Integrating Peritumoral and Intratumoral Radiomics with Deep Learning for Preoperative Prediction of Lymphovascular...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Lymphovascular invasion (LVI) is a key prognostic indicator in breast cancer (BC), but its preoperative assessment using non-invasive methods is challenging.
• Accurate prediction of LVI is crucial for tailoring treatment and surgical planning in BC patients.

Purpose of the Study:

• To design and validate a comprehensive artificial intelligence (AI) system for the preoperative prediction of LVI in invasive breast cancer (IBC).
• To integrate intratumoral and peritumoral radiomic features, deep learning (DL) insights, and clinical data for enhanced LVI prediction accuracy.

Main Methods:

• A retrospective study involving 496 IBC patients with preoperative dynamic contrast-enhanced MRI (DCE-MRI).
• Extraction of intratumoral/peritumoral radiomic features, application of a ResNet-50 DL model, and identification of clinical risk factors.
• Development of an ensemble AI model combining radiomic, DL, and clinical data using LASSO feature selection.

Main Results:

• The ensemble AI model achieved high performance, with AUCs of 0.951 (training) and 0.929 (validation) for LVI prediction.
• The integrated model demonstrated superior discriminative ability compared to single-region radiomic approaches.
• Decision curve analysis (DCA) and calibration curves confirmed the model's clinical utility and accuracy.

Conclusions:

• The AI-driven ensemble model accurately predicts preoperative LVI in IBC patients using DCE-MRI data.
• This AI system has the potential to significantly improve surgical planning and adjuvant therapy decisions.
• Combining radiomics, DL, and clinical factors offers a powerful approach for non-invasive LVI assessment in breast cancer.