Towards a modular decision support system for radiomics: A case study on rectal cancer

Area of Science:

• Oncology
• Medical Imaging
• Data Science

Background:

• Personalized medicine necessitates predictive tools for therapy efficacy.
• Medical images offer valuable data for oncology treatment prediction.
• Radiomics, the analysis of image features for prediction, lacks integrated solutions.

Purpose of the Study:

• To introduce a comprehensive medical agent-based decision support system for the entire radiomics process.
• To evaluate the system's performance in generating predictive models for rectal cancer patients.
• To investigate the impact of data variability from different centers on model performance.

Main Methods:

• Development of a medical agent-based decision support system for end-to-end radiomics.
• Testing the system on two independent datasets of rectal cancer patients.
• Analysis of generated center-specific predictive models and data set differences.

Main Results:

• The system successfully generates highly performant center-specific predictive models.
• Significant issues related to data set differences across centers were identified.
• The impact of inter-center data variability on predictive model performance was demonstrated.

Conclusions:

• The proposed system effectively handles the complete radiomics workflow.
• Addressing data variability is crucial for robust radiomics model development and clinical application.
• The system provides a foundation for advancing personalized medicine in oncology through integrated radiomics analysis.