Related Experiment Video
Updated: Oct 17, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
7.0K
Generalized methodology for radiomic feature selection and modeling in predicting clinical outcomes
Jing Yang1,2, Lei Xu1,2, Pengfei Yang3
1Women's Hospital, Zhejiang University School of Medicine, Hangzhou, Zhejiang, 310006, People's Republic of China.
Physics in Medicine and Biology
|October 11, 2021
Summary
A new generalized radiomic feature selection and modeling (GRFM) method offers a parameter-independent approach for predicting various cancer outcomes, demonstrating promising accuracy across gastric cancer, osteosarcoma, and neuroendocrine tumors.
Area of Science:
- Radiology
- Medical Imaging Analysis
- Oncology
Background:
- Quantitative radiomic features from medical images aid clinical decision-making.
- Existing methods often rely on parameter-dependent feature selection and modeling.
- A generalized approach is needed for diverse clinical outcomes.
Purpose of the Study:
- To develop and validate a generalized radiomic method for feature selection and modeling (GRFM).
- To assess the method's applicability across various clinical outcomes.
- To establish a non-invasive prediction tool for guiding clinical decisions.
Main Methods:
- Proposed a generalized radiomic feature selection and modeling (GRFM) methodology.
- Employed a two-step feature selection: Pearson correlation analysis and sequential forward floating selection.
- Utilized an adaptive searching strategy for globally optimal, preset-free parameters.
Main Results:
- Achieved high Area Under the Curve (AUC) values: 0.9017 (gastric cancer), 0.7652 (osteosarcoma), and 0.8438 (pancreatic neuroendocrine tumors) in training cohorts.
- Validation cohorts showed comparable performance, indicating robustness.
- Identified optimal parameters and feature subsets specific to each cancer type.
Conclusions:
- The GRFM method demonstrates effective prediction of diverse clinical outcomes.
- The approach shows potential as a general, non-invasive prediction tool.
- GRFM can assist clinical decision-making across various cancer sites.
