JMIR Publications

ABSTRACT

Background:

Osteoporosis is one of the diseases that require early screening and detection for its management. Common clinical tools and machine learning (ML) models for screening osteoporosis are attempted, but they have shown limitations, such as low accuracy. Moreover, they are confined to limited risk factors and lack individualized explanations.

Objective:

This paper presents an interpretable deep learning (DL) model for osteoporosis risk prediction. Clinical interpretation with individual explanations of feature contributions is provided using an explainable artificial intelligence (XAI) technique.

Methods:

In this study, we used two separate datasets, namely the National Health and Nutrition Examination Survey datasets from the United States (NHANES) and South Korea (KNHANES) (8274 and 8680 respondents, respectively). The study population was classified with a t-score of bone mineral density at the femoral neck or total femur. A DL model for osteoporosis diagnosis was trained on the datasets, and significant risk factors were investigated with local interpretable model-agnostic explanations (LIME). The performance of the DL model was compared with ML models and conventional clinical tools. Additionally, contribution ranking of risk factors and individualized explanation of feature contribution were examined.

Results:

Our DL model showed AUCs (95% confidence interval (CI)) of 0.851 (0.844 to 0.858) and 0.922 (0.916 to 0.928) for the femoral neck and total femur bone mineral density, respectively, for the NHANES dataset. The corresponding AUC values for the KNHANES dataset were 0.827 (0.821 to 0.833) and 0.912 (0.898 to 0.927), respectively. Through the LIME method, significant features were induced, and each feature’s integrated contribution and interpretation for individual risk were presented.

Conclusions:

We demonstrate that the developed DL model significantly outperforms conventional machine learning models and clinical tools. Our XAI model produces high-ranked features along with the integrated contributions of each feature, which facilitates the interpretation of individual risk. In summary, our interpretable model for osteoporosis risk prediction showed outperforming performance as compared to the state-of-the-art.