Researchers have created an innovative deep learning algorithm that surpassed current computer-based methods for predicting osteoporosis risk, which could result in earlier detection and improved outcomes for individuals at risk of osteoporosis.
Osteoporosis is often challenging to diagnose in its early stages and is known as the “silent disease.” Imagine if artificial intelligence could assist in predicting a person’s likelihood of developing this bone-loss condition even before visiting a healthcare provider?
Tulane University researchers have taken a step toward this vision by developing a new deep learning algorithm that outperformed existing computer-based methods for predicting osteoporosis risk, potentially leading to earlier diagnoses and better outcomes for individuals at risk of osteoporosis.
The findings of their study were recently published in Frontiers in Artificial Intelligence.
Deep learning models are notable for their ability to emulate human neural networks and identify patterns within vast datasets without the need for explicit programming to do so. Researchers evaluated the deep neural network (DNN) model against four traditional machine learning algorithms and a standard regression model. They used data from over 8,000 participants aged 40 and above in the Louisiana Osteoporosis Study. The DNN demonstrated superior predictive performance overall, assessed by evaluating each model’s capability to correctly identify true positives and prevent errors.
Lead author Chuan Qiu, a research assistant professor at the Tulane School of Medicine Center for Biomedical Informatics and Genomics, noted, “The earlier osteoporosis risk is identified, the more opportunity a patient has to take preventive measures. We were pleased to discover that our DNN model surpassed other models in accurately predicting osteoporosis risk in an aging population.”
By analyzing the algorithms using a substantial real-world health data set, the researchers pinpointed the ten most critical factors for predicting osteoporosis risk. These factors include weight, age, gender, grip strength, height, beer consumption, diastolic pressure, alcohol intake, years of smoking, and income level.
Interestingly, the simplified DNN model utilizing these top 10 risk factors performed nearly as effectively as the comprehensive model that encompassed all risk factors.
Although Qiu acknowledged the necessity for further advancements before an AI system can be utilized by the public to predict an individual’s osteoporosis risk, he emphasized that recognizing the advantages of the deep learning model was a significant step in that direction.
Qiu stated, “Our ultimate goal is to enable individuals to input their information and receive highly precise osteoporosis risk scores, empowering them to seek treatment to enhance bone strength and minimize any further harm.”
