The aim of the project is to build a prototype of an intelligent IT system processing and analyzing images with the use of artificial intelligence. The functionality of the intelligent IT system will consist in analyzing images from various sources. The final result of the project will be a prototype of the e-Medicus system for registering and analyzing data from X-ray images and classifying cancer cells. The innovative solution will be directed to medical units. The system will support the activity of health care units conducting radiology exams in the field of recording and analyzing X-ray images and the results of computed tomography scans. The analysis of radiological images will consist in their processing, as well as image identification and segmentation. Due to such analysis, it will be possible to track changes in the patient’s condition over time, and the image analysis process itself will be automated.
The planned solution will be based on the latest scientific achievements in the field of image segmentation, computational intelligence algorithms, mathematical models and theories. The system will be responsible for registering and analyzing medical images in the form of numerical functions. This will allow for the identification of medical changes and appropriate data classification. The multimedia presentation of the process of changes will be carried out with the use of the level set function, which enables the topological change of the properties of the solution. The project will develop and apply new procedures and algorithms in the field of theoretical computer science and numerical mathematics using the issues of neural networks, genetic algorithms, semantic networks, image ontology, rough set theory, level set methods and hybrid algorithms. The designed algorithms will develop methods and concepts in the field of image segmentation, capturing, transferring, collecting and extracting information and they will present them in the appropriate form.

The development of information technology has made images a valuable source of information. Often, however, while solving a specific computer problem, most of this information turns out to be useless. That is why a process of extracting interesting data from the image is needed. The process consists of a pre-processing stage in order to improve the image by binarization or noise removal; the data reduction stage, the main task of which is to extract the features relevant to the study and the stage of analysis of these features.
The medical images, which for more than one hundred years have been recorded in analogue form on X-ray film, have recently become the subject of the intense research. The X-ray film allowed for the recording and presentation of the image as well as its storage. At present radiovisiography, which was initially used only in dentistry, has become the standard. With time, however, despite of the fact it created much controversy, it appeared in other fields of medicine.
The images created in this standard are saved in digital form as a two-dimensional matrix, each element of which constitutes the value that determines the grayscale level of the respective pixel. Digital recording is a great advantage of this method, because it allowed for easier storage of images and their possible transmission for further diagnosis. Moreover, standard software provided with radiovisiography equipment offer basic image operations, such as changing the image presentation manner (rotate by any angle, zoom in or zoom out), changing the contrast or brightness, making any type of measurements (eg. angle and surface area measurements). In addition, the high quality of the sensor significantly reduces the radiation doses to which the patient is exposed during the examination. However, it remains unchanged that diagnosis of tissue anomalies is still the responsibility of the physician. Due to this, along with the development of technology for creating medical images, software supporting their analysis is created. .

The analysis of medical images allows you to locate medical conditions and classify them accordingly. This is an extremely difficult task, because in the case of many diseases, the transformation of healthy cells into diseased ones is a continuous process. Therefore it is not known when it begins and how long it lasts. That is why, the greatest difficulty is being able to detect diseased cells as early as possible.
The process of computer image analysis starts at the level of pixels. The image is presented in a form of two-dimensional pixel array providing only information about the location and color of individual points of the image, but does not contain information determining which pixels create particular objects. In order to analyze an image, it is necessary to go from pixel level to object level. It is done with help of less or more complex segmentation algorithms.

Although there are several types of segmentation, they all share one feature. They consist in partitioning the image according to a selected criterion in such a way which allows to obtain separate and uniform parts of it. It is difficult to determine which method is the best, as they usually used to extract other types of information from an image. However, their importance is crucial, because if poorly performed, they may be the main cause of erroneous results. The resolution of the images has also a significant influence on the quality of their results. If it is too high, it shows many irrelevant details, while being too low, it may lead to their incorrect extraction by contributing to blurring the edges of the object.
Among the simplest segmentation methods, one should mention segmentation by thresholding and by detecting object edges. The first one consists in determining a certain threshold value T, on the basis of which each image pixel is assigned to one of the two categories. Multilevel thresholding is also possible. In turn, the second mentioned method is based on the boundaries between the areas of different brightness, where a large difference between the levels on the grayscale of the adjacent pixels indicates the presence of edges between the objects.

Statistical methods are worth paying more attention to, including the k-means and k-nearest neighbors algorithm, which due to their simplicity and the lack of the need to determine a priori probabilities, have found wide application in image segmentation. This way, the k-means algorithm determines which pixel belongs to which group on the basis of the mutual similarity of the classified image points. The k value is predetermined, which is often a serious drawback of this algorithm. Therefore, it is usually applied multiple times for different k values and the best result is chosen. In turn, in the case of the k-nearest neighbors algorithm, the classification of pixels into individual groups is based on the so-called training data set and on k closest points. In both cases, the value of k indicates the number of groups.
Presented above algorithms were used to look for soft tissues lesions in the dental region. Figures 1 and 2 show the results of their application. It is visible that with a properly selected number of classes and the method of measuring the distance between the points, they are very similar.

The segmentation of X-ray images may be based on the variational level set method, which, compared to its basic version, eliminates the need for time-costly reinitialization. This method is a numerical technique with which it is possible to track the figures of the searched object and optimize their shape. It is presented in figure 3 on the example of a dental image. It shows that objects are located and separated with great accuracy. Separating image fragments sharing common features allows to more precisely define the boundaries between individual tissues and detect anomalies within tissues themselves.

Medical images, obtained with the use of various types of specialized equipment, enable doctors to observe the structures of the human body and the course of a number of physiological processes. Proper analysis of these images and interpretation extracted from them significantly contribute to the correct diagnosis. It should be emphasized that there is no universal imaging method for every field of medicine.