Microsoft Word - Issue-4.docx
161
Frequent Pattern Mining of Risk Factors Predicting Neonatal Seizures
Outcomes
Ionela Maniu
Research and Telemedicine Center in Neurological Diseases in Children, Pediatric Clinical
Hospital
Department of Mathematics and Computer Science, “Lucian Blaga” University
Bulevardul Victoriei 10, Sibiu 550024, Romania
Phone: 0269 216 062
ionela.maniu@yahoo.ro
George Maniu
Department of Mathematics and Computer Science, “Lucian Blaga” University
Bulevardul Victoriei 10, Sibiu 550024, Romania
Phone: 0269 216 062
george.maniu@ulbsibiu.ro
Gabriela Visa
Research and Telemedicine Center in Neurological Diseases in Children, Pediatric Clinical
Hospital
gabiap@yahoo.com
Raluca Costea
Research and Telemedicine Center in Neurological Diseases in Children, Pediatric Clinical
Hospital
ralucacostea2004@yahoo.com
Bogdan Neamtu
Research and Telemedicine Center in Neurological Diseases in Children, Pediatric Clinical
Hospital
Departament of Computer Science and Electrical Engineering, Faculty of Engineering Lucian
Blaga University Sibiu Romania
Preclinical Departament, Faculty of Medicine, Lucian Blaga University, Sibiu, Romania
Bulevardul Victoriei 10, Sibiu 550024, Romania
Phone: 0269 216 062
bogdan.neamtu@ulbsibiu.ro, bogdan.neamtu@pediatriesibiu.ro
Abstract
This study aims to present a possible approach to identify the most common
combinations of possible risk factors for the outcomes following neonatal seizures. First, we
extract important predictor variables from the published studies featuring aspects regarding
neurological outcomes in the context of neonatal seizure. Then, we used association rules to
build prediction models to determine associations of risk factors which are frequently
identified in real-data / evidence based researches. A total of 15 studies and 14 variables were
included to identify frequent patterns and generating association rules processe. We searched
for accurate and valuable interrelationships between various risk factors. The FP-Growth
algorithm generated different itemsets with the largest including four parameters:
electroencephalography (EEG), seizures semiology (SO), aetiology (ET), birthweight (BW),
with support 0.200. The insights regarding our results may help in creating evidence-based
prevention programmes enhancing existing algorithms for diagnosis and treatment of neonatal
seizure outcomes.
Keywords: Neonatal Seizures, Risk Factors, Association Rules, Frequent Patterns.
BRAIN – Broad Research in Artificial Intelligence and Neuroscience
Volume 9, Issue 4 (November, 2018), ISSN 2067-3957
162
1. Introduction
Neonatal seizures seem to have paramount importance in forecasting longterm outcomes on
neurological maturation. These paroxysmal episodes might manifest as subtle patterns (more than
50%), clonic or myoclonic (20-25%) and tonic (~5%). Consequently, it is crucial for a neonatologist
to make an adequate differential diagnosis to exclude other essential pathologies and to intervene in
managing this syndrome effectively. Different medical conditions could mimic neonatal seizures
namely gastroesophageal reflux in Sandifer syndrome, palid reflex anoxic attacks, cyanotic breath
holding spells, hyperplexia, nervous system infections, haemorrhage or mitochondrial disorders,
epilepsy syndromes, cardiac arrhythmias (Cazan et. al., 2014; Cazan et. al., 2017).
Among the prognostic factors, aetiology ranks first followed by perinatal conditions,
seizures patterns, different biomarkers from the medical investigations, therapy type (Costea
et. al. 2017; Azhibekov et. al., 2015).
Current approaches even tackle the possibility to find suggestive biomarkers noninvasively
in epileptic toddlers or neonates (Shahar, 2012; Nicolae et. al., 2016; Pitkänen, 2016; Choy et. al.,
2014). However, the most advanced angles are used for abnormal outcomes prognosis considering
data mining techniques and association algorithms through a list of risk factors.
Many existing models for predicting mortality and or neurodevelopmental outcomes
following neonatal seizures have been described. Usually, the goal is to foresee these outcomes
based on risk factors. Irrespective of the model design to manage the outcomes the results intend to
help the clinicians and researchers but also parents in initiating and planning decisions related to
procedures, treatments, follow-ups, assistances, comparisons between different clinical studies or
populations, institutions.
However, it is difficult to choose risk factors. As can be seen from the reviewed studies, the
initial number of included risk factors (10-20 or more than 50) and different types of factors
(perinatal, aetiology, seizures, investigations, therapy) vary considerably between studies. There are
also (expected) differences in the factors combinations which are selected by the algorithms as
suggestive for high-risk outcomes. In the current study, we propose an approach to identify the most
common combinations of possible risk factors published in the literature which could influence an
abnormal outcome. Our solution is based on frequent pattern mining (FPM), applied on risk factors
identified by different research papers, a data mining technique mainly used for market basket
analysis (MBA or Affinity Analysis) to discover relations among vast amounts of business
transaction records.
2. Methods
Study selection
We reviewed the literature searching for evidence based studies with a focus on the existing
scoring systems to predict outcomes after neonatal seizures. Five scoring systems emerged (Ellison,
1981, 1986; Pisani et. al., 2009; Garfinkle, & Shevell, 2011; Salamon et. al., 2014; Hur & Chung,
2016). A detailed analysis of these research papers was already presented in a previous review
(Maniu et. al., 2017). However, we also considered other articles presenting case studies on
computational models developed to identify risk factors (Miller et. al., 2005; Ambalavanan et. al.,
2006; Nunes, 2008; Pisani, 2012; Yildiz, 2012; Lai, 2013; Vargas, 2013; Anand, 2014; Shah, 2014;
Pisani, 2016). In our previuos work (Maniu et. al., 2018) factor analysis was used to determine
groups of main risk factors considering the articles mentioned above. A total of 15 studies were
included in the current study analysis to identify the most common combinations of possible risk
factors using this time frequent pattern mining and association rules techniques.
Variables
First, we extracted the neurological risk factors from the prediction models presented
in each study in Excel format. Then, we coded each outcome variable binary with 1 if the
I. Maniu, G. Maniu, G. Visa, R. Costea, B. Neamtu - Frequent Pattern Mining of Risk Factors Predicting Neonatal
Seizures Outcomes
163
factor was considered as risk factor by the prediction model and 0 otherwise. A total of 14
variables were investigated. Eventually, we used RapidMiner implementation to build a
model for frequent items/variables set mining and association rules generation. The process
involved the following steps: (1) the Excel table with 14 binomial type attributes and 15
records (for 15 reviewed articles) was imported in RapidMiner, (2) FP-Growth operator was
added to the model, (3) Create Association Rules operator was connected with the FP-Growth
operator. In the case of FP-Growth operator, we considered 0.1 as the min_support value (to
include even outliers in the analysis) while for Association rules operator we considered 0.5
as the min_confidence parameter.
Frequent pattern mining
Association rules mining (ARM) is a branch of data mining methods, including
various algorithms that identify similarities between items/variables. It relies on techniques
such as frequent itemset mining (horizontal, vertical, projected layout), sequential
pattern(pattern-growth, apriori approaches), or structured pattern. Agrawal and Srikant
proposed for the first time in 1994 a FPM algorithm for MBA named AIS, in the form of
association rule mining. Many scalable algorithms followed, having different approaches
regarding the form, passes, reduced data set, candidate itemsets problems used data structures
or storage space. The goal was a rule set simplification and/or performance improvements
(Liu et. al., 1999, Hu & Chen, 2006, Hunyadi, 2009, Hunyadi, 2011, Kiran 2011, & Inan
2013). Beside MBA, ARM techniques have been used in many areas such as medical data
(Downs, 2000, Aswani Kumar, 2010, Chaves, 2013, & Inan 2013), Customer Relationship
Management (CRM) (Chen, 2005), spatial and spatiotemporal data (Koperski, 1995; Shiwei,
2018).
FP-Growth algorithm(frequent pattern growth)
The table no.1 presents basic concepts of FPM technique.
Table 1. Frequent pattern mining: Basic concepts
Set of all items
(ex: in a market basket data)
𝐼 = {𝑖 , 𝑖 , … , 𝑖𝒎}
itemset / k-itemset a collection of zero or more items / itemset containing k
items
transaction a subset of items chosen from 𝐼
set of all transactions 𝑇 = {𝑡 , 𝑡 , … , 𝑡𝒏}
support count the number of transactions that contain a particular itemset
𝜎 (𝑋) = | {𝑡 |𝑡 ∈ 𝑇, 𝑋 ⊆ 𝑡 } |
support fraction of transactions in which an itemset occurs
support (X) =
( )
frequent itemset an itemset (ex. X) is called frequent if:
support(X) > minsup
where minsup – user-defined threshold
An association rule is an implication of the form 𝑋 → 𝑌 where X and Y are sets of
predictors (𝑋 ⊂ 𝐼, 𝑋 ≠ ∅, 𝑌 ⊂ 𝐼, 𝑌 ≠ ∅, 𝑋 ∩ 𝑌 = ∅ ) representing the antecedent and the
consequent part of the rule.
The most common measures to quantify the strength for generated rules are presented
in table 2.
BRAIN – Broad Research in Artificial Intelligence and Neuroscience
Volume 9, Issue 4 (November, 2018), ISSN 2067-3957
164
Table 2. Association rule evaluation metrics
Confidence
Predictability
𝑋 → 𝑌 (𝑐%)
A rule has confidence c in the
dataset if
c% of the dataset transactions
that contain X also contain Y
confidence (𝑋 → 𝑌) =
𝑃 (𝑌|𝑋) =
( ∪ )
( )
how frequently items in Y appear in transactions that contain X
Support
Prevalence
𝑋 → 𝑌 (𝑐%, 𝑠%)
A rule has support s in the dataset
if
s% of the dataset transactions
contain X also contain Y
support (𝑋 → 𝑌) =
𝑃 (𝑋 ∪ 𝑌) =
( ∪ )
where n = number of
transactions
how often a rule is applicable to a given data set
ARM process identifies all the rules with the support and confidence equal or grather
than coresponding support (minsup) and confidence thresholds (minconf). It aims to predict
an item occurrence based on other items occurrences. This process is divided in two subtasks:
frequent itemset generation and rule genaration. The first task objective is to generate all
itemsets having the support >= minsup treshold. The second task objective is to generate high
confidence rules using the frequent itemsets previously identified. There are many algorithms
presented in literature for performing these tasks (as efficiently as possible).
FP-growth method is a frequent itemset mining, projected layout based method,
proposed by Han et. al in 2000. The algorithm strategy involves a two steps approach: (1)
build the FP-tree (a compressed form of the initial dataset), (2) extract frequent item sets
(frequent patterns and corresponding support) from the FP-tree, using the recursively build
conditional pattern base and conditional FP-tree (for each frequent item from the FP-tree)
(Han 2000, 2004, 2006; Borgelt 2013). The FP-tree algorithm implementation is the
following:
procedure FP growth(Tree, α)
(1) if Tree contains a single path P then
(2) for each combination (denoted as β) of the nodes in the path P
(3) generate pattern β ∪α with support_count = minimum support count of nodes
in β;
(4) else for each ai in the header of Tree {
(5) generate pattern β = ai ∪α with support_count = ai .support count;
(6) construct β’s conditional pattern base and then β’s conditional FP_tree Treeβ ;
(7) if Treeβ ≠ ∅ then
(8) call FP_growth(Treeβ , β); }
3. Results
The 15 studies analysis covered a period from 1975 to 2015, representing neonatal
seizure real-data based studies from different states/countries (California, Maryland, New
York, Brazil, Italy, Taiwan, Colombia, India, UK, Canada, Slovenia, South Korea). The
samples size varied from 55 to 403 cases. The type and number of risk factors included in the
models differed in the analysed studies. Most of them considered a range between 10 and 20
factors, but there were papers with over 50 analysed factors (Ambalavanan, 2006). Multiple
logistic regression model was the most widely used technique for risk factor identification. On
average, a study has identified 4 risk factors (with mode values: 4 and 5) while the largest
results had 6 risk factors.
In all this studies the generated model had presence/absence of post neonatal outcomes
(epilepsy, developmental delay, mortality) as target variable and different combination of
I. Maniu, G. Maniu, G. Visa, R. Costea, B. Neamtu - Frequent Pattern Mining of Risk Factors Predicting Neonatal
Seizures Outcomes
165
clinical variables (perinatal factors, etiology factors, seizures characteristics factors,
investigations findings factors, therapy related factors) as predictors. Analyzing
retrospectively the 15 studies, from the point of view of variables identified as risk factors, for
each study we recorded the information for the following variables: birth weight (BW),
gestational age (GA), mode of delivery (MD), Apgar score (AS), resuscitation (RE), seizure
characteristics: type, onset, duration, semiology (SO), status epilepticus (SE), etiology (ET),
early neurologic examination (ENE), EEG findings (EEG), cerebral ultrasound scan findings
(UBS), neuroimaging (NI), treatment duration (TD), response to treatment (TR).
In the context of our data set, the set of all items 𝐼 consisted of the 14 risk factors while the 15
reviewed studies represented the set of all transactions 𝑇.
The implementation in RapidMiner using FP-Growth operator and Association rules
operator effected different patterns. The process results are presented in the table no 3 with a
perspective including frequent pattern generation for 2-itemset, 3-itemset and 4-itemset (with
support >=0.2 for 2 and 3 itemset). The 2-itemset was counting 38 patterns, the 3-itemset 29
while the 4-itemset 8. The largest itemset consisted of 4 items, the combination with the
largest support being EEG, SO, ET, BW items(s0.200). The combination of EEG, SO, ET was
also the most common combination of 3 items (s0.333).
Table 3. Frequent itemset generation
2-itemset 3-itemset 4-itemset
0.400 EEG SO 0.333 EEG SO ET 0.200 EEG SO ET BW
0.400 EEG ET 0.267 SO ET BW 0.133 EEG SO ET MD
0.400 SO ET 0.200 EEG SO BW 0.133 EEG ET AS SE
0.333 SO BW 0.200 EEG ET BW 0.133 EEG BW AS SE
0.267 EEG BW 0.200 EEG AS SE 0.133 SO BW AS SE
0.267 ET BW 0.200 BW AS SE 0.133 BW AS UBS SE
0.267 AS SE 0.133 EEG SO MD 0.133 AS RE TR TD
0.200 EEG AS 0.133 EEG ET AS
0.200 EEG SE 0.133 EEG ET SE
0.200 EEG NI 0.133 EEG ET NI
0.200 SO AS 0.133 EEG ET MD
0.200 BW AS 0.133 EEG BW AS
0.200 BW UBS 0.133 EEG BW SE
0.200 BW SE 0.133 EEG BW ENE
0.200 AS TR 0.133 SO ET MD
The following figures describe the graphical rule visualisation (using KKLayout
format) and associated rules evaluation metrics, in the case of filtering after variables EEG
(figure no. 1) and USB (figure no. 2). Apgar score, birth weight and status epilepticus are
factors that appear in both EEG findings and cerebral ultrasound scan findings (with same
support (0.133) and confidence(0.667)). It is important to remind that the extracted rules are
useful to understand correlations / associations but not causality. For example, the rule X → Y
indicates that Y has a high probability to happen when X is present, but this does not mean
that X is the cause of Y. Aditional information / domain experience is needed to make this
claim.
BRAIN – Broad Research in Artificial Intelligence and Neuroscience
Volume 9, Issue 4 (November, 2018), ISSN 2067-3957
166
Figure 1. Graphical and tabelar rule vizualization, filtering after EEG factor
I. Maniu, G. Maniu, G. Visa, R. Costea, B. Neamtu - Frequent Pattern Mining of Risk Factors Predicting Neonatal
Seizures Outcomes
167
Figure 2. Graphical and tabelar rule vizualization, filtering after UBS factor
4. Discussions and conclusions
This study underlines the applicability and suitability of association rules mining
algorithms to determine the most common combinations of possible risk factors for outcomes
following neonatal seizures. The insights regarding our approach may be useful along with
clinical judgment and existing protocols for the epileptologists, neurophysiologists,
pediatricians, neonatologists throughout the complex decision-making process in managing
these cases. From a medical point of view more complex ET, lower AS or BW values,
moderate or severe abnormal EEG (SO) or UBS, increase the odds for an abnormal outcome.
In the proposed design we intended to provide a simulation of the association rules to identify
combinations of risk factors based on the identified risk factors from 15 literature studies. Our
study does not claim to have considered all possible risk factors for outcomes in the context of
neonatal seizure, but it successfully bundled key factors such as ET, AS, with EEG, SO and
UBS.
The analyzed articles presented models based mostly on the regression method, in
particular logistic regression, but also other methods such as decision trees. To achieve the
dataset of the present paper, the choice of variables did not take into account the prediction
model, but only the fact that certain factors were considered/identified by the model (used in
the study) as risk factors.
This retrospective study has not considered a hierarchy of the included studies in terms
of quality assessment criteria, models reliability and/or validity. We were focused on
proposed predicted models results, with an emphasis on identified risk factors. Also, some
studies considered distinct prediction models for different types of outcome while other
studies considered all outcome variants as one group using the prediction model for all types
of outcomes as a group.
The considered risk factors are very different from each other, in the sense that they
don’t have the same frequency. Some are common type of factors (BW, GA, AS) and they
BRAIN – Broad Research in Artificial Intelligence and Neuroscience
Volume 9, Issue 4 (November, 2018), ISSN 2067-3957
168
have naturally more chance of being frequent than items that are much less likely to be
considered (known as rare items problem, such as UBS).
In this context, our future work will focus on the following aspects: (1) including other
studies in the analysis, (2) classifying studies according to quality assessments, (3)
classifying studies according to prediction model, (4) classifying different individual outcome
variants, (4) including, in the generation of association rules, of variables considered in the
initial phase of the model, beside the risk factors variables identified by the model (5)
considering other algorithms (MSApriori (Liu 1999), CFPGrowth (Hu 2006), CFPGrowth++
(Kiran 2011)), to deal with issues such as items skewed support distribution (rare items)).
Acknowledgment
This work has been conducted in the Pediatric Clinic Hospital Sibiu, within Research
and Telemedicine Center in Neurological Diseases in Children - CEFORATEN project (ID
928 SMIS-CSNR 13605) financed by ANCSI with the grant number 432 / 21. 12. 2012 thru
the Sectoral Operational Programme „Increase of Economic Competitiveness”.
References
Agrawal R. and R. Srikant (1994) Fast algorithms for mining association rules. In VLDB'94,
pp. 487-499
Ambalavanan N, Carlo WA, Shankaran S, Bann CM, Emrich SL, Higgins RD, Tyson JE,
OShea TM, Laptook AR, Ehrenkranz RA, Donovan EF, Walsh MC, Goldberg RN,
Das A; (2006) Predicting outcomes of neonates diagnosed with hypoxemic-ischemic
encephalopathy. Pediatrics;118(5):2084-2093.
Anand V, Nair P. (2014) Neonatal seizures: Predictors of adverse outcome. J Pediatr
Neurosci;9:97-9
Azhibekov T, Soleymani S, Lee BH, Noori S, Seri I. Hemodynamic monitoring of the
critically ill neonate: An eye on the future. Seminars in Fetal & Neonatal Medicine.
2015;20(4):246–254.
Borgelt, C. (2010) An Implementation of the FP-growth Algorithm. Proceedings of the 1st
International Workshop on Open Source Data Mining: Frequent Pattern Mining
Implementations. 10.1145/1133905.1133907
Cazan C, Dobrota L, Neamtu M PO-0813 Sandifer Syndrome – A Challenge For The
Paediatrician Archives of Disease in Childhood 2014; 99:A520.
http://dx.doi.org/10.1136/archdischild-2014-307384.1447
Cazan C, Dobrota L, Neamtu M, Bogdan Neamtu, Gabriela Visa, Nicolae Bodrug, Viorel
Istrate Elevated aminotransferase activity as a clue to muscular dystrophy Archives of
Disease in Childhood 2017;102:A148. http://dx.doi.org/10.1136/archdischild-2017-
313273.385
Chaves, R., Ramrez, J., Grriz, J.M. (2013) Integrating discretization and association rule-
based classification for alzheimers disease diagnosis. Expert Syst. Appl. 40(5), 1571–
1578
Chen R. S.,R. C. Wu and J. Y. Chen, (2005) “Data Mining Application in Customer
Relationship Management Of Credit Card Business”, In Proceedings of 29th Annual
International Computer Software and Applications Conference (COMPSAC'05),
Volume 2, pages 39-40
Choy M, Dube CM, Patterson K, Barnes SR, Maras P, Blood AB, Hasso AN, Obenaus A,
Baram TZ. A novel, noninvasive, predictive epilepsy biomarker with clinical
potential. Journal of Neuroscience 2014; 34: 8672-8684
Costea R., Neamtu M, Neamtu B. (2016) Febrile seizures – EEG study and risks factors for
recurrence and for epilepsy. Eur J Pediatr (2016) 175: 1393.
https://doi.org/10.1007/s00431-016-2785-8
I. Maniu, G. Maniu, G. Visa, R. Costea, B. Neamtu - Frequent Pattern Mining of Risk Factors Predicting Neonatal
Seizures Outcomes
169
Downs, S., Wallace, M. (2000) “Mining Association Rules from a Pediatric Primary Care
Decision Support System.” Proceeding of the 2000 Annual Symposium of American
Medical Informatics Association, Los Angels, CA, USA, November 4-8, pp.200-204
Ellison PH, Largent JA, Bahr JP. (1981) A scoring system to predict outcome following
neonatal seizures. J Pediatr [Internet].;99(3):455–9
Ellison PH, Horn JL, Franklin S, Jones MG. (1986) The results of checking a scoring system
for neonatal seizures. Neuropediatrics [Internet].;17(3):152–7
Garfinkle J, Shevell MI. Prognostic factors and development of a scoring system for outcome
of neonatal seizures in term infants. (2011) Eur J Paediatr Neurol [Internet]. Elsevier
Ltd;15(3):222–9
Han J, Pei J, Yin Y (2000) Mining Frequent Patterns without Candidate Generation, In Proc.
ACM SIGMOD Intl. Conference on Management of Data, 29(2), pages 1-12
Han J, Pei J, Yin Y, Mao R (2004) Mining frequent patterns without candidate generation: A
frequent-pattern tree approach. In Data mining and knowledge discovery. 8(1): 53–87
Han J, Kamber M, Pei J (2006) Data mining: concepts and techniques, Morgan Kaufmann
Hu, YH, Chen, YL. (2006) Mining association rules with multiple minimum supports: a new
mining algorithm and a support tuning mechanism. Decision Support Systems,
42(1):1–24.
Hunyadi D (2009) Improvements of Apriori Algorithms, First International Conference on
Modelling and Development of Intelligent Systems – MDIS, October 22-25, Sibiu,
Romania, pg. 106-114, “Lucian Blaga” University Publishing House, ISSN 2067-3965
Hunyadi D (2011) Performance Comparison of Apriori and FP-Growth Algorithms in
Generating Association Rules, The 5th European Computing Conference (ECC’ 11),
Paris, France, April 28-30 2011, pg. 376-381,ISBN 978-960-474-297-4
Hunyadi D (2011) Performance algorithms in generating association rules, International
Journal of Mathematics and Computers in Simulation, Issue 3, Volume 5, July 2011,
pg. 282-289, ISSN 1998-0159
Hur J.Y, Chung ML. (2016) Risk Factor Analysis and Scoring System for
Neurodevelopmental Outcomes after Neonatal Seizures. J Neonatal Biol, 5(4).
Inan, O., Uzer, M.S., Yılmaz, N. (2013) A new hybrid feature selection method based on
association rules and PCA for detection of breast cancer. Int. J. Innov. Comput. Inf.
Control 9(2), 727–729
Kiran, R. U, Reddy, PK. (2011) Novel techniques to reduce search space in multiple
minimum supports-based frequent pattern mining algorithms. In: Proc. 14th Intern.
Conf. Extending Database Technology, Uppsala, Sweden, 21-24 March, 11–20
Koperski K, Han J (1995) Discovery of spatial association rules in geographic information
databases. In: Proceeding of the 1995 international symposium on large spatial
databases (SSD’95), Portland, ME, pp 47–66
Kumar Aswani, Ch., Srinivas, S. (2010) Mining associations in health care data using formal
concept analysis and singular value decomposition. J. Biol. Syst. 18(04), 787–807
Lai YH, Ho CS, Chiu NC, Tseng CF, Huang YL. (2013) Prognostic factors of developmental
outcome in neonatal seizures in term infants. Pediatr Neonatol; 54: 166-72
Liu, B., Hsu, W., Ma, Y. (1999) Mining Association Rules with Multiple Minimum Supports.
In: Proc. ACM SIGKDD Intern. Conf. Knowledge Discovery and Data Mining, San
Diego, USA, 15-18 August, 337-341
Mahoney, L. G. S. (2016) Assessment of the transitional circulation in late preterm and term
neonates using non-invasive biomarkers: a longitudinal analysis and evaluation of
repeatability (Doctoral dissertation, University of Brighton)
BRAIN – Broad Research in Artificial Intelligence and Neuroscience
Volume 9, Issue 4 (November, 2018), ISSN 2067-3957
170
Maniu I, Maniu G C, Bălan C, Neamțu B (2017) SCORING SYSTEMS FOR PREDICTING
OUTCOMES OF NEONATAL SEIZURE, Acta Medica Transilvanica, vol. 22,
ISSUE No 4, pp 48-50
Maniu I, Maniu G, Dospinescu C, Visa G. (2018) A Factor Analysis Model for Dimension
Reduction of Outcome Factors in Neonatal Seizure Context, BRAIN. Broad Research
in Artificial Intelligence and Neuroscience , Volume 9, Issue 2, ISSN 2067-8957, 95-
103
Miller, S. P., Ramaswamy, V., Michelson, D., Barkovich, A. J., Holshouser, B., Wycliffe, N.,
… Ferriero, D. M. (2005). Patterns of brain injury in term neonatal encephalopathy. J
Pediatr, 146(4), 453–460. https://doi.org/10.1016/j.jpeds.2004.12.026
Neamtu B, Picu O, Martu Stefanache MA, Ibric Cioranu VS (2016) The Comparative
Evaluation of Salivary Biomarkers (Calcium, Phosphate, Salivary pH) in Cario-
resistance Versus Cario-activity Vasile Nicolae, REV.CHIM.(Bucharest), 67, No. 4 ,
2016 http://www.revistadechimie.ro
Nunes ML, Martins MP, Barea BM, Wainberg RC, da Costa JC. (2008) Neurological
outcome of newborns with neonatal seizures. Arq Neuropsiquiatr.;66:168–174
Pisani F, Sisti L, Seri S. (2009) A Scoring System for Early Prognostic Assessment After
Neonatal Seizures. Pediatrics;124(4):e580–7.
Pisani F, Piccolo B, Cantalupo G, et al. (2012) Neonatal seizures and post-neonatal epilepsy:
a seven-year follow-up study. Pediatr Res;72:186e93
Pisani F, Facini C, Pelosi A, Mazzotta S, Spagnoli C, Pavlidis E (2016) Neonatal seizures in
preterm newborns: A predictive model of outcome. Eur J Paediatr Neurol 20:243-251
Pitkänen A, Löscher W, Vezzani A, et al. Biomarkers in epilepsy. Lancet
Neurol 2016;15:843–856
Salamon S.A, Neubauer D, Petrovcic A, Paro-Panjan D. (2014) Risk factors and scoring
system as a prognostic tool for epilepsy after neonatal seizures. Pediatr Neurol
[Internet]. Elsevier Ltd;50(1):77–84
Shahar, E., Nagler R. (2012) Oxidative Stress in Children with Intractable Epilepsy Reflected
by Salivary Glands Anti-Oxidant Profile , Neurology , 78 (1 Supplement) P01.073;
ISSN:1526-632X
Shah, D.K., Wusthoff C.J., Clarke, P., (2014). Electrographic seizures are associated with
brain injury in newborns undergoing therapeutic hypothermia, Archives of Disease in
Childhood - Fetal and Neonatal Edition 2014;99:F219-F224
Shiwei Lian, Jinning Gao, Hongwei Li (2018) A Method of Mining Association Rules for
Geographical Points of Interest, International Journal of Geo-Information
Vargas N.P, Sánchez E.F, Segura O.E (2014) Risk factors associated with a poor response
to first-line treatment in neonatal seizures a case-control study done at Mercy
Foundation Hospital, Revista Academia Nacional de Medicina, Vol 103
Yıldız EP, Tatlı B, Ekici B, et al. (2012) Evaluation of etiologic and prognostic factors in
neonatal convulsions. Pediatr Neurol; 47: 186-92
Ionela Maniu is a lecturer professor Ph.D at Lucian Blaga University of Sibiu, Faculty of Sciences,
Department of Mathematics and Computer Science and researcher in the Neurology Research
Department within the Research and Telemedicine Center in Neurological Diseases in Children
affiliated to the Pediatric Clinical Hospital from Sibiu Romania. Current research is focused on data
mining, machine learning and artificial intelligence techniques, with special interest in medical data
sets.