Advancing osteoporosis opportunistic screening: multicenter validation of a deep learning algorithm using abdominal
Augusto Sarquis Serpa1,2,3, Marcelo Straus Takahashi4, Eduardo Moreno Júdice de Mattos Farina5,6
1Departamento de Diagnóstico por Imagem, Federal University of São Paulo, São Paulo, Brazil. augusto.esd5@gmail.com.
Abdominal Radiology (New York)
|October 31, 2025
Summary
This study developed a deep learning algorithm to detect osteoporosis from abdominal CT scans. The validated model shows high accuracy in identifying osteoporosis, potentially improving early diagnosis.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence
Background:
- Osteoporosis diagnosis typically relies on bone mineral density (BMD) measurements using DEXA scans.
- Abdominal CT scans are widely available and can potentially be used for opportunistic osteoporosis screening.
Purpose of the Study:
- To develop and validate a deep learning (DL) algorithm for osteoporosis screening using abdominal CT images.
- To assess the diagnostic accuracy of the algorithm across multiple institutions.
Main Methods:
- A retrospective diagnostic accuracy study using CT data from January 2022 to July 2022.
- A 2D UNet with a ResNet34 backbone was used for lumbar vertebral body segmentation.
- Multicenter validation was performed incorporating data from five institutions.
Main Results:
- A strong correlation (r=0.62) was found between CT-derived mean slice mean attenuation (MSMA) and BMD.
- The DL model achieved high AUCs for osteoporosis prediction: 0.96 (internal) and 0.82 (external).
- The model demonstrated excellent sensitivity (100% internal, 79% external) and specificity (91% internal, 81% external).
Conclusions:
- A DL model was successfully developed and multicenter validated for osteoporosis prediction from abdominal CTs.
- The algorithm shows promise for opportunistic osteoporosis screening in routine clinical practice.
