Supporting intraoperative margin assessment using deep learning for automatic tumour segmentation in breast...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Complete tumor removal is crucial for curative breast cancer surgery to prevent recurrence.
• Intraoperative margin assessment (IMA) using [18F]FDG micro-PET-CT shows potential for improving surgical outcomes.
• Accurate delineation of invasive carcinoma in lumpectomy specimens is essential for effective IMA.

Purpose of the Study:

• To develop and validate an AI model for delineating invasive carcinoma in [18F]FDG micro-PET-CT lumpectomy images.
• To assess the model's performance in predicting margin status compared to histopathology and human interpretation.
• To advance the development of a decision-support system for AI-assisted IMA in breast cancer surgery.

Main Methods:

• A 2D Residual U-Net model was trained on 53 breast cancer lumpectomy images with histopathology-defined tumor segmentations.
• The model underwent five-fold cross-validation, achieving a Dice similarity coefficient of 0.71 ± 0.20 for tumor segmentation.
• An ensemble model was created to segment tumors and predict margin status, evaluated on 31 independent cases.

Main Results:

• The AI model achieved a Dice similarity coefficient of 0.71 ± 0.20 for segmenting invasive carcinoma in micro-PET-CT images.
• The ensemble model predicted margin status with an F1 score of 84% on a separate test set.
• The AI model's performance in predicting margin status closely matched the average performance of seven human physicians.

Conclusions:

• The developed AI model demonstrates high accuracy in delineating breast cancer tumors and predicting margin status from micro-PET-CT lumpectomy specimens.
• This AI-driven approach shows significant promise as a decision-support tool to enhance intraoperative margin assessment in breast cancer surgery.
• The findings support the clinical adoption of AI-assisted micro-PET-CT for improved intraoperative margin assessment, potentially leading to better surgical outcomes and reduced recurrence rates.