Effect of fully automatic classification model from different tube voltage images on bone density screening: A...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Bone Health Assessment

Background:

• Accurate bone status assessment is crucial for managing osteoporosis and osteopenia.
• Computed tomography (CT) imaging offers potential for bone density evaluation.
• Deep learning and radiomics are advanced techniques for image analysis.

Purpose of the Study:

• To develop bone status prediction models using deep learning and radiomics on standard-dose (SDCT) and low-dose (LDCT) chest CT images.
• To evaluate the impact of tube voltage on radiomics feature reproducibility and predictive model performance.

Main Methods:

• Developed an automated thoracic vertebral cancellous bone (TVCB) segmentation model.
• Extracted 1184 radiomics features and built classifiers for bone mineral density (BMD) assessment.
• Calculated Concordance Correlation Coefficients (CCC) to assess feature reproducibility between LDCT and SDCT.
• Evaluated model performance using Area Under the Curve (AUC).

Main Results:

• Automated segmentation achieved high accuracy (Dice coefficient > 0.97).
• Tube voltage variations led to low reproducibility for 85.05% of radiomic features (CCC < 0.75).
• LDCT and SDCT models showed high predictive performance (AUCs 0.97 and 0.94, respectively), but were sensitive to tube voltage variations.

Conclusions:

• Deep learning and radiomics models can accurately predict bone status from LDCT and SDCT.
• Tube voltage is a critical factor affecting feature reproducibility and model efficacy.
• Consistent tube voltage during image acquisition is essential for reliable bone status prediction models.