Early Breast Cancer Detection based on High-Boost Filtration, Data Fusion and Modified Object Detection.

While object detection has made substantial progress in medical image anomaly identification, the deep learning method fails to extract breast cancer images' hidden features. We thus applied traditional image processing techniques, such as high-boost filtration and image subtraction, to extract such hidden information, and generated four processed image datasets. The four types of generated image datasets, together with the original dataset, can be merged via data fusion technology. Our data fusion method first converts all the source images to matrices and then concatenates the matrices. The original dataset, the four types of generated image datasets and the result matrix of data fusion are trained separately with and without attention mechanisms. The data fusion method can increase the mAP by 19.46 percent when combined with the attention mechanisms. The data fusion method that concatenates the source images and embeds the complete information from the source images into the data fusion result may inspire future research in deep learning with matrix concatenation-based data fusion.