The work is devoted to application of mathematical methods and development of algorithms for diagnostics of the lichen planus forms that located at mucous membrane of the mouth and lips, the differential diagnosis of severe forms of other diseases and its prediction in difficult cases through the use of expert knowledge base, the results of the selection of the most informative features and micronucleus test's interpretation in buccal epithelium. Diagnosis of lichen planus that located at the oral mucosa and the vermilion border has a significant interest among dentists, dermatologists, oncologists and doctors of other specialties. This is due to the lack of clear mechanisms of the disease development, severe, often permanent period, the existing trend towards malignancy elements of destruction, as well as frequent interaction with the General condition of the patient. Treatment of patients applying for dental care with pathology of the oral mucosa and lips, is one of the most challenging problems in dentistry because of the difficulties that occur by diagnosis of diseases in this body's region. For example, the survey of 214 dentists, students of the improvement series in the Stomatology Department of the supplementary professional education Institute, Voronezh state medical University named after N. N. Burdenko, showed that only 30% of them are trying to make a diagnosis and prescribe treatment in cases of pathology of the oral mucosa and lips, and the remaining 70% of physicia ns send patients to the relevant medical Department of the University. Analysis 568 advisory directions to the Department from dentists of the city's hospitals and region shows that the most often reveals of the discrepancy in the diagnosis of diseases such as lichen planus, various forms of cheilitis, erosive-ulcerative lesions of the oral mucosa, glossalgia. During the complex examination of patients by the Department staff revealed that the differences in the diagnoses at referral and the final diagnosis was 28%. In this regard, to improve the treatment of lichen planus of the oral mucosa and the vermilion border relevant has been the development and implementation of the educational and clinical practice of the Department by the algorithm computer system which allows on the basis of the most essential signs to diagnose forms of the disease, differential diagnosis with other diseases, to give the treatment regimen.
Due to the fact that the experience of our clinic covers almost a quarter of the observations’ century, we can state the fact that the incidence of misdiagnosis has no tendency to decrease for years. This is despite the fact that in the literature, including and intended for dentists issues of clinical laboratory diagnostics of the diseases is paid more attention from year to year.
The authors analyzed clinical and laboratory characteristics of lichen planus, established expert knowledge base, the selection of the most informative parameters (erosive and ulcerative elements, the presence of local irritating factors, the presence o f inflammatory infiltrate at the base of erosions, ulcers, nature of complaints, the course of the disease, age and gender of the patients, favourite localization, lesions of the skin, the reaction of the lymph nodes, the possibility of malignancy, specific characteristics, General condition), on the basis of which the basic principles of differential diagnosis of lichen planus with other diseases of the oral mucosa and lips was put forward.
The most common form was in the sample of 212 patients with the typical one (45,7%). Exudative-hyperemic form of lichen planus was diagnosed in 19,3%, erosive – 24,5%, bullous form – at 3.8%, erosive-ulcerative – 6.6% of patients. Significant difficulties in the diagnosis of lichen planus associated with the definition of severe forms of the disease. In determining of the criteria for identifying severe forms of lichen planus took into account the following: prevalence of the process on the mucous membrane of the mouth and lips, duration, and frequency of exacerbations, length of remissions, the severity of subjective sensations (pain), the effect of previous standard therapies, changes in the quality of life of patients.
The algorithm provides two basic modes of operation: 1) using the results of the micronucleus test in buccal epithelium, which informs the patient of lichen planus, and further work on determining the shape of disease and prediction of flow depending on the shape;
2) differential diagnosis of lichen planus with diseases such as ulcus decubitalis ulcers with signs of hyperkeratosis, trophic ulcers with signs of hyperkeratosis, leukoplakia and erosive forms of chronic lupus erythematosus, pemphigus vulgaris,
Mathematical Modeling / O.V. Serikova, V.N. Kalaev, N.A. Soboleva ulcerative-necrotic stomatitis, exudative erythema multiforme, recurrent aphthous stomatitis, and then in identifying of lichen planus implementation of the transition to the first mode.
In addition, in the reference part of the algorithm provides a practical introduction to the dentist with a full description of the knowledge base in the field of diagnostics of lichen planus forms and differential diagnosis with appropriate illustrations from the auther’s own clinical observations.
Micronucleus test in buccal the epithelium of the oral cavity is one of the most widely used methods for the evaluation of
genetic homeostasis, because of its quickness, easiness, non-traumatic, cost-effective, allowing you to survey an unlimited
number of times, requires no special equipment for cultivation of cells [
It should be noted that currently, often for the differential diagnosis of severe diseases and classification of their forms use neural networks, which are attractive from an intuitive point of view, because they are based on the primitive biological model of nervous systems.
The user of a neural network collates representative data and then runs the learning algorithm that automatically adapts to the data structure. However, the user requires a set of heuristic knowledge about how to select and prepare data, the desired network architecture and to interpret the results.
Currently, there are many successful examples of the application of neural network approach to building intelligent
information systems [
The ultimate goal of our research is to create neural network systems that allow for the diagnosis and differential diagnosis of lichen planus.
As it generally known, the creation of a neural network system includes the following stages: studying of the problem;
problem statement; setting of training data and testing examples; training the neural network; optimal scheme; more
experiments; the development and creation of the interface; the connection to the trained neural networks; system test examples
not included in training data; a finish system in these examples [
As the neural structure was chosen as the Kohonen network, as it is, everyone carries out the classification. The Kohonen network can recognize clusters in data, and to establish intimacy and classes. Thus, the user can improve their understanding of data structures, and then to Refine the neural network model. If these recognized classes, they can be described, after which the network can solve classification problems. The Kohonen network can be used in the classification tasks where the classes are already set, then the advantage is that the network can identify similarities between different classes. Another possible area of application is the detection of new phenomena. The Kohonen network can recognize clusters in the training data and classifies all the data to the different clusters. If the network will meet with the dataset, unlike any of the known samples, it will not be able to classify such a set, and thus reveal its novelty. The network is trained by the Kohonen method of successive approximations. Starting from randomly chosen starting location of the centers, the algorithm progressively improves it in or der to capture the clustering of the training data.
The principle of construction of system for differential diagnosis of lichen planus is as follows. Based on the table of differential diagnostics developed by the authors, was composed of simple questions, the answers to which are binary, i.e. "Yes" or "No". In drawing up a "vector of the survey", if the answer should be "Yes", then the vector component is assigned 1 if "No", then 0. This input vectors. Similar is the vector of output values, its components have a binary form.
The most informative characteristics that allow for differential diagnosis of exudative-hyperemic forms of lichen planus are presented in table 1.
Mathematical Modeling / O.V. Serikova, V.N. Kalaev, N.A. Soboleva Table 2. The list of diseases for the differential diagnosis of exudative-hyperemic forms of lichen planus.
Disease
Code of disease Lichen planus, exudative-hyperemic form Chronic mechanical trauma Papular syphillis Leukoplakia, flat shape, Lupus erythematosus chronic Allergic stomatitis Acute hyperplastic candidiasis Early signs of recurrent aphthous stomatitis
D0 - the maximum distance between the weight vectors (columns of the matrix W ). Step 2. Presentation the network a new input signal X .
Step 3. Calculate the distance from input X to all neurons of the network:
n d j ( xi wiNj )2 , j 1, m
i1 Step 4. The choice of a neuron k , 1 k m with the shortest distance d k . Step 5. Adjusting the weights of a neuron k and all neurons that are at a distance of no more than wiNj 1 wiNj N ( xi wiNj ) .
Step 7. Steps 2-6 are repeated until the weight stops changing (or until the total change of all weights will be very small).
After training, the classification is performed by supplying to the input network of the test vector, compute the distance fr om each neuron and then the neuron with the smallest distance as the indicator of correct classification.
For training the neural network were taken 180 cases, whose data were taken from the medical records of patients with already confirmed diagnosis. The data of 185 patients treated in the clinic were left to test the system. Table 4 shows examples of correct recognition diagnosis of a number of diseases as a result of the program.
Number of correctly recognized cases
For nanosistemy presented certain difficulties, for example, the differential diagnosis between lichen planus and erythema multiforme exudative (2 errors) that occurs frequently in clinical practice when doctors make mistakes that occurs in the 35% of cases.
Comparing the encountered incorrect ("guides diagnoses") of medical institutions in patients with diseases of the oral mucosa and the vermilion border, we can say that adequate "smart diagnosis" is recorded only in 72% of cases. In the remaining patients the diagnosis was incorrect. At the same time, the application developed by authors’ algorithm of the neural network allows to obtain a correct diagnosis in 94-97%, which certainly contributes to improve early diagnosis of severe dental diseases.
Thus, the developed and implemented algorithm allows effective diagnostics of the forms of lichen planus and its differential diagnosis with other diseases. The system provides the possibility of reducing the amount of input data identifying the most significant indicators. The system is versatile and can be practical used by doctors to diagnose any other diseases by creating appropriate tests.
Program is performing differential diagnosis of lichen planus with the help of Kohonen network that is implemented in the programming system Delphi.