Skip to main content
European Commission logo print header
Contenu archivé le 2022-12-23

Quantification of magnetic resonance image texture

Objectif

A.BACKGROUND

Since its introduction in the mid-eighties, the growth in the use of magnetic resonance imaging (MRI) in medicine has been spectacular to the present position where it is widely used throughout the USA, Europe and Japan as a primary diagnostic imaging modality. Indeed, Young (1990) has hailed MRI as "possibly the most powerful in-vivo diagnostic tool yet discovered" with "the single most exciting thing about it being its scope". There is considerable opportunity for further advances in utilization and efficacy through making full use of all the quantitative image information available in MRI experiments and this proposal seeks to take advantage of these possibilities.

Reference

Young IR. Magnetic resonance: boundless possibilities or possible boundaries.
Br. J. Radiol 63: 1-13 (1990).

Texture analysis

The "texture" of an area of an image is an ill-defined property - some examples of Brodatz test textures are shown in the figure.

We have an intuitive feeling that it represents the pattern of brightness and darkness and that different textures aid us in recognizing objects in an image or in discriminating between objects. The visual perception of textures has been the subject of extensive psychological study (e.g. Pickett (1964), Callaghan (1989)). A critical point to realise is that the eye-brain complex is only able to appreciate a limited level of (first and second order) complexity in an image (Julesz (1962)).

Computer-based methods of texture analysis were originally developed for use in military applications and a very wide range of techniques are in existence. These methods are able to increase the level of information extracted from the image - this information being inaccessible to human observation. The output is a series of texture parameters representing the texture of the region of interest chosen. Whether this information is useful in discrimination has to be determined for each application studied.

In certain medical imaging techniques, e.g. ultrasound, the visual texture is recognized as conveying diagnostic information. In other methods, e.g. X-ray CAT, or magnetic resonance imaging (MRI), its use is not widely established, except in a few key centres. Medical applications of computerized texture analysis date back to the early 1970s and the explosion in the use of digital techniques has led to the possibility of utilizing texture analysis for most medical imaging modalities. Typical work has been carried out on chest X-rays (e.g. Chien and Fu (1974), Desaga et al (1988), Powell et al (1988)), ultrasound (Lerski et al (1979), Raeth et al (1985), Kratzik et al (1988), Chandrasekaran et al (1989)), X-ray CAT (Coleman et al (1982)) and MRI (Lerski et al (1993), Schad et al (1993)).

To date the studies carried out, although having demonstrated a good measure of success, have not yet been accepted into routine clinical practice. In ultrasound, normal and amyloid myocardial structures have been distinguished (Chandrasekaran et al (1989)), prostatic carcinoma and prostatic hypertrophy separated (Kratzik et al (1988)), and hepatitis differentiated from normal liver - a result which cannot be achieved by conventional ultrasonography (Lerski et al (1979), Schuster et al (1988)). All these successes demonstrate an ability to extract more diagnostic information from the images than can be done by a human observer.

It has been established that, at an individual centre, the application of texture analysis methodologies to MRI data can lead to the separation of pathologies and clinically useful results. However, what is at present not understood, is that some imaging equipments are significantly better at discrimination on the basis of texture than others and, therefore, definition of diagnostic criteria on a multicentre basis is not possible. What is needed (reflected in the aims quoted later) is the systematic

study of the effect of acquisition parameter changes on the texture results. For maximum usefulness, this must, at a minimum, be performed both on a machine which has been shown to give very good results and on one which has given less good discrimination.

The fundamental content of the proposed work programme is the development of appropriate texture methodologies for the MRI task with their subsequent multicentre testing. Also crucial to the success of the project is the addition of different texture methodologies, e.g. structural or fractal approaches. This, also, is demanding of detailed and intensive development work.

Callaghan T C. Interference and dominance in texture segregation: hue, geometric form and line of orientation. Percept. Psychophys. 46: 299 (1989).

Chandrasekaran K, Aylward P E, Fleagle S R et al. Feasibility of identifying amyloid and hypertrophic cardiomyopathy with the use of computerized quantitative texture analysis of clinical echocardiographic data. J. Am. Coll. Cardiol. 13: 832 (1989).

Chien Y P and Fu K S. Recognition of X-ray Picture Patterns. IEEE Trans. Syst. Cybern. SMC-4: 145 (1974).

Coleman A J, Tonge K A and Rankin S C. The Power Spectral Density as a Texture Measure in Computed Tomographic Scans of the Liver. British J. Radiology 55: 601 (1982).

Desaga J F, Dengler J et al. Film digital and texture analysis for digital classification of pulmonary spot opacities. Rontgenblatter 41: 147 (1988).

Julesz B. Visual pattern discrimination. IRE Trans, Info. Theory IT-8: 84 (1962). Katsuragawa S, Doi K et al. Quantitative computer aided analysis of lung texture in chest radiographs. Radiographics 10:257 (1990).

Kratzik C, Schuster E, Hainz A, Kuber W and Lunglmayr G. Texture analysis - a new method for discriminating prostatic carcinoma from prostatic hypertrophy. Urol. Res. 16: 395 (1988).

Lerski R A, Barnett E, Morley P et al. Computer Analysis of Ultrasonic Signals in Diffuse Liver Disease. Ultrasound Med. Biol. 5: 341 (1979).

Lerski R A, Straughan K, Schad L R et al. MR Image Texture Analysis - An approach to Tissue Characterisation. Magnetic Resonance Imaging 11: 873-887 (1993).

Pickett R M. The Perception of a Visual Texture. J Experimental Psychology 68: 13 (1964).

Powell G F, Doi K, Katsuragawa S. Localization of inter-rib spaces for lung texture analysis and computer aided diagnosis in digital chest images. Med. Phys. 15: 581 (1988).

Raeth U, Schlaps D, Limberg B et al. Diagnostic Accuracy of Computerized B-scan Texture Analysis and Conventional Ultrasonography in Diffuse Parenchymal and Malignant Disease. J Clin. Ultrasound 13: 87 (1985).

Schad L R, Bluml S and Zuna I. MR Tissue Characterization of Intracranial Tumors by means of Texture Analysis. Magnetic Resonance Imaging 11: 889-896 (1993).

Schuster E, Knoflach P and Grabner G. Local texture analysis: an approach to differentiating liver tissue objectively. J. Clin. Utrasound 16: 453 (1988).

Image analysis software

A major advance to the possibilities of collaborative image analysis work has been the introduction, within concerted action BMH-CT94-1274, of the software programme NMRWin which runs on PCs of medium to high specification. This has been distributed as a platform on which to perform multi-centre trials of image quantitation and successfully installed at many European centres.

The figure below illustrates a typical screen with MR image displayed and region of interest (ROI) defined. This software allows the display of MR images from a whole series of different commercial equipments. ROI can then be defined, standard texture analysis performed and the resultant texture parameters written to ASCII log file. In the first multicentre trial performed, texture analysis of grey matter in the brain was carried out according to a defined protocol and the results (obtained in seven centres and on four subjects in each) subjected to detailed statistical analysis. The results of this trial showed the need for a more fundamental test object study (see Basic Models section) since the texture parameters measured bore no obvious correlation to the field strength or any other parameter.

Fundamental to this new proposal is the need to further refine and extend this platform. The following areas have been identified where additions to the functionality of NMRWin are crucial to the research programme:

-structural and fractal-based texture analytic methodologies to enhance the possibilities for tissue discrimination. The testing of other techniques, e.g. the Max-Min method;

-the provision of non-linear fitting to calculate T1 and T2 relaxation time maps from acquired image data. Previous calculations of T2 maps employed multi-spin echo data and over-simplified models which neglected RF excitation profiles, diffusion and magnetization transfer and this will be rectified. This will allow multicentre trials in a standardized way of the possibilities for tissue characterization via texture analysis on these maps;

-the use of neural networks to aid classification on the basis of texture.

References

DaPonte J S. Classification of Ultrasonic Image Texture by Statistical Discriminant Analysis and Neural Networks. Computerized Medical Imaging and Graphics 15: 3-9 (1991).

Mitchell O R, Myers C R and Boyne W A. Max-min measure for image texture analysis. IEEE Trans. Computers C-26: 408 (1977).

Basic Models

Fundamental to the overall approach to this proposal is the linkage of the developmental and trial studies to a programme of fundamental studies of the meaning of texture in pathological and tissue structure terms and in test materials. The aim is a development of a detailed understanding of the morphological meaning of tissue texture as measured by MRI. Without this understanding it is not expected that the optimal results can be obtained. A starting point is the study of the optical texture of test materials and the relation of the results to those obtained in MR scanning. This has proved very revealing when applied to reticulated foam and glass bead test textures. Linked to the test object studies and the consideration of optical textures of the same, will be the study of the textures of pathological specimens. These specimens will be obtained on a multicentre basis.

B.OBJECTIVES AND BENEFITS

The fundamental objective of the project is to coordinate and focus recent developments in quantitative magnetic resonance imaging (MRI), in particular texture analysis, to maximize the amount of clinical diagnostic information extracted from this exciting technique. This will improve the effectiveness of this biomedical technology in the European countries.

-The fundamental objective is the progression of image texture analysis as a means of quantification of the subtle differences between normal and diseased tissues by MRI. The recent concerted action "BMH-CT94-1274" has identified several specific areas in which detailed research work is needed to make progress. There is a real and exciting opportunity to preserve the dynamism and vigour of the existing grouping. This proposal would progress new work identified by this concerted action, optimizing efficiency. The work being implemented and planned is only possible in a European context, since multicentre trials of technical and clinical features are crucial to the approach.

-The general clinical focus of all this work will be brain disease and the study of trabecular bone. Both are vital health issues in the European context and techniques able to improve diagnostic ability would be of great economic and social benefit.

-Intermediate measurable targets are readily identified (and are detailed in Section C), i.e. establishment of tissue discriminant functions based on texture for multicentre testing; progression to a clinically useful stage of new techniques of texture analysis in quantitative MRI.

-The expected outcome is the much enhanced usefulness of MR imaging throughout the European Community. There is much useful quantitative data available in this technique which is not fully used if visual interpretation only is applied to the images. Cost effectiveness will be greatly increased and diagnostic capability optimized.

C.SCIENTIFIC PROGRAMME

Introduction

The general activity of the project will be by means of the development of new approaches and measurement protocols through collaborative multicentre research. There will be substantial further developments making use of the continuing advances in MRI technology and in workstation computing. Software already exists to aid this work and it will be the task of the activity to refine and enhance this.

Use will be made of computer facilities already existing to utilize a distributed data base of quantitative image data in order to develop texture software and communicate results to other Institutes. These facilities are crucial to share the use of expensive equipment (the MRI systems) and intellectual resources and to maximize the effectiveness of all the expertise represented within the group. The project coordinator will ensure that the activities of the whole programme and the three subgroups are harnessed to take full advantage of the complementary expertise of the participating centres. The PMG will assist in this task. The general flow of control is indicated in the diagram and the tasks of each group are detailed in Section D.

The general approach in achieving the goals of the project is to establish on a multicentre collaborative basis a standardized set of techniques for subsequent widespread testing. Linked to this will be the use of a database, both of pathologically validated image data, and of derived parameters. In all cases, success of methods will be judged through rigorous statistical analysis. Three general subgroups of participants will be established - one responsible for the development and testing of new texture analysis software, one involved with basic models of pathology, and of tissue and blood relaxation and their link to texture and a third concerned with the collection of data and clinical trials.

Clinical Applications

The focus for clinical application of the research programme will be in four areas carefully selected for maximal potential texture analysis impact, viz.:

Brain disease - Tumours

Over 1% of all deaths are due to intracranial tumours and these represent about 10% of malignant neoplasms in man. An intracranial tumour may occur at any age, although certain types are more common in childhood. The overall prognosis of intracranial tumours is influenced by the nature of the growth and its accessibility to the surgeon. In both these factors, MRI has a vital role to play, firstly in determining the nature and secondly in accurately positioning the growth. This project is concerned with the first factor, i.e. the characterization of the tumour and an improvement of diagnostic performance would be of wide value to the management of intracranial tumours throughout Europe. Of particular interest is the delineation of the margin of the neoplasm. At present this is not reliably possible, either by MRI or CT, but is very important in determining the management. Image texture analysis should allow this delineation. Another feature of great importance is the separation of radiation necrosis after therapy from tumour recurrence. This is at present impossible from the visual examination of MR images alone, but texture has the potential to play a role.

Brain disease - Multiple Sclerosis

Two main lines can be distinguished about the aetiology of Multiple Sclerosis (MS). One line pointing to evidence for genetic factors, and the second line advocating environmental factors, both contributing probably to explain the high prevalence of the disease in North America and northern Europe, in the extreme south of Africa, and in the epidemiological studies of migrants in Australia for instance. The high prevalence of the disease in Europe, extending from Switzerland and the Rhine Valley, to Finland, with a peak in the UK, justifies, of course, the support of the European Community.

Conventional T1, T2 or diffusion-weighted MRI, and qualitative analysis of image amplitude, is a very sensitive means of detecting lesions and, in association with biological and clinical signs, means more accurate diagnosis, but pathologists regret it's apparent poor specificity. Nevertheless, the discrepancy between MRI and clinical data is rather puzzling. Indeed, the topography of the plaques, even when their degree of activity is well observed by using a contrast agent, is not in agreement with neurological signs. The total volume of brain lesions appear to be independent from the severity of the disease, and conversely, during a relapse, follow-up MRI scans may not show any modifications.

Finally, long-term studies of the time-course of the disease reveal the slow growth of the lesion pool, which would appear at once at the onset of the disease. Quantitative analysis of T2 images has improved the insight into brain characterization, by describing the structural heterogeneity of the lesions, but especially by suggesting the existence of fine anomalies in the apparently normal brain matter.

Unfortunately, the performance of relaxivity studies are compromised by instrumental conditions in human investigations, such as the use of shorter interpulse delays, larger number of echoes, energy deposition in magnetization transfer technologies, etc. Texture analysis can be considered as a good alternative to overcome these difficulties, and better delineate lesions and characterize apparently normal brain tissues, while following up the pathology.

Brain disease - Epilepsy

Epilepsy, as a permanent clinical syndrome arising from a synchronous electrical hyperactivity of the brain, affects 1-2% of patients. Among the different categories of epilepsy, it would be of interest to study with an imaging method partial or complex epilepsy, depending on a localized brain area defect, generally in the frontal or temporal lobes, which affect 40% of the patients.

Drug therapy is totally successful in 40% of patients, but 20% are resistant to drug therapy, and in that case, surgical resection of the epileptogenic area remains the one curative treatment for the patient. Surgery requires a precise preoperative work-up. Metabolic PET techniques, and more recently, Nuclear Medicine SPECT techniques using perfusion tracers have been suggested as a method of surgical selection, by comparing the brain regional hyperemia of the epileptogenic area during a seizure with the corresponding hypo perfusion during the interictal period.

Morphological and structural anomalies have also been detected by MRI, noticeable on T2-weighted images, reflecting probably brain matter sclerosis due to a hypertrophy of the connective tissue, or to congenital dysplasia of the cerebral cortex, the latter anomaly being not easy to detect. Using texture analysis in that case should allow the visualization of such low contrast lesions, at the origin of the seizures, in order to determine their extent in the perspective of a surgical approach.

Osteoporosis

Osteoporosis is one of the most common health problems in the world. The disease is characterized by decrease of the quantity of bony tissue in the skeleton, in the thickness and number of trabeculae, and in the cortical thickness. Trabecular or cancellous bone, because of its high surface-to-volume ratio, has a turnover rate about eight times that of compact or cortical bone. This high turnover rate makes it a prime site for detection of early bone loss.

In recent years, considerable effort has been expended in the development of methods for quantitative assessment of the skeleton in osteoporosis and there has been a European Concerted Action on this topic (Dequeker et al (1995)). Numerous methods have been used with variable precision, accuracy and sensitivity. Examples are radiogrammetry, photon absorptiometry (which measure primarily cortical bone) and quantitative computed tomography (QCT) which looks at purely trabecular bone. There is also dual-photon absorptiometry (DPA) which measures an integral of compact and cancellous bone. Six of the partners have access to such methods for comparative studies.

More recently, MRI has been utilized (de Bisschop et al (1996)) and it is the aim of the work to progress this study to the stage of clinical usefulness, including texture analysis in the methodology.

Dequeker J, Pearson J, Reeve J et al. Dual X-ray absorptiometry: cross calibration and normative reference ranges for the spine. Results of a European Community Concerted Action. Bone 17: 247-254 (1995).

de Bisschop E, Luypaert R, Allein S and Osteaux M. Quantification of trabecular structure in the distal femur using magnetic resonance phase imaging. Magnetic Resonance Imaging 14: 11-20 (1996).

D.ORGANIZATION AND TIMETABLE

To achieve the objectives, the following working groups will be responsible for coordinating the research work within the three work packages described in Section C:

Working Group 1-Development of new texture software.

Working Group 2-Basic models of pathology and the link to texture.

Working Group 3-Collection of data and clinical trials.

A flexible format will be maintained in order to be responsive to input from other interested partners. In particular, it is hoped to take advantage of the strength in basic mathematics existing in Eastern Europe and the countries of the former Soviet Union.

The duration of the action will be four years.

Cross-disciplinary interactions with other COST technical committees: e.g. physics and materials, will be investigated.

E.ECONOMIC DIMENSION

The following COST countries have actively participated in the preparation of the Action or otherwise indicated their interest: Austria, Belgium, Denmark, France, Germany, Italy, Norway and the United Kingdom. On the basis of national estimates provided by the representatives of these countries and taking into account the coordination cost to be covered over the COST budget of the European Commission, the overall cost of the activities to be carried out under the Action has been estimated, at 1997 prices, at ECU 4 million.

National research funds are provided by nearly every European country to study magnetic resonance imaging techniques.

Appel à propositions

Data not available

Régime de financement

Data not available

Coordinateur

N/A
Contribution de l’UE
Aucune donnée
Adresse


Royaume-Uni

Voir sur la carte

Coût total
Aucune donnée