Hybrid transformer convolutional neural network-based radiomics models for osteoporosis screening in routine CT
Jiachen Liu1, Huan Wang1, Xiuqi Shan1
1Department of Orthopedics, Shengjing Hospital of China Medical University, 110004, Shenyang, People's Republic of China.
BMC Medical Imaging
|March 15, 2024
Summary
A novel hybrid transformer convolutional neural network (HTCNN) model accurately screens for osteoporosis using routine CT scans. This vertebrae radiomics approach shows superior diagnostic performance compared to traditional methods, aiding early fracture prevention.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging
Background:
- Osteoporosis diagnosis is critical for preventing vertebral fractures and surgical complications.
- Routine CT scans offer a potential avenue for osteoporosis screening.
Purpose of the Study:
- To develop and evaluate a hybrid transformer convolutional neural network (HTCNN)-based radiomics model for osteoporosis screening using routine CT scans.
- To compare the diagnostic performance of the HTCNN radiomics model against traditional Hounsfield Unit (HU) values.
Main Methods:
- A HTCNN algorithm was developed for precise segmentation of vertebral bodies and trabecular compartments.
- Radiomics features were extracted from 283 vertebral bodies (204 training, 79 testing).
- Area under the receiver operating characteristic curves (AUCs) and decision curve analysis (DCA) were used for performance evaluation.
Main Results:
- The HTCNN achieved high segmentation accuracy (Dice scores of 0.968 and 0.961).
- The vertebrae radiomics score demonstrated superior efficacy in discriminating osteoporosis (AUC=0.97 in the test group).
- The HTCNN model significantly outperformed HU values and trabecular radiomics scores.
Conclusions:
- The HTCNN-based vertebrae radiomics model shows significant superiority for osteoporosis discrimination in routine CT scans.
- This AI-driven approach offers a promising tool for early osteoporosis detection and management.
