Related Experiment Video
Updated: May 6, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
43.4K
Assessing an Automated Noncontrast CT-based Pipeline for Sacral Tumor Classification Using a Hip Bone Reference
Fei Zheng1, Ping Yin1, Kewei Liang2
1Department of Radiology, Peking University People's Hospital, No. 11 Xizhimen South Street, Xicheng District, Beijing 100044, PR China.
Radiology. Imaging Cancer
|January 2, 2026
Summary
A new automated hybrid model predicts sacral tumor types from CT scans with high accuracy. This deep learning approach, CL-MedImageNet, outperforms expert radiologists in classifying sacral tumors.
Area of Science:
- Medical Imaging and Artificial Intelligence
- Oncology and Radiology
- Deep Learning in Healthcare
Background:
- Accurate classification of sacral tumors from noncontrast CT (NCCT) images is crucial for preoperative planning.
- Existing methods may lack automation or struggle with precise tumor localization and classification.
- Deep learning offers potential for improving diagnostic accuracy and efficiency in medical image analysis.
Purpose of the Study:
- To develop and validate a fully automated hybrid deep learning model for predicting sacral tumor types using preoperative NCCT images.
- To integrate tumor segmentation and classification into a streamlined pipeline, utilizing hip bone as a novel reference frame.
- To compare the model's performance against expert radiologists in classifying six types of sacral tumors.
Main Methods:
- A retrospective, multicenter study included 690 patients with histopathologically confirmed sacral tumors.
- A hybrid model combined two deep convolutional neural networks: Model 1 for segmentation (tumors and hip bones) and Model 2 (CL-MedImageNet) for six-class classification using image, clinical, and location data.
- Model performance was evaluated using Area Under the Curve (AUC), F1 score, precision, and sensitivity, with comparisons to radiologist performance.
Main Results:
- The automated segmentation achieved high Dice coefficients (0.81-0.82) across validation and test sets.
- The CL-MedImageNet classifier demonstrated strong performance with macro average AUCs of 0.87-0.89 and macro average F1 scores of 0.56-0.63.
- The automated pipeline outperformed expert radiologists in classification accuracy (AUC 0.87 vs. 0.80, P=.002 for one comparison).
Conclusions:
- The fully automated, NCCT-based CL-MedImageNet pipeline provides accurate segmentation and robust six-class classification of sacral tumors.
- This deep learning approach significantly enhances diagnostic capabilities, outperforming expert human interpretation.
- The model represents a promising advancement for automated tumor analysis in preoperative CT imaging.
