Corneal confocal microscopy (CCM) has been advocated as a non-invasive technique for objective diagnosis of very early neuropathy in patients by scanning the corneal subbasal nerve plexus. The obtained images provide a range of research opportunities to be explored. Current research revolves around providing automated solutions for nerve segmentation in CCM images. A common problem in CCM nerve segmentation is that the total number of nerve pixels is much smaller than the total number of background pixels. Thus, accuracy is not a very suitable evaluation metric for this problem. At the same time, the loss function used in the deep learning network influences the performance of the network. It is an important constituent of a deep learning segmentation network. In this project, we address the problem of low sensitivity of nerves in automatic segmentation caused by imbalanced pixel distribution in the CCM images. We evaluate different loss functions for this problem.
A comparison is made between cross entropy, dice loss, and Tversky loss using the most popular CNN for biomedical image segmentation, U-Net.
Members
Lead Principal Investigator (LPI):
Student:
-
Tooba Salahuddin, PhD student, Computer Science and Engineering
Department, Qatar University (Email: t.salahuddin@qu.edu.qa)
Publications
T. Salahuddin and U. Qidwai, "Evaluation of Loss Functions for
Segmentation of Corneal Nerves," 2020 IEEE-EMBS Conference on
Biomedical Engineering and Sciences (IECBES), 2021, pp. 533-537, doi:
10.1109/IECBES48179.2021.9398843.
.
Funding (if applicable)
Funded By:
Collaborating Institution(s)
Datasets (if applicable)
The datasets are available upon request from the corresponding author
of the following publications:
M. A. Dabbah, J. Graham, M. Tavakoli, Y. Petropoulos, and R. A. Malik,
“Nerve fibre extraction in confocal corneal microscopy images for
human diabetic neuropathy detection using gabor filters,” Med. Image
Underst. Anal., vol. 254–258, 2009
Source Code (if applicable)