BI-RADS-NET-V2: A Composite Multi-Task Neural Network for Computer-Aided Diagnosis of Breast Cancer in Ultrasound...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Computer-aided diagnosis (CADx) using explainable artificial intelligence (XAI) is crucial for enhancing radiologist trust and improving diagnostic accuracy.
• Breast cancer diagnosis in ultrasound images requires accurate differentiation between malignant and benign tumors.

Purpose of the Study:

• To propose BI-RADS-Net-V2, a novel machine learning approach for fully automatic breast cancer diagnosis in ultrasound images.
• To provide both semantic and quantitative explanations for diagnostic decisions, aligning with clinical standards.

Main Methods:

• Development of BI-RADS-Net-V2, a machine learning model for breast cancer diagnosis.
• Integration of Breast Imaging Reporting and Data System (BI-RADS) morphological features for explanation generation.
• Validation using a dataset of 1,192 Breast Ultrasound (BUS) images.

Main Results:

• BI-RADS-Net-V2 accurately distinguishes malignant tumors from benign ones in ultrasound images.
• The system provides clinically relevant semantic and quantitative explanations based on BI-RADS features.
• Improved diagnosis accuracy and efficiency demonstrated through experimental validation.

Conclusions:

• BI-RADS-Net-V2 enhances breast cancer diagnosis by leveraging AI and clinical knowledge (BI-RADS).
• The explainable AI approach fosters trust and improves consultation efficiency for radiologists.
• The method offers a robust solution for automatic breast cancer diagnosis in ultrasound imaging.