機械学習を用いた椎体の圧縮強度推定手法の検討

抄録

It is well known that osteoporosis causes different types of bone fractures. Among them, vertebral compressive fracture tends to easily take place due to the structural weakness of osteoporotic vertebrae. It is therefore clinically important to develop an assessment system for the fracture risk of vertebrae. The aim of this study is to develop a 3D convolutional neural network for estimating the fracture risk of vertebral body. we examined the extraction of vertebral region using U-net on CT images. Sensitivity, specificity, accuracy, and IoU were as high as 90%, 97%, 97%, and 52% or more. In addition, from the results of OS and US, it was found that the spinal region was estimated to be smaller.