Automatic Vertebral Fracture Assessment System

Automatic Vertebral Fracture Assessment System (AVFAS) for Spinal Pathologies Diagnosis Based on Radiograph x-ray Images

Abstract.

Nowadays, medical imaging has become a major tool in many clinical trials. This is because the technology enables rapid diagnosis with visualization and quantitative assessment that facilitate health practitioners or professionals. Since the medical and healthcare sector is a vast industry that is very much related to every citizen`s quality of life, the image based medical diagnosis has become one of the important service areas in this sector. As such, a medical diagnostic imaging (MDI) software tool for assessing vertebral fracture is being developed which we have named as AVFAS short for Automatic Vertebral Fracture Assessment System. The developed software system is capable of indexing, detecting and classifying vertebral fractures by measuring the shape and appearance of vertebrae of radiograph x-ray images of the spine. This paper describes the MDI software tool which consists of three main sub-systems known as Medical Image Training & Verification System (MITVS), Medical Image and Measurement & Decision System (MIMDS) and Medical Image Registration System (MIRS) in term of its functionality, performance, ongoing research and outstanding technical issues.

Keywords:

spine x-ray pathologies, image processing, ASM model, B-spline, AVFAS, MIMDS, MITVS, MIRS.

1 Introduction

The role of medical imaging in diagnostic features is innumerable and computer assisted imaging is now the challenging work of many researchers. Medical images play important role to detect functional information of the body part for diagnosis, medical research and education [1]. Modern standards such as Digital imaging and communication (DICOM) [2] and picture archival and communication systems (PACS) [3], enable easy storage and transportation of these images and thus, increase interoperability. Medical images of diverse modalities such as computerized tomography (CT), magnetic resonance image (MRI), positron emission tomography (PET), and signal photon emission computed tomography (SPECT) ultrasound, microscope pathology and histology images are generally complex in nature and require extensive image processing techniques for computer aided diagnosis [4]-[5]. Due to this reason, in most instances physicians or radiologists examine images in the conventional ways based on their individual experiences and knowledge and such practice is exhaustive to the physicians or radiologists. Therefore, there is an urgent need to automate this process. An ideal scenario would be where medical images generated by an imaging station can be automatically compared with the existing images stored in a data base. Then possible abnormalities can be identified and suggested by the system. With such capabilities, role of medical imaging would expand and the focus could shift from image generation and acquisition to more effective post processing, organization and interpretation. To approach this kind of automation two main technologies, namely image retrieval and image registration need to be addressed and integrated in a computer assisted diagnostic environment. Image retrieval in the field of medical domain has been one of the most exciting and fastest growing research areas over the last decade [4], [5].

2 Objectives of the research

Pathologies found on the spine x-ray images that are of interest to the medical researchers are generally expressed along the vertebral boundary. These pathologies include anterior osteophytes (AO), intervertebral disc degeneration and resulting disc space narrowing, subluxation and spondylolisthesis. However, the work presented in this paper will only focus on anterior osteophytes (AOs). The AOs or Osteoporosis is a tricky bones disease in which one may not even know that they have it until they break a bone. When this occurs it means that the disease is already well advanced [6].Therefore, the long-term goal of this work is to develop a computer-aided system for screening and classifying radiograph x-ray images of the spine presence with/without bone fractures. To do so, we have to the followings:

· Acquire and annotate a large database of spine images (cervical /lumbar).

· Develop algorithms that automatically locate and measure all the vertebrae in the image.

· Develop new techniques for classifying and matching vertebrae.

· Produce a user-friendly software tool that can be used by regular clinicians.

· Develop a fully automatic mode for use in large-scale clinical trials and epidemiological studies

3 Current AVFAS Design Prototype System

The AVFAS prototype software is implemented in MATLAB version 7. Fig. 1 shows the design of the main interface of AVFAS prototype screening system (MIMDS, MIRS and MITVS) whereas Fig. 2 shows the block diagram of AVFAS system component.

In this current system the user manipulates the main GUI tools to select either one of the three sub- system interfaces. 1-Medical Image Measurement & Decision System (MIMDS), 2- Medical Image Training & Verification System (MITVS), 3-Medical Image Registration System (MIRS). The MITVS system performs the following functions:

· Image pre-preprocessing comprising the region-of-interest (ROI)-localization, and ROI enhancement.

· Modeling process where models extracted from the vertebral boundary shape determination based on active shape modeling (ASM) and B-spline enhancement in terms of list of points selected are built and trained.

· Shape boundary analysis based on extracted features and morphometric measurements

· Build and train classifier models to detect and classify various AOs classes based on the medical classification and grading schemes provided

· Build models for shape query and retrieval based on matching and similarity techniques.

· Establish database reference according to the medical expert observation using Microsoft access database

The MIMDS system tasks are listed as follow:

· Image pre-preprocessing task as before which involves ROI-localization and ROI enhancement

· Evaluating the locate accuracy of the shape models extracted automatically from the modeling stage

· Detecting and diagnosing the pathologies founded based on their classification schemes provided by the medical expert

· Perform matching and similarity between a query shape and those stored in database

· Provide final report for medical use.

· Provide other function such (zooming in and out, image viewer etc)

Finally, the MIRS system main function is to record all related information about the medical diagnostic imaging procedure. The information includes the image database source, its coordinate system and its origin, the anatomy, region of the anatomy, etc Details of patient medical data and physician planning, date and timing etc. are also part of the information stored in the MIRS. A unique object identifier is used to archive and store information of segmented objects produce during diagnosis.

4 MITVS System Overview

The MITVS inputs are x-ray spine images (cervical/lumbar) selected from the database and its main functions are, pre-processing stage, modeling stage, boundary shape determination, training and testing classifier system, reporting and database reference establishment are briefly described as follows.

4.1 Pre-Processing Stage

In general the quality of computer segmentation is affected by three important factors that is, first factor is region of interest (ROI), the second is image quality, the third image size/resolution in spine x–ray images, pre-processing stage plays important role in the AVFAS system including x-ray images acquisition, ROI-localization, ROI-enhancement and be outlined as follows

4.2 Image Acquisition

Medical images are multi-modal, where each modality reveals anatomical and/or functional information of different body parts and has its own set of requirements such as file format, size, spatial resolution, dimensionality, and image acquisition and production techniques [12],[13]. In this work, more than 500 cervical and lumbar spine image were selected from The Lister Hill Center for Biomedical Communications, an intramural research and development division of the U.S. National Library of Medicine (NLM), who maintains a digital archive of 17,000 cervical and lumbar spine images collected from the second National Health and Nutrition Examinations Survey (NHANES II) conducted by the U.S National Center for Health Statistics (NCHS) [7].

4.3 Region Localization

Initially, a manual technique is being developed to select the ROI from the x-ray images. Using the mouse, two separate landmarks points are chosen to mark the object position within the region of interest. Automatically, the system responds by returning a display of those two landmarks points associated with the x-ray ROI images desired.

4.4 ROI-Enhancement

The image quality resulting from the ROI selection in spine x-ray image is poor with ambiguous vertebral boundaries, making a reliable segmentation a challenging task. In order to detect the presence of the vertebrae and obtain a good detection, it is necessary to enhance the localized region. Various enhancement techniques such adaptive histogram-based equalization, adaptive gamma value and adaptive contrast enhancement were implemented and evaluated based on the threshold approach and visual inspection. The cervical vertebrae C1, C2 and C7 are basically left out and not considered because these structures are often not visible on the radiograph and hence it is difficult to characterize them. However, lumbar vertebrae of (L1/L2/L3/L4/L5) can be clearly observed. Pre-processing steps for both (cervical/lumbar).

5 Boundary Shape Determination

Shape is an important characteristic for describing pertinent pathologies in various types of medical image and it is a particular challenge regarding vertebra boundary segmentation in spine x-ray images. It was realized that the shape representation method would need to serve the dual purpose of providing a rich description of the vertebra shape while being acceptable to the end user community consisting of medical professionals. ASM model [8] has been used to obtaining a boundary shape determination of a shape vertebra in terms of a list of points, two schemes list points were used at this stage. A 9-anatomical points shape (9-APS) assigned by an expert that is indicative of the pathology found to be consistent and reliable in detecting the image collection, where B- spline technique for smoothing have been implemented successfully to accurately and robustly locate vertebrae in lateral spinal x ray images based on the 9-anatomical pseudo points shape (9-APPS) which is constructed of 27 pseudo points on the vertebrae between the selected 9-anatomical points.

The vertebrae boundary points are extracted as (x ,y) coordinates in the image space. These are then presented in a suitable form for archiving, indexing, classification, similarity and matching .The segmentation results from the ASM technique based on the two schemes indicated that include the template and segmentation object are stored with unique object identifier. Each stored object records the information about the vertebrae and the images. Fig.4 below shows the block diagram of the shape boundary determination which consists of the Training Stage, Finding the Weight and 3-Verifying Stage) using active shape model associated with B-spline algorithm.

6 Boundary Shape Analysis

At this stage, three schemes of boundary shape analysis are being implemented. The first scheme is the shape analysis based feature vector extraction where the second is the shape analysis based on morphometric measurement including angles and intra-bone ratio measurement. The third and last scheme is the matching and similarity analysis. The index resulted from this analysis is then used as input for the classifier systems. The aforementioned schemes outlined are briefly described below.

6.1 Analysis Based Feature Vector Extraction

The system provide three techniques based feature vector as input for the classification system where the feature vectors output size invariant base on the each technique applied.

· Feature vector based eigenvector extracted from the ASM model with variant size where size of (18 x1) features based on 9-anatomical points shape and size of (100x1) features based on 9-anatomical pseudo points shape enhanced using b-spline for each unique vertebra .

· Feature vector based Gabor wavelets filter bank extracted of size (150x150) from convolution of each vertebra shape with suitable selection of mask size (7x7) and frequency level between 10 to15 for both shape boundary determination methods

· Feature vector based Gray Level Co-Occurrence Matrix

· Feature vector based Orientation Histogram

6.2 Analysis Based Morphometric Measurement

In order to distinguish between normal and abnormal vertebrae effectively and efficiently, an analysis based on morphometric measurements was determined through experiments. This analysis involves two types of measurements, namely angle measurements and intra-bone ratio measurements. The angle based measurements comprise three angles measurements from a shape that can be used to distinguish AOs are selected and it's called Horizontal angle (HƟ), vertical angle (VƟ) and corner angle (CƟ). The second measurement which is the intra-bone ratio of the anterior, medial and posterior height form vertebra shape are then computed to produce distinguish index of AOs classes.

6.3 Analysis Based Matching and Similarity

Shape matching is an important component in shape retrieval, recognition and classification, alignment and registration, and approximation and simplification. This analysis treats various aspects that are needed to solve shape matching problems. It involves choosing the precise problem, selecting the properties of the similarity measure that are needed for the problem, choosing the specific similarity measure, and constructing the algorithm to compute the similarity.

6.4 Classifier and Matching Models

The classifier system implemented for the MITV system based on shape analysis techniques where,

· K-fold cross validation (KCV), K-nearest neighbor (KNN) and Support vector machine (SVM) for classification based feature vector extraction.

· Fuzzy logic and rule-base models for classification and evaluation based morphometric measurement analysis.

· Minimum average correlation energy (MACE) filter and K-mean clustering for matching and similarity.

Main MITVS Subsystem screening design and functions is.

7 MIMDS System Overview

Currently, this sub-system is a semi-automated system and access to the subsystem is via the graphical interface that allows the users to load medical images organized as field of rational database that can be measured, processed and classified. The system function-modules are briefly described and the MIMDS screen interface system is shown in Fig.6.

7.1 Pre-processing stage

The pre-processing stage methods in MIMDS system are identical to those in the MITVS system which have been previously discussed.

7.2 The AO classes and grading scheme

In order to evaluate the system performance, it is necessary to provide the interpretation of these shapes a priori. There are two common classification schemes for the AOs. One is the Macnab classification [14], [15], [16] where the data set adopt two osteophytes type (claw and traction) indexed from (0-2), in which index 0 reflects normal vertebra whilst index 2 reflects claw. In most cases, claw often possesses a triangular shape and is curved at the tips. A class 2 reflects a traction that consists of spur that protrudes horizontally with moderate thickness and does not curve at the tips. The second classification is the severity grading system [17] where three severity levels are defined namely as slight, moderate and severe. The grades ranging from 1 to 3 is used for assessing the severity level of the AOs. Grade 1 reflects slight i.e. no narrowing or the angle by the osteophyte from the expected normal face of the vertebra is less than 150. Grade 2, on the other hand, reflects moderate severity. This represents the middle narrowing or can also be interpreted as the condition where the angle by the osteophyte from the expected normal anterior face of the vertebra is between 150 to 450. Finally, grade 3 reflects severe condition with sharp narrowing or when the angle by the osteophyte from the expected normal anterior face of the vertebra is equal or greater than 450. By combining the two classification schemes mentioned above, six categories of pathology can be established to assist the radiologist in vertebrae trusting. Radiologist can use the grade assignment examples for the vertebral fractures from the Online Digital Atlas Version 2.0 developed by the Communications Engineering Branch of the National Library of Medicine [7].

7.3 Decision and Reporting

Data set consisting of 276 vertebrae spine of both cervical and lumbar images were used as test dataset. The performance of the developed software system is very encouraging and promising since it can classify and match them correctly.

8 MIRS System Overview

Medical image registration can be defined as an establishment of correspondences between image and physical space [9], [10]. This kind of correspondence has been studied and practiced, such as in monitoring changes of a pathological object, images guided surgery, combining information from multiple imaging modalities and comparing individual's anatomies to standard. However, to achieve realistic application of the registration technology, there are still many technical barriers to overcome [10], [11]. Brief overviews of the approaches including related requirements to render proper registration, the main research focus and approaches are given. The registration system is currently a semi-automated process and it is done via a graphical interface that allows two types of data: patient archive data such as name, sex, age etc and feature classification which includes pathology on the basis of shape, labelling of the segmented structure by proper anatomical name, and classification of the segmented and measurement labelled structures into groups corresponding to high level semantic features of interest.

9 Conclusion

This paper presents the progress of an on-going research effort in medical image retrieval and indexing. The work also points out some promising research directions which are to develop combined system architecture for automatic and efficient diagnosis in a hospital or clinical environment. The need for an accurate and practical image based diagnosis system is growing rapidly as more healthcare professionals utilize image guided diagnosis in their daily activities and research. Although to build an enterprise class system that is reliable and robust is a complex task; the benefits it would bring to the heath community are unimaginable.

Reference

1. H. D. Tagare, C. Jafe, J. Duncan, “Medical image databases: A content based retrieval approach”, Journal of the American Medical Informatics Association, 4 (3), pp. 184-198, (1997).

2. W. Dean Bidgood, Jr., MD, MS, Steven C. Horii, MD, Fred W. Prior, PhD, and Donald E. Van Syckle, “Understanding and Using DICOM, the data Interchange Standard for Biomedical Imaging” J Am Med Inform Assoc. 4 (3), pp. 199–212 ,(1997).

3. T. M. Lehmann, M. O. Gould, C. Thies, B. Fischer, M. Keysers, D. Kohnen, H. Schubert, B. B. Wein, “Content- Based Image Retrieval In Medical Applications For Picture Archiving And Communication Systems”, Medical Imaging, of SPIE Proceedings, Vol. 5033, San Diego, California, USA, (2003).

4. H Müller N Michoux, D Bandon, A Geissbuhler, “A Review of Content-based Image Retrieval Applications Clinical Benefits And Future Directions”, International Journal of Medical Informatics, 73, pp. 1-23,(2004).

5. Richard A. Robb, Wiley-Liss, Biomedical Imaging, Visualization and Analysis, Wiley-Liss (1999).

6. Spine universe website, http://www.spineuniverse.com/article/osteoporosis-4040.html

7. National Library of Medicine , http://archive.nlm.nih.gov/

8. T. F. Cootes and C. J. Taylor,"Slatislica Models of Appearance for Computer Vision" , of Machester, Wolfson Image Analysis Unit, and Biomedical Engineering, University of, MI2 9PT. U.K., Tech. Rep, February (2001).

9. David Hawkes, Registration Methodolgy – Introduction, Chap 2 of Medical Image Registration, edited by Joseph Hajnal, David Hawkes and Derek Hill, (2000).

10. J. B. Antoine Maintz, Max A. Viergever; “A Survey of Medical Image Registration”, Medical Image Analysis,(1998).

11. Joseph V. Hajnal, Derek LG. Hill and David J. Hawkes, A View Of The Future, Chap16 of Medical Image Registration, edited by Joseph Hajnal,David Hawkes and Derek Hill,(2000).

12. Wong STC, Tjandra D, “A Digital Library for Biomedical Imaging on the Internet”, IEEE Communications Magazine 37(1), pp. 84–91, (1999).

13. Richard A. Robb, "Biomedical Imaging Visualization and Analysis", Wiley-Liss, December (1999).

14. M. Heggeness and B. Doherty, “Morphologic study of lumbar vertebral osteophytes," Southern Medical Journal, vol. 91, no. 2, pp. 187-189, Feb. (1998).

14. D. Pate, J. Goobar, D. Resnick, P. Haghighi, D. Sartoris, and M. Pathria, "Traction osteophytes of the lumbar spine: Radiographic-pathologic correlation", Radiology, vol. 166, no. 3, pp. 843-846, March (1988).

15. Macnab, "The traction spur: An indicator of segmental instability", Journal of Bone and Joint Surgery, vol. 53, no. 4, pp. 663{670, June (1971).

16. J. Shou, S. Antani, L. L.R., and G. Thoma, "Evaluating partial shape queries for pathology-based retrieval of vertebra", Proc. of the 8th World Multi-conference on Systemic, Cybernetics and Informatics, vol. 12, pp. 155{160, July 18-21 (2004).

Appendix

Table 1. MITVS system screening and salidation sools

Index

Description

1

ROI-localization method selection

2

ROI-localization resulting display

3

ROI-Enhancement and filtering method selection

4

ROI- Enhancement resulting display

5

Shape boundary determination method selection

6

Shape analysis methods selection

7

Shape matching and similarity methods selection

8

Processing stages display and screening

9

Models measurement and verification display

10

Main screen for x-ray shape boundary determination and verification

11

Training different classifier system

12

Testing and validation different classifier system

Table 2. MIMDS system screening and salidation sools

Index

Description

1

Pre-processing stage toolbar (ROI-selection, ROI-enhancement)

2

Image Acquisition display

3

ROI-enhancement display

4

Shape boundary determination selection method

5

Cervical bones selection (full spine /partial )

6

Lumbar bones selection (full spine/ partial)

7

Select model overlaid evaluation

8

Classification & matching techniques toolbar

9

Select other pathologies (DSN/subluxation / spondylolisthesis)

10

Classification, matching an measurement report for unique vertebra

11

Main screening shows model overlaid verification using model 1

12

Main screening shows model overlaid verification using model 2

Table 3. MIRS system screening and salidation sools

Index

Description

1

Image viewer toolbar

2

Image viewer display

3

Patient information key in

4

Hospital and clinic information key in

5

Date and planning key in

6

Registration type

7

Calendar

[1]1 MITVS sub-system screening & validation tools description on table 1 (Appendix)

[2] MIMDS system screening & validation tools description on table 2 (Appendix)

[3] MIRS system screening & validation tools description on table 3 (Appendix)

Please be aware that the free essay that you were just reading was not written by us. This essay, and all of the others available to view on the website, were provided to us by students in exchange for services that we offer. This relationship helps our students to get an even better deal while also contributing to the biggest free essay resource in the UK!