Related Experiment Video
Updated: Mar 23, 2026
06:45
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
264
Deep learning ensemble models for CT-based differentiation of malignant and benign sacral bone tumors: development
Ping Yin1, Fei Zheng1, Ke Liu2
1Department of Radiology, Peking University People's Hospital, Beijing, P. R. China.
Insights Into Imaging
|March 3, 2026
Summary
This study developed an artificial intelligence-radiologist ensemble model to differentiate benign from malignant sacral tumors using noncontrast computed tomography (CT) scans. The model significantly improved diagnostic performance for all radiologists, especially junior ones.
Area of Science:
- Musculoskeletal Oncology
- Radiology
- Artificial Intelligence
Background:
- Differentiating benign from malignant sacral bone lesions is challenging due to similar imaging features.
- Preoperative diagnosis is crucial for effective treatment planning.
Purpose of the Study:
- To develop an ensemble deep learning (DL) model for preoperative differentiation of benign and malignant sacral tumors using noncontrast CT.
- To evaluate the diagnostic performance and clinical utility of the DL model in conjunction with radiologists.
Main Methods:
- Analysis of 569 preoperative sacral CT scans from three centers.
- Development and testing of ensemble DL models integrating 3D-DenseNet121 with human interpretation.
- Assessment of diagnostic performance using AUC, F1 score, precision, recall, and accuracy.
Main Results:
- The ensemble model achieved high performance, with AUCs of 0.9139 (internal) and 0.8713 (external).
- All radiologists showed improved diagnostic metrics (AUC, accuracy, sensitivity, specificity) when using the DL model.
- Junior radiologists experienced more significant performance gains compared to senior radiologists.
Conclusions:
- The developed DL model enhances diagnostic efficiency for radiologists in classifying sacral tumors.
- This AI-radiologist ensemble offers a reliable approach for NCCT-based sacral tumor diagnosis, potentially reducing reliance on contrast-enhanced imaging.
