A sensitivity analysis of probability maps in deep-learning-based anatomical segmentation.

Journal: Journal of applied clinical medical physics
Published Date:

Abstract

PURPOSE: Deep-learning-based segmentation models implicitly learn to predict the presence of a structure based on its overall prominence in the training dataset. This phenomenon is observed and accounted for in deep-learning applications such as natural language processing but is often neglected in segmentation literature. The purpose of this work is to demonstrate the significance of class imbalance in deep-learning-based segmentation and recommend tuning of the neural network optimization objective.

Authors

  • Noah Bice
    Department of Radiological Sciences, UT Health San Antonio, San Antonio, TX, 78229, USA.
  • Neil Kirby
    Department of Radiological Sciences, UT Health San Antonio, San Antonio, TX, 78229, USA.
  • Ruiqi Li
    Department of Radiation Oncology, UT Health San Antonio, San Antonio, TX, USA.
  • Dan Nguyen
    University of Massachusetts Chan Medical School, Worcester, Massachusetts.
  • Tyler Bahr
    Department of Radiological Sciences, UT Health San Antonio, San Antonio, TX, 78229, USA.
  • Christopher Kabat
    Department of Radiation Oncology, UT Health San Antonio, San Antonio, TX, USA.
  • Pamela Myers
    Department of Radiation Oncology, UT Health San Antonio, San Antonio, TX, USA.
  • Niko Papanikolaou
    Department of Radiological Sciences, UT Health San Antonio, San Antonio, TX, 78229, USA.
  • Mohamad Fakhreddine
    Department of Radiological Sciences, UT Health San Antonio, San Antonio, TX, 78229, USA.