GFNet: A Deep Learning Framework for Breast Mass Detection

Background: Breast mass is one of the main symptoms breast cancer. Effective and accurate detection masses at an early stage would be great value for clinical cancer analysis. Methods: We developed a novel framework named GFNet. The GFNet comprised three modules, including patch extraction, feature detection. high robustness generality that can self-adapted to images collected by different imaging devices. patch-based deployed improve performance. A extraction technique based on gradient field convergence features (GFCF) proposed enhance information and, therefore, provide useful following module. false positives reduction method designed combining texture morphological in patch. This first attempt fusing positive reduction. Results: Compared other state-of-the-art methods, showed best performance CBIS-DDSM INbreast with accuracy 0.90 2.91 per image (FPI) 0.99 only 0.97 FPI, respectively. Conclusions: effective tool detecting mass.