Deep Learning-Based Rectum Segmentation on Low-field Prostate MRI to Assist Image-guided Biopsy
Published in Proceedings of SPIE, Medical Imaging 2023: Image-Guided Procedures, Robotic Interventions, and Modeling, 2023
This study evaluates whether radiomics features from CT scans can predict 5-year survival in gastric cancer patients.
- Data: 406 patient CT scans; 168 survived, 238 did not.
- Approach: Tumor segmentation followed by extraction of 103 radiomics features using PyRadiomics.
- Two Methods:
- 2D: Features from a single CT slice.
- Quasi-3D: Weighted average features from multiple slices.
- Feature Reduction: PCA reduced features to 11 (2D) and 7 (3D).
- Models Tested: Logistic Regression (LRM), ANN, and SVM.
- Best Performance: LRM with quasi-3D features (AUC = 0.72), but not significantly better than 2D (AUC = 0.70, p = 0.21).
Conclusion
Recommended citation: Pham, H., Le, D. B. T., Sadinski, M., Narayanan, R., Nacev, A., & Zheng, B. (2023). "Deep Learning-Based Rectum Segmentation on Low-field Prostate MRI to Assist Image-guided Biopsy." In Proceedings of SPIE, Vol. 12466, Medical Imaging 2023: Image-Guided Procedures, Robotic Interventions, and Modeling. https://doi.org/10.1117/12.2654511
Download Paper