Dhaval Dilip Kadia


This presentation was given live via Zoom at at 9:00 a.m. (Eastern Time) on Wednesday, April 22.



Download Project (894 KB)


The recent developments in the field of Medical Imaging, Deep Learning, is a crucial technology to accelerate medical tasks and perform them precisely and automatically. 3D lung segmentation has a significant role in removing the unnecessary volume of 3D CT scans and segments the actual volume of the lungs in three dimensions, to simplify the 3D CT scan for further tasks. Recently, the deep learning network such as U-Net and its variants provides excellent results for biomedical image segmentation. We propose a novel deep neural network architecture based on UNet, for the 3D lung segmentation task. The proposed model helps learn spatial dependencies in 3D and increases the propagation of volumetric information. We have investigated our network with different architectural modules, learning strategy, activation functions, optimizers, loss functions, and appropriate hyperparameters. Our proposed deep neural network is trained on the publicly available dataset - LUNA16 and achieves state-of-the-art performance on the VESSEL12 dataset and the testing set of LUNA16.

Publication Date


Project Designation

Graduate Research

Primary Advisor

Van Tam Nguyen, Vijayan K. Asari

Primary Advisor's Department

Computer Science


Stander Symposium project, College of Arts and Sciences

United Nations Sustainable Development Goals

Good Health and Well-Being

UNet-based Deep Neural Network for 3D Lung Segmentation