Next Article in Journal
The Potential Use of Radiomics with Pre-Radiation Therapy MR Imaging in Predicting Risk of Pseudoprogression in Glioblastoma Patients
Next Article in Special Issue
Personal Heart Health Monitoring Based on 1D Convolutional Neural Network
Previous Article in Journal / Special Issue
Hand Motion-Aware Surgical Tool Localization and Classification from an Egocentric Camera
Article

Testing Segmentation Popular Loss and Variations in Three Multiclass Medical Imaging Problems

Dei/FCT/CISUC, University of Coimbra, Polo II, 3030-290 Coimbra, Portugal
Academic Editor: Yudong Zhang
Received: 22 December 2020 / Revised: 16 January 2021 / Accepted: 22 January 2021 / Published: 27 January 2021
(This article belongs to the Special Issue Deep Learning in Medical Image Analysis)
Image structures are segmented automatically using deep learning (DL) for analysis and processing. The three most popular base loss functions are cross entropy (crossE), intersect-over-the-union (IoU), and dice. Which should be used, is it useful to consider simple variations, such as modifying formula coefficients? How do characteristics of different image structures influence scores? Taking three different medical image segmentation problems (segmentation of organs in magnetic resonance images (MRI), liver in computer tomography images (CT) and diabetic retinopathy lesions in eye fundus images (EFI)), we quantify loss functions and variations, as well as segmentation scores of different targets. We first describe the limitations of metrics, since loss is a metric, then we describe and test alternatives. Experimentally, we observed that DeeplabV3 outperforms UNet and fully convolutional network (FCN) in all datasets. Dice scored 1 to 6 percentage points (pp) higher than cross entropy over all datasets, IoU improved 0 to 3 pp. Varying formula coefficients improved scores, but the best choices depend on the dataset: compared to crossE, different false positive vs. false negative weights improved MRI by 12 pp, and assigning zero weight to background improved EFI by 6 pp. Multiclass segmentation scored higher than n-uniclass segmentation in MRI by 8 pp. EFI lesions score low compared to more constant structures (e.g., optic disk or even organs), but loss modifications improve those scores significantly 6 to 9 pp. Our conclusions are that dice is best, it is worth assigning 0 weight to class background and to test different weights on false positives and false negatives. View Full-Text
Keywords: computers in medicine; segmentation; machine learning; deep learning; MRI computers in medicine; segmentation; machine learning; deep learning; MRI
Show Figures

Figure 1

MDPI and ACS Style

Furtado, P. Testing Segmentation Popular Loss and Variations in Three Multiclass Medical Imaging Problems. J. Imaging 2021, 7, 16. https://0-doi-org.brum.beds.ac.uk/10.3390/jimaging7020016

AMA Style

Furtado P. Testing Segmentation Popular Loss and Variations in Three Multiclass Medical Imaging Problems. Journal of Imaging. 2021; 7(2):16. https://0-doi-org.brum.beds.ac.uk/10.3390/jimaging7020016

Chicago/Turabian Style

Furtado, Pedro. 2021. "Testing Segmentation Popular Loss and Variations in Three Multiclass Medical Imaging Problems" J. Imaging 7, no. 2: 16. https://0-doi-org.brum.beds.ac.uk/10.3390/jimaging7020016

Find Other Styles
Note that from the first issue of 2016, MDPI journals use article numbers instead of page numbers. See further details here.

Article Access Map by Country/Region

1
Search more from Scilit
 
Search
Back to TopTop