Cargando…

An Edge-Based Selection Method for Improving Regions-of-Interest Localizations Obtained Using Multiple Deep Learning Object-Detection Models in Breast Ultrasound Images

Computer-aided diagnosis (CAD) systems can be used to process breast ultrasound (BUS) images with the goal of enhancing the capability of diagnosing breast cancer. Many CAD systems operate by analyzing the region-of-interest (ROI) that contains the tumor in the BUS image using conventional texture-b...

Descripción completa

Detalles Bibliográficos
Autores principales: Daoud, Mohammad I., Al-Ali, Aamer, Alazrai, Rami, Al-Najar, Mahasen S., Alsaify, Baha A., Ali, Mostafa Z., Alouneh, Sahel
Formato: Online Artículo Texto
Lenguaje:English
Publicado: MDPI 2022
Materias:
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9500621/
https://www.ncbi.nlm.nih.gov/pubmed/36146070
http://dx.doi.org/10.3390/s22186721
_version_ 1784795265245380608
author Daoud, Mohammad I.
Al-Ali, Aamer
Alazrai, Rami
Al-Najar, Mahasen S.
Alsaify, Baha A.
Ali, Mostafa Z.
Alouneh, Sahel
author_facet Daoud, Mohammad I.
Al-Ali, Aamer
Alazrai, Rami
Al-Najar, Mahasen S.
Alsaify, Baha A.
Ali, Mostafa Z.
Alouneh, Sahel
author_sort Daoud, Mohammad I.
collection PubMed
description Computer-aided diagnosis (CAD) systems can be used to process breast ultrasound (BUS) images with the goal of enhancing the capability of diagnosing breast cancer. Many CAD systems operate by analyzing the region-of-interest (ROI) that contains the tumor in the BUS image using conventional texture-based classification models and deep learning-based classification models. Hence, the development of these systems requires automatic methods to localize the ROI that contains the tumor in the BUS image. Deep learning object-detection models can be used to localize the ROI that contains the tumor, but the ROI generated by one model might be better than the ROIs generated by other models. In this study, a new method, called the edge-based selection method, is proposed to analyze the ROIs generated by different deep learning object-detection models with the goal of selecting the ROI that improves the localization of the tumor region. The proposed method employs edge maps computed for BUS images using the recently introduced Dense Extreme Inception Network (DexiNed) deep learning edge-detection model. To the best of our knowledge, our study is the first study that has employed a deep learning edge-detection model to detect the tumor edges in BUS images. The proposed edge-based selection method is applied to analyze the ROIs generated by four deep learning object-detection models. The performance of the proposed edge-based selection method and the four deep learning object-detection models is evaluated using two BUS image datasets. The first dataset, which is used to perform cross-validation evaluation analysis, is a private dataset that includes 380 BUS images. The second dataset, which is used to perform generalization evaluation analysis, is a public dataset that includes 630 BUS images. For both the cross-validation evaluation analysis and the generalization evaluation analysis, the proposed method obtained the overall ROI detection rate, mean precision, mean recall, and mean F1-score values of [Formula: see text] , [Formula: see text] , [Formula: see text] , and [Formula: see text] , respectively. Moreover, the results show that the proposed edge-based selection method outperformed the four deep learning object-detection models as well as three baseline-combining methods that can be used to combine the ROIs generated by the four deep learning object-detection models. These findings suggest the potential of employing our proposed method to analyze the ROIs generated using different deep learning object-detection models to select the ROI that improves the localization of the tumor region.
format Online
Article
Text
id pubmed-9500621
institution National Center for Biotechnology Information
language English
publishDate 2022
publisher MDPI
record_format MEDLINE/PubMed
spelling pubmed-95006212022-09-24 An Edge-Based Selection Method for Improving Regions-of-Interest Localizations Obtained Using Multiple Deep Learning Object-Detection Models in Breast Ultrasound Images Daoud, Mohammad I. Al-Ali, Aamer Alazrai, Rami Al-Najar, Mahasen S. Alsaify, Baha A. Ali, Mostafa Z. Alouneh, Sahel Sensors (Basel) Article Computer-aided diagnosis (CAD) systems can be used to process breast ultrasound (BUS) images with the goal of enhancing the capability of diagnosing breast cancer. Many CAD systems operate by analyzing the region-of-interest (ROI) that contains the tumor in the BUS image using conventional texture-based classification models and deep learning-based classification models. Hence, the development of these systems requires automatic methods to localize the ROI that contains the tumor in the BUS image. Deep learning object-detection models can be used to localize the ROI that contains the tumor, but the ROI generated by one model might be better than the ROIs generated by other models. In this study, a new method, called the edge-based selection method, is proposed to analyze the ROIs generated by different deep learning object-detection models with the goal of selecting the ROI that improves the localization of the tumor region. The proposed method employs edge maps computed for BUS images using the recently introduced Dense Extreme Inception Network (DexiNed) deep learning edge-detection model. To the best of our knowledge, our study is the first study that has employed a deep learning edge-detection model to detect the tumor edges in BUS images. The proposed edge-based selection method is applied to analyze the ROIs generated by four deep learning object-detection models. The performance of the proposed edge-based selection method and the four deep learning object-detection models is evaluated using two BUS image datasets. The first dataset, which is used to perform cross-validation evaluation analysis, is a private dataset that includes 380 BUS images. The second dataset, which is used to perform generalization evaluation analysis, is a public dataset that includes 630 BUS images. For both the cross-validation evaluation analysis and the generalization evaluation analysis, the proposed method obtained the overall ROI detection rate, mean precision, mean recall, and mean F1-score values of [Formula: see text] , [Formula: see text] , [Formula: see text] , and [Formula: see text] , respectively. Moreover, the results show that the proposed edge-based selection method outperformed the four deep learning object-detection models as well as three baseline-combining methods that can be used to combine the ROIs generated by the four deep learning object-detection models. These findings suggest the potential of employing our proposed method to analyze the ROIs generated using different deep learning object-detection models to select the ROI that improves the localization of the tumor region. MDPI 2022-09-06 /pmc/articles/PMC9500621/ /pubmed/36146070 http://dx.doi.org/10.3390/s22186721 Text en © 2022 by the authors. https://creativecommons.org/licenses/by/4.0/Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
spellingShingle Article
Daoud, Mohammad I.
Al-Ali, Aamer
Alazrai, Rami
Al-Najar, Mahasen S.
Alsaify, Baha A.
Ali, Mostafa Z.
Alouneh, Sahel
An Edge-Based Selection Method for Improving Regions-of-Interest Localizations Obtained Using Multiple Deep Learning Object-Detection Models in Breast Ultrasound Images
title An Edge-Based Selection Method for Improving Regions-of-Interest Localizations Obtained Using Multiple Deep Learning Object-Detection Models in Breast Ultrasound Images
title_full An Edge-Based Selection Method for Improving Regions-of-Interest Localizations Obtained Using Multiple Deep Learning Object-Detection Models in Breast Ultrasound Images
title_fullStr An Edge-Based Selection Method for Improving Regions-of-Interest Localizations Obtained Using Multiple Deep Learning Object-Detection Models in Breast Ultrasound Images
title_full_unstemmed An Edge-Based Selection Method for Improving Regions-of-Interest Localizations Obtained Using Multiple Deep Learning Object-Detection Models in Breast Ultrasound Images
title_short An Edge-Based Selection Method for Improving Regions-of-Interest Localizations Obtained Using Multiple Deep Learning Object-Detection Models in Breast Ultrasound Images
title_sort edge-based selection method for improving regions-of-interest localizations obtained using multiple deep learning object-detection models in breast ultrasound images
topic Article
url https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9500621/
https://www.ncbi.nlm.nih.gov/pubmed/36146070
http://dx.doi.org/10.3390/s22186721
work_keys_str_mv AT daoudmohammadi anedgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alaliaamer anedgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alazrairami anedgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alnajarmahasens anedgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alsaifybahaa anedgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alimostafaz anedgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alounehsahel anedgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT daoudmohammadi edgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alaliaamer edgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alazrairami edgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alnajarmahasens edgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alsaifybahaa edgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alimostafaz edgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages
AT alounehsahel edgebasedselectionmethodforimprovingregionsofinterestlocalizationsobtainedusingmultipledeeplearningobjectdetectionmodelsinbreastultrasoundimages