Dual-Branch Fully Convolutional Network (DB-FCN): An End-to-End Model for Nucleus Detection

Abstract

The analysis of medical images depends on nucleus detection as an essential tool for disease diagnosis especially in cancer treatment and histopathology. Modern nucleus detection systems face obstacles in handling differences between cell shapes and sizes together with tissue structures. We created DB-FCN as a new end-to-end deep learning model made for efficient and accurate nucleus segmentation that utilizes dual-branch architectural design. DB-FCN uses two parallel feature extraction branches which helps generate multiple resolution inputs that improve detection efficiency when processing difficult visual information.

Keywords: Deep Learning, Fully Convolutional Network (FCN), Nucleus Detection, Image Segmentation, Dual-Branch Architecture, End-to-End Learning