Artificial Intelligence Methods in Osteoporosis Prediction Problem

ANNA HOVAKIMYAN1, SIRANUSH SARGSYAN1, TATEV HOVAKIMYAN2,

ANI BADALYAN1

1Department of Programming and Information Technologies,

Yerevan State University,

1, Alek Manukyan St., Yerevan 0025,

ARMENIA

2AltMed Medical Center,

3/15 Hakob Hakobyan St., Yerevan 0033,

ARMENIA

Abstract: - Many sectors of human activity have implemented various solutions based on artificial intelligence

methods. These solutions help significantly in decision-making tasks, especially when analyzing a large amount

of relevant data is required beforehand. This paper discusses developing a computer system to assist doctors in

diagnosing osteoporosis based on densitometric exam results. The system was developed using machine

learning and trained on patient data obtained from densitometric examinations. The STRATOS device was used

to collect data at AltMed Medical Center in Armenia. The goal of the system is to provide an accurate diagnosis

of osteoporosis in patients while ensuring that the diagnosis is reliable and effective. During the system's

development, we utilized three prominent machine learning models: Decision Tree, Random Forest, and SVM

(Support Vector Machines). To enhance the accuracy and robustness of the system, these models were selected

based on their effectiveness in solving complex classification problems. The developed system is equipped with

advanced tools to detect potential diseases by exploring unidentified patterns and correlations among

syndromes. The mentioned capability improves the diagnostic capabilities of the system. Achieving the medical

goal requires early detection and accurate diagnosis. The AltMed Medical Center plans to utilize this system to

provide medical professionals with support for informed decisions and improved patient care. The ability of the

system to analyze complex medical data and reveal hidden insights makes it a valuable asset in the field.

Key-Words: - Osteoporosis, machine learning, classification task, osteoporosis diagnosis

Received: June 14, 2022. Revised: September 5, 2023. Accepted: September 29, 2023. Published: October 10, 2023.

1 Introduction

Osteoporosis is a common health issue that affects

approximately 200 million people all around the

world, according to the International Osteoporosis

Foundation. It is more frequent in women than in

men. Around one in every three women and two in

every ten men aged 50 and above experience

fractures related to osteoporosis, [1], [2], [3].

The impact of osteoporosis on individuals and

healthcare systems is significant due to the

associated fractures and their consequences. As the

world's population ages, healthcare providers and

policymakers face a growing challenge due to the

projected rise in the prevalence of this disease.

The early detection and effective management of

osteoporosis is crucial to mitigate its adverse impact

on health and quality of life. The integration of

machine learning can improve early detection rates

and personalized treatment approaches, ultimately

reducing the burden of osteoporosis on a global

scale, [1], [4], [5]. Several factors can increase the

risk of developing osteoporosis at any age, such as

having other health conditions at the same time,

having a genetic tendency for the disease, having

insufficient levels of calcium and vitamin D in the

blood, taking certain hormonal medications,

smoking, drinking too much alcohol, and leading a

sedentary lifestyle. It’s important to diagnose

osteoporosis as early and accurately as possible

because there is a period of time when the disease

may not show any symptoms, and early intervention

can help prevent complications, [2], [3].

The conventional techniques for medical

diagnosis, which rely on research findings, are

facing challenges due to incomplete data and

inaccuracies. Precise analysis of medical data is

crucial for the accurate diagnosis, prognosis, and

treatment of diseases. Artificial intelligence has

emerged as a valuable tool for doctors to overcome

these challenges, [4], [5].

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2023.20.17

Anna Hovakimyan, Siranush Sargsyan,

Tatev Hovakimyan, Ani Badalyan

E-ISSN: 2224-2902

171

Volume 20, 2023

Thanks to rapid advancements in computer

technology, we now have access to a vast array of

algorithms, models, and information technologies

that ensure accurate and reliable data analysis. This

has led to a significant decrease in incorrect

diagnoses, resulting in more precise and dependable

medical outcomes. The use of powerful computing

and machine learning techniques in medical data

analysis has truly revolutionized the field, [6].

Although medical technology has advanced, disease

diagnosis remains a challenge for doctors, [7].

It is important to collect high-quality medical

data and use appropriate methods to effectively

address the issue of noisy, redundant, and

incomplete data that can hinder predictive models in

medical research.

To create intelligent physician assistants using

artificial intelligence techniques, it is essential to

have a substantial and relevant dataset. This data

can be collected through laboratory tests or by

utilizing hardware technologies like magnetic

resonance imaging and computed tomography.

Additionally, data from online sources such as

Kaggle can also be utilized for this purpose, [8].

Obtaining precise medical information for

certain diseases can be challenging, which can

hinder the development of intelligent systems for

disease prevention and diagnosis, [9]. Nevertheless,

some computerized technologies that utilize medical

diagnostic procedures are currently in use,

particularly in countries such as Armenia.

For several years, the AltMed Medical Center in

Armenia has utilized the STRATOS device for

diagnosing osteoporosis, [10]. As a result, a

significant amount of data has been collected, which

necessitates the use of software tools for processing.

Through the use of these tools, we can discover

connections between research indicators that either

confirm or disprove the existence of a disease that

results from a combination of factors and irregular

indicators. The identification of correlations and

relationships among these characteristics has the

potential to bring about a groundbreaking impact on

the medical field. Doctors could benefit greatly from

this, as it would help them prescribe more targeted

treatments for specific diseases. For instance, in

cases where hormonal therapy is used to treat

thyroid gland issues, it's important to consider the

risk of inducing osteoporosis in the patient.

The paper introduces a software system that uses

machine learning to analyze densitometric

measurements acquired from the STRATOS device

at AltMed Medical Center, [10].

The newly developed software system can be a

useful tool for physicians. It can precisely diagnose

or dismiss the presence of osteoporosis in patients

by examining the digital results of densitometric

exams. Furthermore, it can detect patterns among

survey indicators and offer general statistical data

for further analysis and decision-making.

2 Machine Learning Models for

Osteoporosis Prediction System

The system in question is based on machine

learning, which is a subfield of artificial

intelligence, [9], [11]. It involves creating models

that can learn from data and use that knowledge to

make predictions or informed decisions, [6], [11].

During the training process, a machine learning

model adjusts its internal parameters to minimize

the difference between its predictions and the actual

outcomes. This continuous refinement enables the

model to enhance its accuracy and effectiveness in

making future predictions or decisions.

The process of machine learning involves

various crucial steps that are necessary for achieving

successful outcomes:

 Data Collection, Cleaning, Preprocessing, and

Transformation.

 Data Labeling.

 Selecting the Appropriate Machine Learning

Algorithm or Model.

 Model Training on Prepared Data.

 Evaluating the Model on Training and

Testing Datasets.

 Model Tuning and Optimization.

By following these systematic steps, machine

learning practitioners can develop robust and

effective models for a wide range of real-world

applications.

When it comes to analyzing data in densitometry

studies, it’s a smart decision to consider using

machine learning models like Decision Tree,

Random Forest, and SVM (Support Vector

Machine). Each of these models has its unique

strengths that can be advantageous for different

aspects of the analysis, as shown by research in

sources, [11], [12], [13], [14].

3 Structure and Implementation of

Osteoporosis Prediction System

The following sections will describe the structure of

the system that has been developed.

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2023.20.17

Anna Hovakimyan, Siranush Sargsyan,

Tatev Hovakimyan, Ani Badalyan

E-ISSN: 2224-2902

172

Volume 20, 2023

3.1 Data Collecting and Preprocessing

The developed system uses data from densitometry

and lab studies conducted with the STRATOS

device. Each data sample contains about 130

features (Figure 1, Figure 2).

Preprocessing data is a critical step in analyzing

data and building machine learning models. It

involves several tasks, including cleaning and

normalizing the data, encoding categorical

variables, dividing the dataset into training and

testing sets, and selecting important features for the

model, [11], [15].

In cases where there is not enough data available

for model tuning, artificial data generation

techniques can be used to ensure high accuracy and

reliable performance of the models, [15].

Fig. 1: Data received from the STRATOS device

Fig. 2: Densitometric examination results

3.2 Labeling of Collected Data

Data labeling is an important step in supervised

machine learning.

This step entails linking each data sample with

its corresponding target output or label.

The machine learning model is trained using the

labeled dataset to learn patterns and relationships

between input features and target labels. After

training, the model can predict labels for new data

based on learned patterns, [11].

The labeling of data relies on indicators such as

Z-score and T-score, as shown in Figure 3, [2], [3].

Fig. 3: Grades of osteoporosis on Z_score and

T_score

3.3 Tuning and Evaluation of Machine

Learning Models

The system uses machine learning models,

including Decision Tree, Random Forest, and

Support Vector Machine (SVM), which are

implemented with Python’s sci-kit-learn library.

This library offers various machine learning

algorithms and tools for analyzing data and

developing models, [11], [12], [13], [14].

We have trained and evaluated three machine

learning models - Decision Tree, Random Forest,

and SVM - using a standard method of dividing the

data into training and testing sets. The dataset was

split into 70% for training and 30% for testing,

ensuring accurate results.

The Decision Tree model achieved a training

dataset accuracy score of 99.8%, while the Random

Forest model scored 98.6%. On the testing dataset,

the Decision Tree model scored 92.3% accuracy,

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2023.20.17

Anna Hovakimyan, Siranush Sargsyan,

Tatev Hovakimyan, Ani Badalyan

E-ISSN: 2224-2902

173

Volume 20, 2023

compared to the Random Forest model's accuracy

score of 95.8%.

The SVM classifier achieves a remarkable 89%

accuracy using the Gaussian kernel, [14].

4 Correlation Data Analysis

The system analyzes patient data by focusing on age

group distribution, gender, and Body Mass Index

(BMI) to investigate the relationship between Z-

score, T-score, and BMI characteristics and uncover

potential connections and patterns. Medical

practitioners can gain deeper insights into patients'

health, identify hidden diseases early, and provide

more targeted and effective care by studying the

joint increase and decrease of Z-score, T-score, and

BMI. The visual representation of connections in

Figure 4 can help convey complex information and

improve communication between healthcare

professionals and patients.

Fig. 4: Dependence of Z-score, T-score, and BMI

The system's ability to detect connections and

correlations between different indicators from

densitometric research is powerful in diagnosing

and treating osteoporosis. Osteoporosis is a

condition where bones become less dense and more

prone to fracture, often without symptoms.

The cited examples provide valuable insights for

diagnosing and treating osteoporosis. For instance,

the BMI index and percentage of bone volume in

the total body volume are independent of each other.

Additionally, low bone density of the lumbar region

is strongly correlated with low bone density of the

wrist.

Using the connections and correlations it

discovers, this system has the potential to improve

osteoporosis management. It can help detect the

condition early and assist in creating personalized

treatment plans for people with different risk factors

and bone health characteristics. In the end, these

insights will help enhance patient care and outcomes

when it comes to diagnosing and treating

osteoporosis.

The correlation matrix shows the degree of

correlation between indicators. Each point's color

represents the correlation level (Figure 5). In a

correlation matrix, each element represents the

correlation coefficient between two variables. The

matrix is usually presented in a square format. The

correlation coefficient measures the strength and

direction of the linear relationship between two

variables, [16].

Fig. 5: Correlation between densitometric markers

In the matrix, the diagonal elements are always

set to 1. This is because a variable has a perfect

correlation with itself, resulting in a positive

correlation of 1.0. In simpler terms, a variable

always has the highest correlation with itself,

creating a flawless linear relationship.

By studying the average masses of various bones

per 1 cm² area (Figure 6), we can gain valuable

insights into the variations in bone density among

individuals. This study has revealed an important

correlation between the densities of the lumbar and

left ribs, with an 80% coincidence rate.

The diagram in Figure 7 shows that the densities

of the sternum and lumbar bones have a lower

coincidence rate of 20% in the examined

individuals.

This finding suggests a weak correlation between

bone densities in the sternum and lumbar regions,

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2023.20.17

Anna Hovakimyan, Siranush Sargsyan,

Tatev Hovakimyan, Ani Badalyan

E-ISSN: 2224-2902

174

Volume 20, 2023

emphasizing the importance of separate evaluation

for diagnosis and treatment planning.

Fig. 6: Dependencies of bone density in different

parts of the body

Fig. 7: Dependencies of bone density in different

parts of the body

5 System's Interface

An easy-to-use interface has been developed for the

intelligent diagnosis system, allowing doctors to

efficiently utilize its functionality (Figure 8).

The doctor can easily download the patient's

densitometry data and quickly receive the system's

diagnosis for osteoporosis (Figure 9, Figure 10).

The doctor can access statistical analysis,

calculate BMI, understand disease relationships, and

train the system for improvement.

Fig. 8: Doctor’s interface

Fig. 9: Disease prediction (negative result)

Fig. 10: Disease prediction (positive result)

6 Conclusion

A new technique for automating medical diagnoses

using artificial intelligence and machine learning

has been introduced in a recent paper. The software

relies on three sturdy models and can accurately

detect the presence of osteoporosis, given a

sufficient amount and quality of medical

examination results. This breakthrough has the

potential to improve diagnostic effectiveness and

enhance patient outcomes in osteoporosis detection.

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2023.20.17

Anna Hovakimyan, Siranush Sargsyan,

Tatev Hovakimyan, Ani Badalyan

E-ISSN: 2224-2902

175

Volume 20, 2023

The system is capable of discovering new

connections and dependencies within research data.

Its advanced data analysis capabilities can reveal

previously unknown relationships and correlations,

providing valuable insights and discovering novel

patterns in medical research. This feature creates

exciting opportunities for advancing scientific

knowledge and improving decision-making

processes in the field of medicine.

As the system continues to develop, it provides

doctors with numerous valuable capabilities, such

as:

 Identifying the underlying causes of disease,

suggesting appropriate tests including the thyroid

gland, diabetes, and genetic predisposition analysis

for pediatric patients.

 Allowing doctors to seamlessly integrate

and access additional test results from various

devices, enabling more targeted and personalized

treatment plans for patients.

 Utilizing the findings from investigations

that concentrate on a specific body part and

examining the connections that are uncovered, it can

now be possible to anticipate the potential spread

and severity of the illness in other parts of the body.

This remarkable feat was once thought to be

impossible, but thanks to the system's analysis, it is

now attainable.

The system revolutionizes medical practice by

offering advanced functionalities that enhance

diagnostic accuracy and enable more precise and

effective treatment strategies, ultimately leading to

improved healthcare outcomes for patients.

References:

[1] Ann E. Kearns, Mayo Clinic on

Osteoporosis: Keep your bones strong and

reduce your risk of fractures, Mayo Clinic

Press, 2021.

[2] Pauline M. Camacho, Updates on

Osteoporosis. An Issue on Endocrinology and

Metabolic Clinics of North America, Elsevier,

2021.

[3] "What is Osteoporosis?" by the National

Osteoporosis Foundation,

https://www.nof.org/ patients/what-is-

osteoporosis/ (Accessed Date: 09/10/2023)

[4] "Osteoporosis Prevention: What You Need to

Know" by WebMD,

https://www.webmd.com/

osteoporosis/features/preventing-osteoporosis-

tips (Accessed Date: on 09/10/2023)

[5] "Osteoporosis: Causes, Symptoms, and

Treatments" by Medical News Today,

https://www.medicalnewstoday.com/articles/1

55646 (Accessed Date: on 09/10/2023)

[6] Pranav Rajpurkar, Matthew P. Lungren, The

Current and Future State of AI Interpretation

of Medical Images, The New England Journal

of Medicine, May 25, 2023, pp.1981-1990,

DOI: 10.1056/NEJMra2301725

[7] Kyoung Hyup Nam, Il Seo, Dong Hwan Kim,

Jae Il Lee, Byung Kwan Choi, In Ho Han,

Machine Learning Model to Predict

Osteoporotic Spine with Hounsfield Units on

Lumbar Computed Tomography, J Korean

Neurosurg Soc., Vol.62, No.4, 2019, pp.442-

449. DOI: 10.3340/jkns.2018.0178.

[8] Kaggle: Your Machine Learning and Data

Science Community, www.kaggle.com

(Accessed Date: on 09/10/2023)

[9] Mariya Yao, Adelyn Zhou, Marlene Jia,

Applied Artificial Intelligence: A Handbook

for Business Leaders, TOPBOTS, 2018.

[10] Altmed, “AltMed” Medical Center: a unique

combination of modern and traditional

medicine methods!, www.altmed.am

[11] Andreas C. Müller, Sarah Guido, Introduction

to Machine Learning with Python: A Guide

for Data Scientists, O’Reilly, 2017.

[12] Christopher M. Bishop. Pattern Recognition

and Machine Learning, Springer, 2006.

[13] Aurélien Géron, Hands-On Machine Learning

with Scikit-Learn and TensorFlow: Concepts,

Tools, and Techniques to Build Intelligent

Systems, O’Reilly, 2017.

[14] Ingo Steinwart, Andreas Christmann, Support

Vector Machines, Springer Science+Business

Media, 2008.

[15] Prateek Joshi, Artificial intelligence with

Python, Packt, 2019.

[16] Displayr, What is a Correlation Matrix?,

https://medium.com/@displayr/what-is-a-

correlation-matrix-b2196f728bc2

www.altmed.am (Accessed Date: on

09/10/2023)

Contribution of individual authors to the

creation of a scientific article (ghostwriting

policy)

- Anna Hovakimyan has proposed methods for

solving the problem and was responsible for data

collecting.

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2023.20.17

Anna Hovakimyan, Siranush Sargsyan,

Tatev Hovakimyan, Ani Badalyan

E-ISSN: 2224-2902

176

Volume 20, 2023

- Siranush Sargsyan organized the experiments and

was responsible for results statistical analysis.

- Tatev Hovakimyan is an endocrinologist who

treats osteoporosis.

- Ani Badalyan has implemented Machine Learning

models in Python.

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflict of Interest

The authors have no conflicts of interest to declare

that are relevant to the content of this article.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on BIOLOGY and BIOMEDICINE

DOI: 10.37394/23208.2023.20.17

Anna Hovakimyan, Siranush Sargsyan,

Tatev Hovakimyan, Ani Badalyan

E-ISSN: 2224-2902

177

Volume 20, 2023