AI-Based Ultrasound Nomogram for Differentiating Invasive from Non-Invasive Breast Cancer Masses

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Artificial Intelligence in Medicine

Background:

• Distinguishing ductal carcinoma in situ (DCIS) from invasive ductal carcinoma (IDC) is crucial for effective breast cancer treatment.
• Ultrasound is a primary imaging modality, but differentiating these lesions can be challenging.
• AI-driven analysis of imaging features offers potential for improved diagnostic accuracy.

Purpose of the Study:

• To develop and validate a predictive nomogram for differentiating mass-type DCIS from IDC using AI-based BI-RADS lexicons and lesion-to-nipple distance (LND) on ultrasound.
• To assess the nomogram's performance in terms of discrimination and calibration.

Main Methods:

• A retrospective analysis of 175 pathologically confirmed malignant breast lesions (26 DCIS, 149 IDC) from 170 women.
• Integration of AI-derived BI-RADS features (S-Detect) and ultrasound LND measurements.
• Development of a logistic regression-based nomogram, with performance evaluated using ROC curves and calibration plots.

Main Results:

• Smaller lesion size, irregular shape, LND ≤ 3 cm, and non-hypoechoic echogenicity were independent predictors of DCIS.
• The AI-based nomogram demonstrated strong discriminative power (AUC training: 0.851, AUC validation: 0.842).
• Excellent calibration was observed, with non-significant Hosmer-Lemeshow tests and low mean absolute errors.

Conclusions:

• The developed AI-based nomogram is a reliable tool for non-invasively distinguishing mass-type DCIS from IDC.
• This predictive model can enhance preoperative planning and improve diagnostic accuracy in breast cancer.
• AI integration with ultrasound features shows promise for advancing breast cancer diagnostics.