Bilkent University
Department of Computer Engineering


Distance transforms for organ segmentation in thorax CT scans


Haya Shamim Khan Khattak
Master Student
(Supervisor: Assoc. Prof. Çiğdem Gündüz Demir )
Computer Engineering Department
Bilkent University

Abstract: Organ segmentation plays a crucial role for planning therapy in radiation oncology. In this work, the role of various distance transforms has been explored to show that using distance transform regression alone or in conjunction with classification improves overall performance of the organ segmentation network. These maps can be the shortest distance between each pixel and the center of the organ, or to the boundary of the organ, or between two organs; such as heart and lungs, and signed or unsigned. It is shown that when used in a single task regression model, they may improve the segmentation results. Moreover, when used in a multitask architecture with classification being the other task, they act as regularizers for the classification task and yield improved segmentations.


DATE: 04 April 2022, Monday @ 15:50 Zoom