AI-Based Predictive Analysis of Osteoporosis: A Machine Learning Approach for Early Diagnosis

Abstract

In underserved regions like Sub-Saharan Africa, Osteoporosis, a debilitating disease remains one of the pathologies that goes undetected owing to limited access to advanced diagnostic equipment like Dual-Energy X-ray Absorptiometry (DEXA) scans. The wave of Artificial Intelligence (AI) offers the capacity to utilize its predictive power harnessed by training models on datasets composed of demographic, medical and lifestyle variables to assess the risks of Osteoporosis in these regions. This study employs the Random Forest algorithm based on reduced tendency for overfitting and its efficiency in handling categorical and numerical variables to evaluate a machine learning model, OsteoModel using a dataset of 1,958 patient records downloaded from Kaggle. The model achieved a predictive accuracy of 84.69% (95% Confidence Level (Cl): 84.47%-84.91%) with a recall value of 0.75(95%Cl: 74.78%- 75.22%) for Osteoporosis cases. Analysis of feature importance showed age, race, medical history and lifestyle as the key predictors. However, the dataset can potentially be biased in composition and lack diversity, necessitating the need for further model training and evaluation with an independent dataset for future studies. The outcome of this study reveals the potential and key role AI diagnostic tools can play in narrowing the gap caused by lack of access to conventional diagnostic tools in regions of low resources. It is imperative that emphasis be placed on the need for the complete and urgent integration of AI based tools into Osteoporosis screening especially in various Primary Health Care facilities in Sub-Saharan Africa with limited access to advance screening tools. It is also important for an AI regulatory framework to exist that will ensure compliance to the ethics during test and deployment.