Exploring a frequency-domain attention-guided cascade U-Net: Towards spatially tunable segmentation of vasculature

Nan Mu, Zonghan Lyu, Xiaoming Zhang, Robert McBane, Aditya S. Pandey, Jingfeng Jiang

Abstract: Highlights•This paper proposes multiple attention modules in frequency domain for CNN model.•Our study is the first to explore spatially tunable segmentation of medical images.•We designed frequency-domain-based channel, spatial, and content attention modules.•We exploited feasibility of quantitatively tuning high/low frequency information.•We analyzed how tuning the frequency helped us to achieve user-desirable outcomes.