Integrating radiomics into predictive models for low nuclear grade DCIS using machine learning.

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Accurate preoperative prediction of low nuclear grade ductal carcinoma in situ (DCIS) is crucial for optimizing treatment and patient management.
• Current diagnostic methods, often relying on invasive biopsies, face limitations due to DCIS heterogeneity and incomplete tumor characterization.

Purpose of the Study:

• To develop and validate an ensemble machine learning model for the preoperative diagnosis of low nuclear grade DCIS.
• To assess the model's performance in integrating diverse data types, including clinical, imaging, and Radiomic features.

Main Methods:

• An ensemble machine learning model was developed using Elastic Net, Generalized Linear Models with Boosting (glmboost), and Ranger algorithms.
• The model integrated preoperative clinical data, ultrasound images, mammography images, and Radiomic scores from 241 DCIS cases.
• Performance was evaluated using AUC, integrated discrimination improvement, and net reclassification improvement.

Main Results:

• The ensemble model achieved an Area Under the Curve (AUC) of 0.92 on the validation set, outperforming models using clinical data alone.
• Significant improvements in integrated discrimination improvement and net reclassification improvement were observed (p < 0.001).
• The Radiomic ensemble model demonstrated effectiveness in stratifying DCIS patients based on disease-free survival risk.

Conclusions:

• Integrating Radiomic features into machine learning models significantly enhances the preoperative prediction of low nuclear grade DCIS.
• This approach offers improved diagnostic accuracy and personalized risk stratification, paving the way for tailored treatment and clinical management strategies for DCIS.