CC BY
205. Общие вопросы структурообразования при улучшении грунтов лесных дорог цементом / Аки-нин Д.В., Борисов В.А., Фокина Е.А., Казначеева Н.И. / Успехи современной науки. 2016. Т.8. №12. С.30-32
6. Некоторые результаты экспериментальных исследований при укреплении местных грунтов фурфурол-анилиновыми реагентами для лесного дорожного строительства / Акинин Д.В., Борисов В.А., Фокина Е.А., Казначеева Н.И. / Успехи современной науки. 2016. Т.8. №12. С.7-12.
7. Оценка качества обустройства и инженерного оборудования лесовозных автомобильных дорог / Никитин В.В., Акинин Д.В., Борисов В.А., Казначеева Н.И., Зарубина А.Н. / в сборнике: Приоритеты мировой науки: эксперимент и научная
дискуссия / Материалы Х международной научной конференции. 2016. С. 92-98.
8. Некоторые вопросы использования местного грунта в лесном дорожном строительстве / Акинин Д.В., Борисов В.А., Фокина Е.А., Казначеева Н.И., Данилова А.А. / В сборнике: Наука сегодня: Проблемы и перспективы развития / Материалы международной научно-практической конференции: в 2 частях. Научный центр «Диспут». 2016. С. 16-17.
9. Баженов Ю.М., Шубенкин П.Ф., Дворкин Л.И. Применение промышленных отходов в производстве строительных материалов. - М.: Стройиз-дат, 1986.
COMPUTER-AIDED RECOGNITION OF COMPLEX DERMATOGLYPHIC ELEMENT IMAGES IN DIAGNOSIS OF
HEREDITARY DISEASES
Dmitriev A. V.
Candidate of Technical Sciences, associate professor Tver State Technical University, Tver
Dmitriev G.A.
Ph.D, Professor, Tver State Technical University, Tver,
Vetrov A.N.
Candidate of Technical Sciences, associate professor, Tver State Technical University, Tver
ABSTRACT
Dermatoglyphic analysis is an efficient method for early diagnosis of a patient's genetic predisposition to a particular group of hereditary diseases. Its use encounters difficulties due to high labor inputs required for processing of dermatoglyphic images. Plus, determining a patient's predisposition to certain hereditary diseases requires an expert — a skilled dermatoglyphic analysis professional. The article proposes a new approach to address the early diagnosis problem, which relies on utilization of computer-aided dermatoglyphic element recognition, and cites its application effects.
Keywords: hereditary diseases, dermatoglyphic analysis, computer-aided image processing, diagnosis
Dermatoglyphic analysis is a method applied in early diagnosis of hereditary diseases (such as diabetes, epilepsy, schizophrenia, etc.). Predisposition to such conditions could be revealed from studying a patient's genotype by measuring the anthropometric traits. In that case, dermatoglyphic study offers the most relevant practical implications [1, 2, 7, 9].
The problem of dermatoglyphic analysis is to single out a disease classifiers by investigating the palm impressions and fingerprints. Papillary ridge and inter-ridge furrow stand out as structural units of human finger, palm, and sole skin texture. The ridges evolve into various images with varying pattern frequency. The individual diversity of papillary texture structures has been overwhelmingly wide. However, despite overwhelming diversity separate dermatoglyphic elements are groupable into a relatively small number of classes, thus facilitating their analysis. Multiple studies prove
that the following appear to be the main features (descriptors) for early diagnosis of a patient's genetic predisposition to a specific disease: distal phalanx pattern type and subtype, finger papillary pattern orientation, total ridge count, flexor furrows, palm angle (atd angle), direction of palmar lines.
The distal phalanges of human fingers generally contain papillary line patterns of three major configurations: arch (A), loop (L), whorl (W). By pattern orientation on the finger tip surface, the images can be ul-nar (u), radial (r) and symmetric (s). The pattern is divided into ulnar and radial loops by flow orientation of papillary lines. Should the flow open toward a little finger, the pattern shall be ulnar, and radial if opening toward a thumb. If the left and right papillary line flows run symmetrically, such pattern is called symmetric. See Fig. 1 for major pattern types.
Fig. 1. Major types of human finger patterns. A — arch, L — loop, W — whorl, W" — double loop whorl, o —
pattern core, d — delta
The pattern type and orientation do not alter with age, the same as the ridge count (local and total) — a quantitative measure in dermatoglyphics. The local ridge count (for individual fingers) shall be the number of ridges in the central fragment of a complex pattern in a line linking delta to the core, whereby even dot fragments of lines are covered. The arch quantitative value is zero since it has no deltas. Commonly recognized in the whorl pattern during statistical processing shall only be the ridge count from the longer delta-to-core distance. The total ridge count shall be the sum of its local values. The ridge count is a trait of finger derm-atoglyphics that is stable and unaffected by age. How-
ever, in morphogenesis it shall be a derived feature, dependent on the ridge width and the pattern core size. The ridge width is comprised by the widths of the ridge and the furrow; similarly to the pattern core size it changes throughout life. Yet, the ridge width and the pattern core size are the primary morphogenetic structures of friction ridge skin. In the distal palmar part, under II-V fingers you can find digital triradii; these serve as the starting points for main palmar lines that end in a definite field (Fig. 2). The axial or main palmar trira-dius t is in the palm central part. Position of triradius t relies either on its elevation indices or the value of atd angle formed by the arms originating from point t and crossing triradii a and d.
Fig. 2. Palmar topography layout: 1—13 — palmar fields; a, b, c, d — digital triradii and main palmar lines A, B, C, D that run from them; t — main (axial) palmar triradius; k — palmar pattern areas
Complex images of dermatoglyphic elements need to be singled out and recognized for decision-making and diagnosing a patient's predisposition to a certain hereditary disease. Apart from being painstaking and wearisome, this problem requires availability of skilled professionals who can confidently recognize those elements. The application of computer-aided image processing methods [8,11] is likely to drastically
boost the dermatoglyphic study-based diagnosing speed and promote its widespread application in the genetic counseling practice.
Computer-aided recognition methods Analysis of the dermatoglyphic study procedure [3, 4] enables to single out the following major stages of diagnosing inherited predisposition to diabetes while using digital image processing methods.
1. 2.
3.
quality;
4.
5.
Obtaining a fingerprint card;
Image scanning and digitizing;
Detecting defects and improving digital image
Image morphological processing; Image processing to generate a descriptor vector from the multitude of finger papillary patterns and optional configurations of palmar lines (dermatoglyphic traits);
6. Preparing a conclusion on inherited predisposition to a certain disease by processing the descriptor vector values.
Defects are likely to occur in scanning of dermatoglyphic images. Low quality of the images caused by such defects gives rise to significant errors in determining the classifiers. Spatial (3D) and frequency-domain processing methods are used for image enhancement (spatial and frequency methods, respectively).
The images can also be refined in its morphological processing, by using the opening and closing operation.
Image binarization is the next processing stage. Binarization is about breaking the grey-level images into two fields, one of them containing all pixels below a defined threshold, and the other - above such threshold. The pixel intensity histogram is used for that purpose.
The optimal threshold segmentation is based on image histogram approximation to a certain curve, by using the weighted sums of two or more normally distributed intensity probabilities. The thinning procedure that reduces binary objects to separate pixel-thick lines is used in further image analysis to determine the classifiers. Thinning is based on image morphological processing methods, utilizes dilatation and erosion operations, and lookup tables.
The images should have the same size and orientation to enable classifier recognition. In real world the images have arbitrary size and orientation, and need to be matched. Generalized eigenvectors of C z matrix that create an orthonormal basis in n-dimensional Euclidean space are used to orientate the objects among their major directions. Cz matrix is expressed as
1 к
Cz Zk mz )(Zk mz )\
K -1k=1
where zk is the column vector comprised by intensities of homonymous pixels g of related images. With K = MN
1 к
m = — Hz, . z Kk=i k
Transformation of the main components shall be determined under the formula
y = A(z- mz).
Generalized eigenvectors of Ck matrix shall be rows of A matrix. This transformation is used to approximate the original vector
Л A T
z = A y- m
q J z
where only q of eigenvectors is used. The vector approximate recovery error amounts to
a = j - j, j=i j=i
where Xj is comprised by eigenvalues of C z matrix.
Templates (masks) containing a particular type and subtype of dermatoglyphic feature image shall be generated for texture image recognition (Fig.3).
Fig. 3. Source image (left) and templates (masks)
Texture image databases used in analysis and recognition shall contain either the full template representation, or a condensed integral description of topo-logical properties of images. So, the recognition procedures are based on template comparison method or on discriminant decision rules in a selected feature space.
Matched filtering based on correlation alignment is a powerful solution to the problem. The correlation alignment is about finding positions on image f(x, y) which best match the given template w(x, y). To achieve that, we use a correlation metric - a generalized correlation factor
2((f(x,y) f0 )(w( x, y) w0 )
R =
t ,w
l((f(x,y) f0 )2 / Цw(x,y) w0 )2
where f0, w0 are average intensity values for the source image and the mask, respectively.
Given n templates {wi}, i = 1,...,n, each of them falling within i-th class, recognizing fragment of image
f(x, y) is subject to solving an optimization problem of the type as follows:
j = argmax(Rfw).
This is the problem of seeking a maximum value of generalized correlation factor in the set {wi}
An alternative approach is about implementing the correlation in the frequency domain. In that case the spatial correlation problem comes down to reducing the converted images, since the following statement is valid
f (x, y) O w(x, y) ^F(u, v)W(u, v),
where f(x, y)ow(x, y) is the spatial correlation.
Thus, the spatial correlation could be obtained using inverse Fourier transform applied to the products of transforming one function and the conjugated transformation of the other. The studies have proven that the fragment is thereby detected much faster.
The descriptors determined in the image processing shall be classifiers for assigning a patient's predisposition to diabetes. The Bayesian approach is used in the classification. As per the Bayes' theorem
the equation
pmo=--
( i 1 q,P(C | T) + q2P(C | T)
shall be true for any random distribution Z.
The Bayesian classification procedure involves deriving the ratio of observation vector Z to ¥i if
P(T10 >P(T10
and to ¥2 if
P(T10 < P(T10.
where qi is the a priori probability than the individual belongs to population ¥i, i = 1, 2 (it is assumed that the sum of a priori probabilities qi + q2 is equal to 1;
P(Z|¥i) is the conditional probability for obtaining a certain observation vector Z, if we know that an object belongs to population ¥i. P(¥1|Z) and P(¥2|Z) values are a posteriori probabilities.
Since the multilayer perceptron-type classifier trained by the back-propagation algorithm in the finite set of independent and evenly distributed examples ensures asymptomatic approximation of the relevant a posteriori probability of the class, a neural network was used [5, 6] as a classifier. The neural network architecture included an input layer, one hidden and one neuron in the output layer. The following logistic function type was used as the activation function in the hidden and output layers:
x) = --7-\.
1 + exp^- ax)
The number of neurons in the hidden layer was selected by experiment. Numerical experiments showed that training error is decreased monotonically with increased number of iterations, whereas the generalization error decreases only to a certain point to start growing afterwards. So, the training duration also influences the network generalization capability. A neural network with 100 neurons in the inner layer and a logistic activation function for inner and outer layers was chosen as a compromise. For neural network training results see Fig. 4. The training quality was checked by calculation of the standard square error (SSE).
R^Training with TRAINLM
| E.i|e Edit View Insert Tools Desktop Window Help H
Performance is 0.00822113. Goal is 0.01 10 -,-,-,-,-,-
Fig. 5. Neural network training results
The application of trained network in test examples indicated that the proper classification probability was 84.47%, and the standard square error stood at 0.243.
The approach offered has its advantage in no need to build additional hypotheses related to conditional probabilities.
The methods for processing dermatoglyphic images, obtaining descriptors and classifying a disease
comprise the foundation of the information system intended to computerize the dermatoglyphic studies and reveal a patient's predisposition to diabetes. For information system architecture see Fig.6.
The system implements the above computer-aided image processing methods and has modular structure. The image preprocessing and recognition modules implement the following functions:
- image quality determination and correction;
- image binarization;
- image orientation and matching; - morphological analysis and singling out individ-
ual image elements (palmar lines);
Fig.2. Architecture of information system for dermatoglyphic image processing and early diagnosis of diabetes
- image thinning (skeletonization) to determine quantitative characteristics, such as the ridge count;
- Fourier transform-based matched filtering to single out image fragments using masks.
The image processing sequence is subject to the dermatoglyphic analysis scenario and governed by a supervisor that comprises the information system kernel. The modules for image processing, descriptor generation and diagnostics are connected one by one, as the analysis algorithm proceeds. Besides, the system contains auxiliary modules that control the task execution, provide description for processing scenarios, data visualization, and a module to handle some of the errors that occur. The image processing stage ends with generation of classifiers to become the basis for a conclusion of a patient's predisposition to diabetes. The classifier values are transmitted to the classifier module for further processing and conclusion generation. The system database contains patient data and digital images of their right and left palmar impressions, and digital images of templates (masks) to search for the required fragments.
Defined requirements to loggers have been laid down to ensure quality of sources images, subject to the physical and structural properties of log objects. The requirements relate to image size, spatial resolution and contrast.
Outcome analysis
Efficiency of developed dermatoglyphic image recognition algorithms was verified in early diagnosis of predisposition to type 1 and type 2 diabetes. A sampling of 120 inherited diabetes patients (69 subjects with type 1 diabetes and 51 subjects with type 2 diabetes) from the neurosurgery early treatment center was made. For comparison, we drew a sampling of 120 persons without congenital anomalies. We determined the probability for correct classification of dermatoglyphic features, using methods for processing statistical data by calculating the ratio between values of dermato-glyphic features determined with computer assistance and visually.
The frequency of correct feature classification is 90-98%, subject to the palmar pattern type.
The share of correctly diagnosed patient conditions was correlated with the total diagnose count (without breakdown into disease types and separately for type 1 and type 2 diabetes patients) to evaluate the quality of computer-aided diagnosis of predisposition to diabetes. In the former case, the misclassification frequency totaled 15%, in the latter - 14.7% for type 1 and 16% for type 2 diabetes.
Conclusion
Testing the prototype with actual data showed efficiency and applicability of the software suite in early diagnosis of predisposition to type 1 and type 2 diabetes.
References
1. Abbasi S., Einollahi N., Vaez-Zadeh F. Study of dermatoglyphic patterns of hands in women with breast cancer //Pak J Med Sci. - 2006. - Vol.22. - pp. 18-22
2. Comparative analysis of dermatoglyphic traits in hungarian and gypsy populations /А. Nagy, M. Pap //Human Biology. - 2004. - Vol.76, N3. - pp. 383-400.
3. . G.A. Dmitriev, A.A. Azazi. Systemic analysis of images in dermatoglyphic study // Vestnik KGU. Vol.16, No.2 Kostroma, 2010. - pp. 37 -39.
4. G.A. Dmitriev, N.A. Semenov, A.A. Azazi. Computer processing of papillary images in diagnosing diabetes // Software and Systems, 2011. No. 4. pp. 193 - 196.
5. A.V. Dmitriev, A.N. Vetrov. Recognizing dermatoglyphic features based on neural networks // Medical Information Systems MIS-2006. Taganrog State University of Radio Engineering.- Taganrog.-2006.- pp. 162-167.
6. A.V. Dmitriev, G.A. Dmitriev. Using neural networks in computer-aided disease diagnosing system //From collection of research papers "Computer-aided
technology in management, healthcare, education" TSTU, Tver, 2006- pp. 112-116.
7. V.G. Solonichenko. Human adaptive pheno-types and analytical dermatoglyphics //Biomedical technology and radio electronics. - 2003. - No. 11. -pp.16-18.
8. I.N. Spiridonov. Methods for computer-aided morphometry of biomedical images // Biomedical technology and radio electronics. - 2003. - No. 11. - pp.313.
9. Yohannes S. Dermatoglyphic meta-analysis indicates early epigenetic outcomes & possible implications on genomic zygosity in type-2 diabetes. F1000Research, 2015
10. D. Forsyth, J. Ponce. Computer Vision: A Modern Approach. Transl. from English. - M.: Williams, 2004. - 928 pages
11. A.A. Khrulyov, I.A. Apollonova. Developing a mathematical model to set papillary line directions //Education via science: Abstract from international conference report. - M., 2005. - pp.360-361.
К ОЦЕНКЕ ТОЧНОСТИ АСИМПТОТИЧЕСКОГО ПРЕДСТАВЛЕНИЯ РЕШЕНИЯ ЗАДАЧИ О НАПРЯЖЕННО -ДЕФОРМИРОВАННОМ СОСТОЯНИИ ВОДОНАСЫЩЕННОГО ПОЛУПРОСТРАНСТВА, К ВЕРХНЕЙ ГРАНИЦЕ КОТОРОГО ПРИЛОЖЕНА ВЕРТИКАЛЬНАЯ РАСПРЕДЕЛЕННАЯ ПО ПЛОЩАДИ КРУГА НАГРУЗКА
Мосичева И.И.
Одесская государственная академия строительства и архитектуры, г. Одесса, старший преподаватель Шаповал А.В.
Приднепровская государственная академия строительства и архитектуры, г. Днепр, к.т.н., доцент
BY ASSESSING THE ACCURACY OF THE ASYMPTOTIC REPRESENTATION OF THE SOLUTION OF THE PROBLEM OF STRESS - STRAINED STATE OF WATER-SATURATED HALFSPACE, TO THE UPPER LIMIT OF WHICH IS ATTACHED ON THE VERTICAL DISTRIBUTION OF THE LOAD AREA OF A
CIRCLE
Mosicheva I.I.
Odessa State Academy of Civil Engineering and Architecture,
Odessa, Senior Lecturer Shapoval A.V.
Prydniprovs *'ka State Academy of Civil Engineering and Architecture,
Dnepr, Dr. Ph., docent
АННОТАЦИЯ
В рамках модели упругого водонасыщенного полупространства, к верхней границе которого приложена вертикальная распределенная по площади круга равномерная нагрузка, впервые получено асимптотическое представление решения задачи о перемещениях грунтовой толщи по глубине в процессе фильтрационной консолидации. Для оценки его точности выполнено сопоставление точного решения задачи об осадки верхней границы полупространства и его асимптотического представления. Сделан вывод об удовлетворительном соответствии точного и приближенного решений.
