American Journal of Applied Science and Technology
8
https://theusajournals.com/index.php/ajast
VOLUME
Vol.05 Issue 06 2025
PAGE NO.
8-12
10.37547/ajast/Volume05Issue06-02
Opportunities of Artificial Intelligence in The Detection
and Prognosis of Viral Hepatitis in Children
Qodirova Dilafruz Abdusamat qizi
2nd-year PhD student at Tashkent University of Information Technologies named after Muhammad al-Khwarizmi, Uzbekistan
Received:
11 April 2025;
Accepted:
07 May 2025;
Published:
09 June 2025
Abstract:
This article provides a comprehensive overview of the clinical and laboratory diagnostics of viral hepatitis
types (A, B, C, D, E) in children, methods for evaluating viral load, and the potential of artificial intelligence (AI)
models for prognosis. It discusses the use of machine learning algorithms like LSTM, GRU, and random forest for
analyzing, forecasting, and classifying hepatitis dynamics based on viral load data obtained through serological and
molecular testing. Key aspects such as data preparation, platforms, and clinical integration of AI models are also
considered.
Keywords:
Viral hepatitis, children, artificial intelligence, machine learning, diagnostics, prognosis, medical AI.
Introduction:
Viral hepatitis in children is a serious health issue, as
these infections can lead to chronic liver diseases,
cirrhosis, and hepatocellular carcinoma. Hepatitis
types A, B, C, D, and E exhibit different clinical courses
in children and require molecular and immunological
tests for diagnosis. In this regard, qPCR, dPCR, ELISA,
CLIA, and newer methods such as NGS and LAMP play
a central role. In addition, the use of artificial
intelligence and machine learning algorithms is
increasing in analyzing viral load over time,
automating result interpretation, and enabling
forecasting. Below is a brief description of the types
of viral hepatitis to provide an overview.
Hepatitis A (HAV): HAV is usually transmitted through
contaminated water or food. It is mostly
asymptomatic in children. It does not progress to a
chronic form and lifelong immunity develops after
recovery. According to WHO, 90% of children under
10 in countries with poor sanitation are infected with
HAV.
Hepatitis B (HBV): Mainly transmitted perinatally or
during childhood through blood and div fluids. It
becomes chronic in 90% of infants and 30
–
50% of
children aged 1
–
5. Chronic HBV can lead to cirrhosis
and hepatocellular carcinoma. Its spread is
decreasing thanks to vaccination.
Hepatitis C (HCV): Transmitted through blood. Often
asymptomatic in children, but it progresses to a
chronic stage. Approximately 3.5
–
5 million children
live with chronic HCV. There are medications for
treatment, but no vaccine is available.
Hepatitis D (HDV): Only infects alongside HBV. It
aggravates chronic HBV and increases the risk of liver
cancer. HDV can be prevented by HBV vaccination.
Hepatitis E (HEV): Transmitted via the fecal-oral
route. It is usually mild in children. Fulminant cases
are rare, mainly dangerous for pregnant women. HEV
vaccines are available in some countries.
Related Research
In recent years, a number of studies have been
conducted on the use of modern technologies,
including artificial intelligence (AI) models, in the
diagnosis and prognosis of viral hepatitis. According
to data from the World Health Organization (WHO),
millions of children worldwide are infected with
hepatitis B and C viruses, with most cases being
asymptomatic and progressing to chronic forms.
Therefore, early detection and monitoring are of
great importance.
Tang et al. (2018) described methods for accurately
determining HBV DNA levels using digital PCR (dPCR)
and real-time qPCR in their research. Similarly,
American Journal of Applied Science and Technology
9
https://theusajournals.com/index.php/ajast
American Journal of Applied Science and Technology (ISSN: 2771-2745)
Alavian et al. (2016) presented results on the early
detection and monitoring of hepatitis C infections in
children using serological tests.
In terms of AI-based approaches, LSTM (Long Short-
Term Memory) networks proposed by Hochreiter and
Schmidhuber (1997) have been successfully used in
medicine for forecasting time-varying data. Cho et al.
introduced the GRU (Gated Recurrent Unit) model,
which is considered lighter and more efficient
compared to LSTM.
Ahmad et al. (2021) developed a random forest
model based on HBV data that predicted the
likelihood of disease progression with high accuracy
(AUC ~0.85). Additionally, it was found that SVM and
LSTM models produced significantly lower error rates
in predicting hepatitis incidence compared to the
ARIMA model.
All of this demonstrates that AI technologies can
enhance medical diagnostics and provide effective
management of hepatitis infection dynamics.
However, for these models to be widely applied in
clinical practice, further research is needed
concerning data quality, ethical standards, and
interpretability challenges.
METHODS
Determining viral hepatitis A, B, C, D, and E in children
and assessing their viral load requires a complex and
multi-step diagnostic process. Traditional serological
and molecular tests provide important clinical
indicators of the patient's condition, but they are not
sufficient to fully analyze or predict changes in viral
load over time. There is a need to use artificial
intelligence models to detect viral load early and
predict the stages of disease development.
Molecular diagnostics (PCR methods): Quantitative
PCR (real-time PCR, qPCR) is used to accurately detect
viral load (such as Hepadnavirus DNA or Flavivirus
RNA) in the blood. qPCR measures the increase in
DNA during the reaction by tracking fluorescence.
Digital PCR (dPCR) divides the sample load into
thousands of individual mini-reactions, each of which
determines whether the virus is present and
calculates the exact copy number. These methods are
used to detect not only HBV DNA, but also HCV and
HDV RNA.
Serological markers: Immunological tests are used to
confirm hepatitis at an early stage. For example, in
hepatitis A and E, the detection of anti-HAV IgM and
anti-HEV IgM antibodies (rapid ELISA/ CLIA) indicates
a recent infection. In hepatitis B, the main markers
are HBsAg (viral surface antigen) and HBeAg (a
marker of high viral activity), the presence of HBsAg
indicates the presence of infection. IgM anti-HBc
indicates a recent B infection. In the diagnosis of
hepatitis C, an anti-HCV antidiv test is first used; if
it is positive, an HCV-RNA (PCR) test is performed.
That is, if anti-HCV is positive, the presence of an
active infection (blood virus) is determined using
RNA-PCR. The clinic also monitors the level of liver
enzymes (ALT, AST), as they indicate liver damage.
Together, these biomarkers provide the basis for
determining viral infection and assessing viral load.
Test type
Biomarkers
Main purpose
Advantages
Disadvantages
Serologik
testlar
(ELISA,
CLIA)
HBsAg, HBeAg,
Anti-HCV, Anti-
HAV IgM, Anti-
HEV IgM
To detect the
presence of the virus
Fast, relatively
affordable, widely used
Does not measure
viral load
precisely,
potential for false
positive/negative
qPCR
(Quantitative
PCR)
HBV DNK, HCV
RNK, HDV RNK
Quantitative
measurement of viral
load
Highly accurate, real-
time measurements,
allows monitoring of
viral dynamics
Expensive,
requires
laboratory
infrastructure
dPCR (Digital
PCR)
HBV DNK, HCV
RNK
Detection of very low
viral loads
High precision, detects
low loads, highly
sensitive
Very expensive,
time-consuming
American Journal of Applied Science and Technology
10
https://theusajournals.com/index.php/ajast
American Journal of Applied Science and Technology (ISSN: 2771-2745)
NGS (Next-
Generation
Sequencing)
HBV, HCV, HDV
genomes
Genomic profiling of
the virus,
identification of
mutations
Provides complete
genomic data, enables
detection of resistance
mutations
Very expensive,
requires analysis
of large datasets
LAMP (Loop-
mediated
Isothermal
Amplification)
HCV RNK, HBV
DNK
Rapid detection of
viral load
Affordable, quick, does
not require complex
instruments
Less sensitive,
prone to false
positives
ALT/AST
o‘lchovlari
ALT, AST (liver
enzymes)
Evaluation of liver
damage
Affordable, easy to use
Does not detect
viral load, only
indicates liver
damage
FibroScan /
FibroTest
Liver fibrosis stage
Assessment of liver
tissue damage
Non-invasive, easy to
apply
Does not measure
viral load, only
assesses liver
condition
Table 1. Serological and molecular diagnostics of viral hepatitis types
In addition, there are several artificial intelligence
models for the detection of viral hepatitis. LSTM
(Long Short-Term Memory) network: This is a type of
recurrent neural network specifically designed for
working with sequential data (time series). An LSTM
network learns long-term relationships between data
over time. For example, LSTM is useful for predicting
future viral levels based on previously obtained viral
loads and clinical indicators in children. An LSTM
consists of a single cell, in which the “input”, “forget”
and “output” capabilities control the flow of
information. With the help of these capabilities, the
network remembers the important signal and
discards unnecessary information, thus making
predictions based on the time series.
GRU (Gated Recurrent Unit): GRU is also a type of
RNN designed for sequential data like LSTM, but its
internal structure is somewhat simpler. It consists
only of “update” and “forget” capabilities, so it has
fewer parameters. GRU also learns long-term
relationships between data, but is computationally
lighter than LSTM.
Regression models: Simple regression (linear
regression) is used to study linear relationships in
data. For example, it is possible to predict the amount
of virus (count) based on the ALT level of children,
age, and other clinical parameters. Logistic regression
predicts the outcome in terms of binary (infected/not
infected). Regression models are intuitive and fast,
but may sometimes not be sufficient to model
complex time series well.
Other ML methods: Classical algorithms such as
decision trees and random forests, SVM, and k-NN
have also been applied to hepatitis data. For example,
in one study, a random forest model based on
laboratory parameters predicted the development of
HBV disease with high accuracy (AUC ~0.85).
However, nonlinear models such as LSTM and SVM
performed better in predicting hepatitis incidence
than classical ARIMA (LSTM showed the lowest error).
Since the problem of viral load prediction is usually
related to the analysis of time-varying data, RNN
models such as LSTM/GRU are preferable.
Model type
Task type
Advantages
Disadvantages
Best application area
LSTM / GRU
Time series
analysis,
forecasting
Learns long-term
dependencies,
understands dynamic
Learns long-term
dependencies,
understands dynamic
Forecasting viral load
over time
American Journal of Applied Science and Technology
11
https://theusajournals.com/index.php/ajast
American Journal of Applied Science and Technology (ISSN: 2771-2745)
Model type
Task type
Advantages
Disadvantages
Best application area
changes
changes
CNN
(Convolutional
Neural Network)
Image analysis,
segmentation
Automatic feature
extraction, high
accuracy
Requires large datasets
and GPUs
Detecting liver
diseases from biopsy
images
Transformer
(BERT, GPT)
Text analysis, NLP
Parallel processing,
deep contextual
understanding
Requires massive data
and computational
resources
Analyzing clinical
notes related to viral
load
Random Forest
Classification,
regression
Low risk of
overfitting, easy to
interpret
Slows down with many
trees
Classifying hepatitis
types based on clinical
features
XGBoost /
LightGBM
Classification,
regression, forecast
Very fast, high
accuracy, handles
noise well
Hard to learn, many
parameters
Exploring relationships
between parameters for
viral load prediction
AutoML (H2O.ai,
AutoKeras)
Model creation and
optimization
Easy to use, no-code
ML, rapid prototyping
Limited for complex
datasets
Building simple
models to predict
hepatitis viral load
ARIMA /
SARIMA
Time series
analysis
Simple, intuitive,
effective for small
datasets
Not suitable for non-
linear data
Month-by-month
forecasting of hepatitis
viral load
SVM (Support
Vector Machine)
Classification,
regression
Good for non-linear
data, high accuracy
Slow with large
datasets
Classifying viral load
into low/high levels
DNN (Deep
Neural Network)
Complex
classification and
prediction
Learns complex
relationships, high
accuracy
Requires very large
datasets
Predicting hepatitis
based on multiple
biomarkers
FastAPI / Flask
Model deployment
Easy API
development,
convenient integration
Not suitable for
complex scaling
Creating APIs for viral
load prediction
Docker /
Kubernetes
Model
containerization
and scaling
Easy integration,
independent execution
Complex to learn
Deploying AI models
for large-scale data
environments
Table 2. Types of Artificial Intelligence Models and their Applications
RESULTS
Data Types: The data used in the prediction model
consist of clinical and laboratory parameters. Clinical
parameters
include children’s age, gender, medical
history, symptoms, and physical examination results;
laboratory parameters include liver enzymes (ALT,
AST), hematological parameters (hemoglobin, plt),
viral biomarkers (HBV DNA, HCV RNA), and other
tests; and time-series variables include the above
American Journal of Applied Science and Technology
12
https://theusajournals.com/index.php/ajast
American Journal of Applied Science and Technology (ISSN: 2771-2745)
parameters obtained from multiple measurements.
Longitudinal data (e.g., viral levels at hospital
admission) are processed using a time-series model.
Data Preparation: First, beginners need to clean the
data: fill in missing or incorrect values (e.g., with
previous or subsequent data), remove unnecessary
columns, and check for anomalous values. Numerical
data (e.g., ALT scores) are normalized (standardized,
normalized)
–
this is important for machine learning.
Categorical data (e.g., gender, symptoms) should be
converted to a numeric format (one-hot encode or
numeric codes). If there is a time series, the data is
sorted in time order and, if necessary, processed
together with indicators obtained over a period of
time. If the database is large, it is also useful to first
select the “most important” columns (feature
selection).
Tools for building and training AI models
AutoML tools: For beginners in the software industry,
there are AutoML platforms (e.g., H2O.ai, TPOT,
AutoKeras, AutoGluon, Ludwig, and others). These
tools automatically analyze the data, select the most
suitable model, and make it easy to adjust its
parameters (hyperparameter tuning). For example,
TPOT offers tree-based optimization, and AutoKeras
offers auto-building of graph block-type neural
networks.
Python libraries: The most commonly used libraries
for AI and ML model building are scikitlearn (simple
regression, classification, preprocessing), pandas and
NumPy (data analysis and preparation), TensorFlow
and Keras or PyTorch (building differentiated neural
network architectures). For new learners, scikit-learn
has a lighter interface and is convenient for quickly
testing statistical machine learning and regression
models. It is easy to experiment with these tools in
the Jupyter Notebook environment. It is also
recommended to use matplotlib or seaborn for data
visualization.
Brief programming environments: If you do not want
to write code, there are also graphical interface
platforms such as Google AutoML, Azure AutoML for
beginners. They allow you to visually control the
machine learning process. However, for advanced
research, it is usually preferable to work with the
above Python libraries.
CONCLUSION
Molecular and serological tests play a crucial role in
the analysis of viral load, but their predictive value is
limited. Artificial intelligence models, especially LSTM
and GRU, show high efficiency in forecasting taking
into account time variability. Classical ML algorithms
such as Random Forest can be a tool for automating
diagnostics and assisting doctors. However, the
application of AI models in healthcare poses
challenges in terms of data security, clinical
validation, and interpretation.
In the future, AI-based decision-making systems are
expected to be widely implemented in clinical
practice, but this process should be carried out with
caution.
REFERENCES
World Health Organization (WHO). Hepatitis A.
https://www.who.int/news-room/fact-
sheets/detail/hepatitis-a
World Health Organization (WHO). Hepatitis B.
https://www.who.int/news-room/fact-
sheets/detail/hepatitis-b
World Health Organization (WHO). Hepatitis C.
https://www.who.int/news-room/fact-
sheets/detail/hepatitis-c
WHO. Hepatitis D and E.
room/fact-sheets/detail/hepatitis-d
Tang et al., “Digital PCR and real
-time PCR in detecting
HBV DNA,” Journal of Clinical Viro
logy, 2018.
Alavian SM et al. "Epidemiology of Hepatitis C in
children." Pediatrics Infect Dis J, 2016.
Hochreiter & Schmidhuber. “Long Short
-Term
Memory.” Neural Computation, 1997.
Cho et al., “Learning Phrase Representations using
RNN Encoder
–
De
coder,” arXiv, 2014.
Breiman, L. “Random Forests.” Machine Learning,
2001.
Ahmad et al., “Machine learning methods for
predicting hepatitis.” Health Informatics Journal,
2021.
AutoML platforms. Google, H2O.ai, AutoKeras
documentation.
Chollet F. Deep Learning with Python. Manning, 2018.
Pedregosa et al., “Scikit
-learn: Machine Learning in
Python.” JMLR, 2011.
