Improving tuberculosis severity assessment in computed tomography images using novel DAvoU-Net segmentation and deep learning framework

Alebiosu David Olayemi, Anuja T. Dharmarathne, Chern Hong Lim

Abstract: Highlights•Multi-Scale Residual Blocks and Receptive Dense Connection improve segmentation.•DAvoU-Net solves the problem of insufficient discriminative features in CT images.•Rotation operation from 2-dimensional to 3-dimensional improves transfer learning.•Three-dimensional neural network is good for tuberculosis severity assessment.