at Department of Internal Medicine and Gerontology at the University Hospital in Kraków. His research interests centre around ankylosing spondylitis including diagnostic tools and treatment intervention.

    Kamila Sprężak is graduated in medicine from the Jagiellonian University Medical College in 2009. She is working in the Department of Radiology at the University Hospital in Kraków as a radiologist in training. Her professional interests centre around musculoskeletal imaging.

    View full text