24.11.2014 Views

Elektronika 2009-11.pdf - Instytut Systemów Elektronicznych

Elektronika 2009-11.pdf - Instytut Systemów Elektronicznych

Elektronika 2009-11.pdf - Instytut Systemów Elektronicznych

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Medical pattern intelligent recognition based<br />

on linguistic modelling of 3D coronary vessels<br />

visualisations<br />

(Lingwistyczne modelowanie przestrzennych rekonstrukcji unaczynienia<br />

wieńcowego w inteligentnym rozpoznawaniu zmian patologicznych)<br />

mgr inż. MIROSŁAW TRZUPEK, prof. dr hab. MAREK R. OGIELA,<br />

prof. dr hab. inż. RYSZARD TADEUSIEWICZ<br />

AGH University of Science and Technology, Institute of Automatics, Bio-Cybernetics Laboratory, Krakow<br />

Characteristic for the coronary vessels in different patients with<br />

various pathologic forms observed on flat, i.e. 2D, images is a<br />

certain level of repetitiveness and regularity. The same images<br />

registered by a machine rendering 3D images (helical CT<br />

scanner) feature a much greater number of visible details,<br />

which, however, results also in the increase of the number of<br />

both individual differences and those between various examinations<br />

of the same patient. Having considering these difficulties,<br />

a decision was made to apply the methods of automatic<br />

image understanding for the interpretation of the images considered,<br />

which consequently leads to their semantic descriptions.<br />

Thanks to such posing of the problem, the diagnosis put<br />

forth (being naturally but a suggestion for the physician making<br />

the actual decision) will account for a greater number of<br />

factors and use better the available medical information.<br />

Additionally, it is worth pointing out that - even though this<br />

is not the main focus of the work presented here - that the solution<br />

of problems related to the automatic production of intelligent<br />

(cognitive) descriptions of the 3D medical images in<br />

question may be a significant contribution to solving at least<br />

some of the problems connected to the smart archiving of this<br />

type of data and the finding of semantic image data meeting<br />

the semantic criteria delivered through sample image patterns<br />

in medical multimedia databases.<br />

The goal of research and description<br />

of the problem<br />

Problems of modelling biomedical structure shapes are extremely<br />

interesting from the scientific point of view, but also in<br />

connection with the development of medical diagnostic support<br />

systems. They concern various structures diagnosed<br />

using various modalities. However, in such research, we extremely<br />

frequently see the use of an approach based on neutral<br />

networks for shape modelling. For instance, such models<br />

can be built using Kohonen networks [1]. However, it is worth<br />

noting that such solutions are usually limited by the network<br />

topology, which, although it allows you to operate on vectors<br />

of features informative for the objects researched, does not<br />

allow the image to be analysed or the organ to be reconstructed<br />

in a holistic fashion, i.e. taking into account the full<br />

dimensions of the image [2-4]. This only becomes possible<br />

when we create linguistic descriptions using image grammars<br />

based on graph formalisms. By definition, such grammars<br />

have been dedicated for analysing image scenes, which in the<br />

case of medical visualisation means that they can be used to<br />

model multi-object structures or to conduct the meaning analysis<br />

of spatial reconstructions of heart tomograms, which will be<br />

demonstrated later in this publication.<br />

Graph-based semantic models<br />

of the heart’s coronary vessels<br />

In order to analyze a 3D reconstruction (visualisations for various<br />

patients obtained during diagnostic examinations of the<br />

heart with a helical CT scanner with 64 detectors), it becomes<br />

necessary to select the appropriate projection showing lesions<br />

in vessels in a way that enables them to be analysed on<br />

a plane. In our research we have attempted to automate the<br />

procedure of finding such a projection by using selected geometric<br />

transformations during image processing. Next, to enable<br />

a linguistic representation of the spatial reconstructions<br />

studied, the coronary vessels shown in them had been subjected<br />

to the operation of thinning, referred to as skeletonizing.<br />

This operation allows us to obtain a skeleton of the arteries<br />

under consideration with the thickness of one unit. This skeleton<br />

can then be subjected to the operation of labelling, which<br />

determines the start and end points of main and surrounding<br />

branches of coronary arteries in it. These points will constitute<br />

the peaks of a graph modelling the spatial structure of the<br />

coronary vessels of the heart.<br />

The next step is labeling them by giving each located informative<br />

point the appropriate label from the set of peak labels<br />

which unambiguously identify individual coronary arteries<br />

forming parts of the structure analysed. For the left coronary<br />

artery they have been defined as follows: LCA - left coronary<br />

artery, LAD - anterior interventricular branch (left anterior descending),<br />

CX - circumflex branch, L - lateral branch, LM - left<br />

marginal branch. And for the right coronary artery they have<br />

been defined as follows: RCA - right coronary artery, A - atrial<br />

branch, RM - right marginal branch, PI - posterior interventricular<br />

branch, RP - right posterolateral branch. This way, all<br />

initial and final points of coronary vessels as well as all points<br />

where main vessels branch or change into lower level vessels<br />

have been determined and labelled as appropriate. After this<br />

operation, the coronary vascularization tree is divided into<br />

sections which constitute the edges of a graph modelling the<br />

examined coronary arteries. Mutual spatial relations that may<br />

occur between elements of the vascular structure represented<br />

by a graph are described by the set of edge labels. The elements<br />

of this set have been defined by introducing the appropriate<br />

spatial relations; vertical - defined by the set of labels<br />

α, β,…, µ and horizontal - defined by the set of labels 1, 2,…,<br />

24 on a hypothetical sphere surrounding the heart muscle.<br />

These labels designate individual final intervals, each of which<br />

has the angular spread of 15°. Then, depending on the location,<br />

terminal edge labels are assigned to all branches identified<br />

by the beginnings and ends of the appropriate sections of<br />

coronary arteries.<br />

ELEKTRONIKA 11/<strong>2009</strong> 9

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!