=Paper=
{{Paper
|id=Vol-3041/70-74-paper-12
|storemode=property
|title=Increasing the Accuracy of the Diagnosis of Mental Disorders Based on Heterogeneous Distributed Data
|pdfUrl=https://ceur-ws.org/Vol-3041/70-74-paper-12.pdf
|volume=Vol-3041
|authors=Alexander Bogdanov,Alexander Degtyarev,Natalia Zalutskaya,Natalia Gomzyakova,Sergey Belavin,Anastasia Khokhryakova
}}
==Increasing the Accuracy of the Diagnosis of Mental Disorders Based on Heterogeneous Distributed Data==
https://ceur-ws.org/Vol-3041/70-74-paper-12.pdf
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
INCREASING THE ACCURACY OF THE DIAGNOSIS OF
MENTAL DISORDERS BASED ON HETEROGENEOUS
DISTRIBUTED DATA
A. Bogdanov1, A. Degtyarev1,2,a, N. Zalutskaya3 c , N.
Gomzyakova3, S. Belavin1, A. Khokhryakova1
1
Saint Petersburg State University, 7-9 Universitetskaya emb., Saint Petersburg, 199034,
Russia
2
Plekhanov Russian University of Economics, 36 Stremyanny lane, Moscow, 117997, Russia
3
V.M.Bekhterev National Research Medical Center for Psychiatry and Neurology, 3
Bekhterev str., Saint Petersburg, 192019, Russia
E-mail: a a.degtyarev@spbu.ru
There is no single diagnostic marker for neurodegenerative diseases. The biomedical data obtained
during these studies have heterogeneous nature, which greatly complicates their collection, storage and
complex analysis. Special methods of statistical analysis due to the described specifics of the data
must be applied. The results obtained indicate that for a correct diagnosis, it is necessary to use a
comprehensive assessment of all tests.
Keywords: mental disorders, Big data, clustering, diagnostic criteria
Alexander Bogdanov, Alexander Degtyarev, Natalia Zalutskaya,
Natalia Gomzyakova, Sergey Belavin, Anastasia Khokhryakova
Copyright © 2021 for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
70
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
1. Introduction
Mental disorders of a neurodegenerative nature are one of the most pressing medical,
economic and social problems today. The most common cause of neurocognitive disorders in elderly
patients is Alzheimer's disease (AD) [1]. AD is an incurable, steadily progressive heterogeneous
disease that leads to permanent disability of the patient, which requires constant care from relatives
and medical personnel.
The causes of the development of the disease remain not fully understood, which determines
the need to search for new diagnostic methods, develop algorithms and subsequent data analysis.
Dynamics of the development of the disease is very individual, but a timely diagnosis provides an
opportunity for an early start of treatment, which can significantly slow down the progression of the
patient's disease and improve the social and labor prognosis. In this regard, the verification of early
differential diagnostic criteria acquires special scientific and practical significance.
2. Features of biomedical data
There is no single diagnostic marker for neurodegenerative diseases. It is not known in
advance how much information will be sufficient for making a diagnosis and which of the diagnostic
tools will play a key role in this decision, since the procedure for determining it is not strictly
formalized. This process is individual and often requires the collection of a big amount of information.
Diagnostics consists of a comprehensive assessment of the clinical picture when doctor observes and
interviews a patient, laboratory instrumental, experimental psychological research and neuroimaging
data. In particular, neuropsychological examination [2], magnetic resonance imaging (MRI) [3],
electroencephalography (EEG) [4], blood tests [5], genotyping [6] are used in the field of brain
pathologies research. The biomedical data obtained during these studies have heterogeneous nature,
which greatly complicates their collection, storage and complex analysis. At the same time, the
collected and processed data may have a different volume with comparable significance. The data is
stored and presented in different formats, even for the same kind of survey. As a result of patients
study, several groups of data appear from various measuring devices (EEG, MRI, etc.), which have
different formats and are processed by different software. Some of the data may have gaps for various
reasons, including those depending on the patient. Medical information is strictly confidential and
must be anonymized for further mathematical and statistical analysis. Due to the volume of research
required for one person, its high cost, collecting a huge amount of data from such studies is very labor
intensive.
In any case, in accordance with the CAP theorem and the introduced classification [7], the
data under consideration belongs to the one of Big Data types. It combines diversity, incompleteness
and lack of clear models in the subject area. Therefore, we need to use appropriate working methods.
3. Features of biomedical data processing
Modern methods of data analysis can help to solve the problem of sharing heterogeneous
clinical and biological sources in brain research. However, with such processing, specific problems
arise that must be considered in the process of work:
1. The need to consolidate data due to their heterogeneity and distribution
2. The data must be anonymized due to administrative requirements for the protection of
personal information
3. Special methods of statistical analysis due to the described specifics of the data must be
applied.
The nature of biomedical research is characterized by the work with poorly formalized
information and the work with simple or heuristic models. Therefore, the collected information is
71
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
required to distinguish certain classes. In this case, it is initially important to separate the class of
healthy patients from the classes of sick patients.
4. Processing results
Attempts to cluster using available data for only one type of tests proved to be insufficiently
effective. Despite certain successes [8,9] achieved based on MRI data processing, further
improvement in the accuracy of classifying and making a diagnosis faced problems with a limited
sample data. To this end, it was decided to involve the data of other biomedical studies of the same
patients. For these purposes, the following data was used:
Wechsler test data and blood tests
Data of MRI examinations of the brain – preprocessing with the FreeSurfer software
package
EEG data – coherent analysis
It was decided by experts to distinguish primary and secondary features in the source data. The
division of all patients into groups is carried out in several stages. The groups are preliminarily divided
according to the main characteristics in each of the tests. At the second stage, the division into classes
is refined by comparing the results and adding secondary features to the processing.
To divide into groups, the ISODATA1 [10] algorithm was applied. Its advantage is that it does
not require preliminary analysis of input data and a priori setting of the number of clusters. ISODATA
can dynamically change the number of clusters into which the data is divided, depending on the
features that characterize each element of the sample. As a result, we get the optimal number of
clusters for the given parameters. A blood test was chosen for the study in terms of determining the
number of resistant classes of patients. Cholesterol, triglyceride, glucose, HDL and LDL were selected
as the primary signs on the blood test. They were used to compare the elements. The secondary signs
were gender, age, educational level, use of psychoactive substances, traumatic brain injury and
hypertension. By dividing patients into clusters so that in each cluster there are patients with the most
similar, and patients from different clusters with the most dissimilar indicators. Such a breakthrough
will help to identify the characteristic features of a particular disease. According to secondary signs, it
is possible to determine risk groups and prescribe an additional examination for the purpose of early
detection of the disease. Initially, 5 clusters were set, but during the work this number decreased by 1.
As a result, 4 groups of patients were obtained. In one of the groups, patients from the control group
were absent, in the other, such patients were in the majority. This suggests that there are some signs
that help distinguish a healthy person from a sick person. It should be noted that such a division added
confidence in the expert assessment of specialists in the presence of exactly four groups.
Next, machine learning methods were applied to classify the data. In the pictures the
classification results for each of the studies are presented: the Wechsler test, MRI and EEG. To
separate sick patients from the control group of healthy people, the results were pooled taking into
account additional features. As a result, the best result was obtained by selecting the characteristics
given in table (significance of differences between clusters). In this case, it was possible to reduce the
result to three clusters, in one of which the control group is completely absent, and in the other healthy
people prevail over patients with dementia and Alzheimer's disease.
1
ISODATA – Iterative Self-Organizing Data Analysis Techniques
72
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
Figure 1. Classification results
Table 1. Significance of differences between clusters in accordance with Dan test
Indicator\Cluster 1-2 1-3 2-3 Indicator\Cluster 1-2 1-3 2-3
Right caudalmiddlefrontal 0,730 0,001 0,004 Right lateralorbitofrontal 0,707 0,175 0,150
ThickAvg ThickAvg
Left parahippocampal GrayVol 0,000 0,000 0,523 Right lateraloccipital 0,003 0,067 0,206
SurfArea
Left-Hippocampus 0,000 0,000 0,246 C4-Av Alfa 0,001 0,096 0,000
wm-rh-lateralorbitofrontal 0,002 0,284 0,032 F7-Av Alfa 0,091 0,000 0,000
Right lateraloccipital GrayVol 0,001 0,000 0,591 F8-Av Alfa 0,055 0,001 0,000
Left cuneus ThickAvg 0,484 0,114 0,399 T6-Av Alfa 0,004 0,000
Left frontalpole SurfArea 0,182 0,508 0,508 T3-Av Alfa 0,000 0,012 0,000
Left parsopercularis ThickAvg 0,145 0,556 0,061 T5-Av Alfa 0,008 0,019 0,000
Right lateraloccipital ThickAvg 0,024 0,024 0,917 F3-Av Alfa 0,119 0,000 0,000
Left-Accumbens-area 0,297 0,096 0,559 T4-Av Delta 0,035 0,072 0,000
Right entorhinal SurfArea 0,021 0,165 0,277 O2-Av Delta 0,998 0,003 0,004
Right parsorbitalis ThickAvg 0,080 0,171 0,003 В III 0,004 0,031 0,329
wm-lh-parsorbitalis 0,020 0,000 0,230 В Va right 0,855 0,955 0,855
WM-hypointensities 0,006 0,000 0,225 В VIIb diff 0,000 0,036 0,036
Left parahippocampal 0,000 0,001 0,512 Right isthmuscingulate 0,095 0,391 0,391
ThickAvg ThickAvg
73
Proceedings of the 9th International Conference "Distributed Computing and Grid Technologies in Science and
Education" (GRID'2021), Dubna, Russia, July 5-9, 2021
5. Conclusion
When considering each test separately, it is not possible to isolate biomarkers that allow a
clear division of patients into groups.
Clustering according to selected features of the Wechsler test, MRI and EEG allows us to
identify a group in which healthy patients were absent, which suggests that these features significantly
distinguish the group of healthy patients and those with cognitive impairments. The results obtained
indicate that for a correct diagnosis, it is necessary to use a comprehensive assessment of all tests.
The chosen method for analyzing data from examinations of the brain of patients with
cognitive impairments and the results obtained give grounds to assert that research in this area is
promising and requires further continuation, and it would be advisable to increase the number of tests
under consideration.
References
[1] Budson A.E., Kowall N.W. Handbook of Alzheimer’s disease and other dementias. Wiley-
Blackwell, 2013; 387 p.
[2] Belleville S., Fouquet C., Hudon C. et al. Neuropsychological Measures that Predict Progression
from Mild Cognitive Impairment to Alzheimer"s type dementia in Older Adults: a Systematic Review
and Meta-Analysis. Neuropsychol Rev. 2017; 27: pp.328–353
[3] De Flores R., La Joie R., Chételat G. Structural imaging of hippocampal subfields in healthy
aging and Alzheimer’s disease. Neuroscience. 2015; 309: pp. 29–50
[4] Nardone R., Sebastianelli L., Versace V., Saltuari L., Lochner P., Frey V., Golaszewski S, Brigo
F, Trinka E, Holler Y. Usefulness of EEG Techniques in Distinguishing Frontotemporal Dementia
from Alzheimer’s Disease and Other Dementias. Dis Markers. 2018 Sep 3;2018:6581490
[5] Anstey K.J., Ashby-Mitchell K., Peters R. Updating the Evidence on the Association between
Serum Cholesterol and Risk of Late-Life Dementia: Review and Meta-Analysis. J Alzheimers Dis.
2017; 56(1): pp. 215–228
[6] Neurodegenerative diseases: from the genome to the whole organism. Ed. by M.V.Ugrymov,
Moscow: Nauchni mir, 2014, 848 p.
[7] Bogdanov A., Degtyarev A., Korkhov V., Thurein K., Shchegoleva N. Big data as the future of
information technology. CEUR Workshop Proceedings, 2018, 2267, pp.26-31
[8] Korkhov V., at all Data storage, processing and analysis system to support brain research. LNCS,
2018, 10963, pp. 78-90.
[9] Volosnikov V., at all Data consolidation and analysis system for brain research. CEUR Workshop
Proceedings, 2018, 2267, pp. 388-392
[10] Tou J.T., Gonzalez R.C. Pattern Recognition Principles. Addison-Wesley PC, 1974, 378p.
74