CNN-Based Cross-Modality Fusion for Enhanced Breast Cancer Detection Using Mammography and Ultrasound.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer remains a significant global health concern, necessitating improved early detection methods.
• Current non-invasive imaging techniques like mammography and ultrasound have limitations in diagnostic accuracy when used independently.
• Combining imaging modalities offers potential to overcome individual limitations.

Purpose of the Study:

• To enhance breast cancer detection accuracy by integrating mammography and ultrasound imaging data.
• To develop and evaluate advanced convolutional neural network (CNN) architectures for cross-modality image fusion.
• To compare the performance of different CNN-based models for breast cancer classification.

Main Methods:

• Utilized public breast imaging datasets (RSNA, PAS, Kaggle) for training and validation.
• Employed data augmentation to address class imbalance in ultrasound data.
• Developed and compared three CNN-based approaches: pre-trained CNNs with ML classifiers, transfer learning CNNs, and a custom 17-layer CNN.

Main Results:

• The custom 17-layer CNN achieved the highest performance, with an accuracy of 0.964 and a Kappa score of 0.927.
• Transfer learning CNNs showed moderate results (accuracy 0.846, Kappa 0.694).
• Cross-modality fusion effectively leveraged complementary information from mammography and ultrasound.

Conclusions:

• Tailored CNN architectures combined with cross-modality imaging significantly improve breast cancer detection reliability.
• The custom-designed CNN model presents a practical solution for early breast cancer diagnosis.
• This approach has the potential to reduce diagnostic errors and improve patient outcomes.