Automatic segmentation of teeth using X-ray images

Dissertation for Master's Degree in Artificial Intelligence

Abstract

One of the most complicated tasks in digital image processing is image segmentation. Due to increasing attention to this technique by researchers and turning it into a vital role, it is used in many practical fields such as medical applications. Today, in modern dentistry, techniques based on the use of computers, such as planning and planning before surgery, are being developed day by day. In order to achieve and implement the mentioned processes, the automatic segmentation of teeth is one of the important and primary steps. In this thesis, a multi-step method for automatic segmentation of teeth in digital dental images is presented.

The main goal of this thesis is to use subbands of wavelet coefficients to improve segmentation. Each of these subbands contains important information that can be used in image segmentation. This important information is ignored in image segmentation. The main idea is to somehow add this information to the original image. Subbands of wavelet coefficients are added to the first subband of wavelet transform coefficients, corresponding to approximation coefficients that are closer to the original image in terms of value and appearance, using integration methods. After that, the wavelet transform image is done. In this case, the obtained image contains more information than the original image and the segmentation is done better and more accurately.

In this thesis, the EM algorithm is used for the segmentation of dental radiology images, and to improve this algorithm, the k-means algorithm is used for the initial estimation of the parameters of the EM algorithm. Despite its simplicity, this algorithm is considered a basic method for many other clustering methods. Morphological operators have been used to improve the segmentation. Keywords: segmentation, wavelet transform, EM algorithm, K-means algorithm, morphological operators. It is useful in engineering techniques. Extensive studies and researches have been done in this field for a long time and many advances have been made. The speed of development of these developments has been such that now and after a short period of time, the effect of image processing can be clearly seen in many sciences and industries. While some of these applications are so dependent on image processing, they cannot be used without it. In today's world, the science of image processing in a comprehensive and specialized way is gaining a more basic and important role day by day, and it is at the beginning of the way in our country. The issue of the size of image data and the effort to remove noise and image disturbances such as the parameters resulting from inappropriate light sources, the disproportion of the combination of colors and many other factors in the received images are very important issues in working with images and their processing. Segmentation[2] is an important step of image processing, the input of which is the image and the output of those features extracted from the image. Segmentation divides the image into its constituent regions or constituent objects. The level of detail at which segmentation is performed depends on the problem to be solved, i.e., segmentation should stop when objects or regions of interest in the application have been identified. Segmentation accuracy determines the ultimate success or failure of computer analysis procedures. For this reason, great care must be taken to improve the possibility of accurate segmentation.

Today in dentistry, computer-based techniques such as preoperative planning and planning, implant placement, and surgical evaluation are being developed day by day [1]. Segmentation of teeth in digital representation plays an important role in computer algorithms for feature extraction and measurement, and in orthodontic simulation for reordering teeth. In order to achieve and implement the mentioned processes, the automatic segmentation of dental images is one of the important and primary steps. Segmentation of teeth is also used in the field of identity recognition, orthodontic direction planning and facial cosmetic surgery.

Separation of dental structures is very important both anatomically and pathologically.Finding the size, volume and sometimes the location of the structures is also very important in terms of diagnosing the disease. Most of the efforts have been made on two-dimensional images of maxillary and dental surfaces and curves. Segmentation of teeth is separating different parts of a tooth from each other in dental radiology images[3]. (Relationships and formulas are available in the main file) 1-2-Segmentation Most segmentation algorithms are based on one of the two main properties of brightness intensity values, discontinuity and similarity; In the first category, the method of dividing the image is based on rapid changes in light intensity, such as edges. The main methods in the second category are based on segmenting the image into regions that are similar based on a set of predefined criteria. Thresholding, area growth and area division and integration are examples of methods in this category.

Image segmentation is also done using optimization methods. Among the important optimization algorithms, we can mention the evolutionary algorithms, genetic algorithm, ant colony algorithm, bird collective movement algorithm. Image segmentation can be considered as a process that divides R into n subregions Rn, ., R2, R1, so that:

A.

(Relations and formulas are available in the main file)

B. Ri is the connected set, for n, .,2,i=1

p.  For every j, i that

t.  For n, .,2,1 = i

Th.  For each neighboring region Ri and Ri.

Here, is a rational formula defined on the points of the set RK and is the null set. The symbols are, respectively, community, community, collection. Two regions Ri and Rj are adjacent if their union forms a connected set.

Condition (a) indicates that the segmentation must be complete [2], that is, two pixels must be in the same region, condition (b) requires that the points in a region must be connected based on predefined rules (for example, points must have a 4-way or 8-way connection. Condition (p) indicates that the regions must The condition (t) deals with the properties that must be satisfied by the pixels in the segmented region. The condition (c) indicates that the two adjacent regions Ri and Rj must be different according to the formula. Therefore, the main problem in segmentation is to divide the image into regions that satisfy the previous conditions. Segmentation algorithms for gray images are usually based on one of two categories of brightness such as discontinuity and similarity. In the segmentation based on discontinuities, it is assumed that the boundaries of the regions are completely different from each other and different from the background, as a result, it is possible to detect the border based on the local discontinuities in the light intensity. The edge-based segmentation is the main method used in this category. Area-based segmentation methods are based on dividing the image into regions that are similar based on a set of predefined criteria. 1-2-2-Edge detection [4]

In this section, we emphasize on segmentation methods that are based on the detection of sharp local changes in brightness [2]. Edge pixels are pixels where the brightness of the image varies greatly. Edge detectors are image processing methods designed to detect edge pixels. A line can be thought of as an edge where the background luminance intensity on either side of the line is much higher or much lower than the luminance intensity of the line pixels. By line, we mean thin structures and usually they are one pixel thick. Edge detection is often used to segment images based on rapid changes in brightness. An edge consists of a transition between two levels of luminance that ideally occurs over a distance of one pixel. The size of the first derivative can be used to detect the edge in a region of the image. Among the more advanced techniques for edge detection, we can mention the Mar-Hildreth edge detector [5] and the edge detector using the sufficient transformation method [6]. In this section, techniques based on brightness intensity values ??or the properties of these values ??are discussed for directly dividing images into multiple regions.