Esophageal Tumor Segmentation in CT Images Using DDAUnet

Authors

Sahar Yousefi, Hessam Sokooti, Mohamed S. Elmahdy, Irene M. Lips, Mohammad T. Manzuri Shalmani, Roel T. Zinkstok, Frank J. W. M. Dankers, and Marius Staring

Overview

Esophageal tumor delineation in CT is challenging because tumors can have low contrast against nearby tissue, anatomy varies between patients, and scans may include air pockets or feeding tubes. DDAUnet is a convolutional neural network designed to automatically segment the gross tumor volume using CT images alone.

Approach and results

The network combines dense blocks with dilated convolutions and spatial and skip-connection attention gates. Dilated layers expand the receptive field, while the attention gates help focus on relevant regions and feature maps. The study evaluated the method on 792 CT scans from 288 patients. On 287 test scans, the paper reports a mean Dice similarity coefficient (DSC) of 0.79 ± 0.20.

Scans

792

CT scans in the study

Patients

288

Distinct patients

Mean test DSC

0.79 ± 0.20

Across 287 test scans

Project links