Risk-Adapted Breast Cancer Screening: Combining DOT and Ultrasound with AI to Overcome Mammography Limitations
Shaik Abdul Hafeez, R Mahaveerakannan · 2025
For more than four decades, clinicians have used mammography, the gold standard imaging technique, to screen populations for breast cancer. Nevertheless, the uncritical character of population-based screening is challenged by mammography's sensitivity limitations and high false-positive rates, especially in high-risk women. There may be new possibilities for screening approaches thanks to recent developments in breast imaging technology, such as contrast material enhanced mammography (CEM), ultrasound (US) (automated-breast US, Doppler, elastography US), besides MRI in particular. Additionally, risk-adapted screening could be improved through the combination of AI and radiomics methods. One exciting new tool for studying tumor angiogenesis is diffuse optical tomography, or DOT. Reconstructing a breast lesion's DOT function map, however, is an illposed and undecided inverse operation. To enhance the localization and precision of DOT reconstruction, a co-registered ultrasonography (US) technology that offers structural information about the breast lesion can be utilized. Further improvement in cancer detection based on DOT alone is possible with the well-known US features of benign and malignant breast lesions. to build a new neural network for diagnosis by combining US features extracted from six different neural networks (Xception, InceptionV3, VGG16, MobileNet, and ResNet50) with images reconstructed from a DOT deep learning auto-encoder-based model. This approach was inspired by a fusion model deep learning strategy. The F1-score, recall, precision, and accuracy were the metrics utilized to assess each model. Collection of records Improved accuracy in both the suggested model and the fine-tuned pretrained models was a result of the beneficial effects of boosting, preprocessing, and balancing on breast cancer detection and classification.