AI Model Shows Promise in Predicting Fracture Risk Based on DEXA Exams
AI Model Shows Promise in Predicting Fracture Risk Based on DEXA Exams - Express Healthcare Management Researchers from the University of Melbourne in Australia have developed an AI model that can predict fracture risk in patients based on dual-energy x-ray absorptiometry (DEXA) exams. The study, published in JBMR Plus, aims to develop AI as a screening tool for fracture risk prediction. The researchers combined a vision transformer (ViT) model with a standard convolutional neural network (VGG-16 and Resnet-50), which have proved more powerful than previously used. They collected whole-body DEXA images, as well as isolated images of the hip, forearm, and spine from fallers and non-faller controls. The model's accuracy in classifying low, moderate, or high risk of fractures was assessed using these images. The top-line results showed that the model correctly classified risk with an average area under the receiver-operating characteristic curve score of 74.3%.
Published : 2 years ago by Sandeep Kunchikor in Tech
A group of researchers from the University of Melbourne in Australia has developed an AI model that has the potential to predict fracture risk in patients based on dual-energy x-ray absorptiometry (DEXA) exams. The study, published in JBMR Plus, aims to develop AI as a screening tool for fracture risk prediction.
Fractures can have devastating consequences, and accurate prediction of fracture events can lead to prompt preventive strategies, reducing the number of occurrences. DEXA scans are commonly used to calculate bone mineral density and are the gold standard for diagnosing osteoporosis. However, assessing fracture risk using DEXA image features is rarely done due to the complexity of data analysis. Clinicians often rely on fall history to determine risk.
To evaluate fracture risk, the researchers combined a vision transformer (ViT) model with a standard convolutional neural network (VGG-16 and Resnet-50). These technologies, used together, have proved to be more powerful than when used alone.
The researchers collected whole-body DEXA images, as well as isolated images of the hip, forearm, and spine, from fallers and non-faller controls. They used landmarks on the scans, BMD measurements, and clinical variables as model input.
The dataset was split into a training set (90% of the images) and a hold-out test set (10% of the images). The model’s accuracy in classifying low, moderate, or high risk of fractures was assessed. The top-line results showed that the model correctly classified risk with an average area under the receiver-operating characteristic curve score of 74.3%.
This study demonstrates that artificial neural networks, combined with DEXA images and patient clinical data, can accurately classify fracture risk in fallers. Further research using larger high-resolution image datasets may improve fracture risk assessment and aid in identifying image-based regions of interest that indicate fracture risk.
Topics: AI