Although dual-energy x-ray absorptiometry (DEXA) studies and extremity exams both utilize a small amount of radiation, DEXA provides no additional diagnostic information. As a result, the research team from Thomas Jefferson University led by presenter Dr. Simukayi Mutasa sought to utilize convolutional neural networks (CNNs) to obtain bone mineral density from diagnostic imaging studies performed for other purposes, beginning with hip radiographs.
They first gathered a dataset of over 4,400 patients over the age of 25 who had received a DEXA exam and a hip or pelvic radiograph within a year. A script was then applied to extract quantitative labels -- the average of the hip, L-spine, femur, and forearm when available -- as well as qualitative labels indicating normal bone density, osteopenia, or osteoporosis.
The researchers then manually annotated a bounding box of Ward's triangle on the anteroposterior view for each radiograph. These annotations were subsequently used to train a CNN to isolate the Ward's triangle on radiographs. Next, a second CNN was developed to classify the hip images as either normal, containing osteoporosis, containing osteopenia, or not containing the hip.
In preliminary validation, the CNN produced classification accuracy of 81.2%, sensitivity of 83.8%, specificity of 77.2%, and an area under the curve (AUC) of 0.809.
"A deep-learning system, when applied to extremity radiographs such as the hip, can diagnose abnormalities in bone mineral density similar to a DEXA scan," the authors wrote.