IMPLEMENTATION OF MACHINE LEARNING ALGORITHMS FOR CLASSIFICATION OF BONE MINERAL DENSITY TYPES BASED ON QATAR BIOBANK DATA

Abstract

Bone Mineral Density (BMD) test measures the amount of calcium and other minerals in specific areas of bone. Low BMD is a well-known problem and results in bone fractures in millions of people around the world. BMD can be affected by demographic factors (such as age, gender, etc.) and clinical features (such as Vitamin D level, Calcium, etc.). A large population is known to have issues related to bones due to low vitamin D levels. BMD can be generally classified into normal and low (Osteopenia) by using the BMD t-scores. It is of interest to know which factors can affect BMD and help in classification of BMD types. We applied machine learning techniques to classify BMD levels into "Normal" or "low" using Qatar Biobank dataset. The aim is to highlight the most important variables in classifying BMD levels, and to identify which machine learning algorithm has the ability to accurately and precisely classify BMD levels. Results showed that Random Forest (RF) was the best performing algorithm followed by Gradient Boosting. While the most important variables are "BMI", "Testosterone", "Hip-Waist ratio", "Uric Acid", "eGFR", "Ferritin", "Gender" and "Age". Research showed that we could rely on machine learning algorithms for early diagnosis of low BMD issues, which will spare time and cost