An Attention-PiDi-UNet and Focal Active Contour loss for Biomedical Image Segmentation
Minh-Nhat Trinh, Do-Hai-Ninh Nham, Van-Truong Pham, Thi-Thao Tran · 2022 RIVF International Conference on Computing and Communication Technologies (RIVF) · 2022
Biomedical image analysis has been of paramount importance in modern computer vision recently. While up-to-date procedures have notched up success in medical imaging, there is still necessity to optimise the network and the loss function. Motivated by the PiDiNet architecture and the Active Contour methods, we propose a new Attention-PiDi-UNet architecture and a new Focal Active Contour loss. Our proposed method is evaluated on the three popular datasets: the MRI cardiac ACDC dataset (3-D images), the Skin Lesion ISIC 2018 dataset and the PH2 dataset (both in 2-D images). Several experiments have comfirmed our proposed method effectiveness with outstanding segmentation results.