Comparative Analysis of Deep Learning Architectures for Macular Hole Segmentation in OCT Images: A Performance Evaluation of U-Net Variants

This study presents a comprehensive comparison of U-Net variants with different backbone architectures for Macular Hole (MH) segmentation in optical coherence tomography (OCT) images. We evaluated eleven architectures, including U-Net combined with InceptionNetV4, VGG16, VGG19, ResNet152, DenseNet12...

Full description

Saved in:
Bibliographic Details
Main Authors: H. M. S. S. Herath, S. L. P. Yasakethu, Nuwan Madusanka, Myunggi Yi, Byeong-Il Lee
Format: Article
Language:English
Published: MDPI AG 2025-02-01
Series:Journal of Imaging
Subjects:
Online Access:https://www.mdpi.com/2313-433X/11/2/53
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850081758984273920
author H. M. S. S. Herath
S. L. P. Yasakethu
Nuwan Madusanka
Myunggi Yi
Byeong-Il Lee
author_facet H. M. S. S. Herath
S. L. P. Yasakethu
Nuwan Madusanka
Myunggi Yi
Byeong-Il Lee
author_sort H. M. S. S. Herath
collection DOAJ
description This study presents a comprehensive comparison of U-Net variants with different backbone architectures for Macular Hole (MH) segmentation in optical coherence tomography (OCT) images. We evaluated eleven architectures, including U-Net combined with InceptionNetV4, VGG16, VGG19, ResNet152, DenseNet121, EfficientNet-B7, MobileNetV2, Xception, and Transformer. Models were assessed using the Dice coefficient and HD95 metrics on the OIMHS dataset. While HD95 proved unreliable for small regions like MH, often returning ‘nan’ values, the Dice coefficient provided consistent performance evaluation. InceptionNetV4 + U-Net achieved the highest Dice coefficient (0.9672), demonstrating superior segmentation accuracy. Although considered state-of-the-art, Transformer + U-Net showed poor performance in MH and intraretinal cyst (IRC) segmentation. Analysis of computational resources revealed that MobileNetV2 + U-Net offered the most efficient performance with minimal parameters, while InceptionNetV4 + U-Net balanced accuracy with moderate computational demands. Our findings suggest that CNN-based backbones, particularly InceptionNetV4, are more effective than Transformer architectures for OCT image segmentation, with InceptionNetV4 + U-Net emerging as the most promising model for clinical applications.
format Article
id doaj-art-a59fcc005e8b4e31a90e51a09d8737bf
institution DOAJ
issn 2313-433X
language English
publishDate 2025-02-01
publisher MDPI AG
record_format Article
series Journal of Imaging
spelling doaj-art-a59fcc005e8b4e31a90e51a09d8737bf2025-08-20T02:44:39ZengMDPI AGJournal of Imaging2313-433X2025-02-011125310.3390/jimaging11020053Comparative Analysis of Deep Learning Architectures for Macular Hole Segmentation in OCT Images: A Performance Evaluation of U-Net VariantsH. M. S. S. Herath0S. L. P. Yasakethu1Nuwan Madusanka2Myunggi Yi3Byeong-Il Lee4Department of Industry 4.0 Convergence Bionics Engineering, Pukyoung National University, Busan 48513, Republic of KoreaFaculty of Technology, Sri Lanka Technological Campus, Padukka 10500, Sri LankaDigital Healthcare Research Center, Pukyong National University, Busan 48513, Republic of KoreaDepartment of Industry 4.0 Convergence Bionics Engineering, Pukyoung National University, Busan 48513, Republic of KoreaDepartment of Industry 4.0 Convergence Bionics Engineering, Pukyoung National University, Busan 48513, Republic of KoreaThis study presents a comprehensive comparison of U-Net variants with different backbone architectures for Macular Hole (MH) segmentation in optical coherence tomography (OCT) images. We evaluated eleven architectures, including U-Net combined with InceptionNetV4, VGG16, VGG19, ResNet152, DenseNet121, EfficientNet-B7, MobileNetV2, Xception, and Transformer. Models were assessed using the Dice coefficient and HD95 metrics on the OIMHS dataset. While HD95 proved unreliable for small regions like MH, often returning ‘nan’ values, the Dice coefficient provided consistent performance evaluation. InceptionNetV4 + U-Net achieved the highest Dice coefficient (0.9672), demonstrating superior segmentation accuracy. Although considered state-of-the-art, Transformer + U-Net showed poor performance in MH and intraretinal cyst (IRC) segmentation. Analysis of computational resources revealed that MobileNetV2 + U-Net offered the most efficient performance with minimal parameters, while InceptionNetV4 + U-Net balanced accuracy with moderate computational demands. Our findings suggest that CNN-based backbones, particularly InceptionNetV4, are more effective than Transformer architectures for OCT image segmentation, with InceptionNetV4 + U-Net emerging as the most promising model for clinical applications.https://www.mdpi.com/2313-433X/11/2/53optical coherence tomographysegmentationconvolutional neural networksmacular hole
spellingShingle H. M. S. S. Herath
S. L. P. Yasakethu
Nuwan Madusanka
Myunggi Yi
Byeong-Il Lee
Comparative Analysis of Deep Learning Architectures for Macular Hole Segmentation in OCT Images: A Performance Evaluation of U-Net Variants
Journal of Imaging
optical coherence tomography
segmentation
convolutional neural networks
macular hole
title Comparative Analysis of Deep Learning Architectures for Macular Hole Segmentation in OCT Images: A Performance Evaluation of U-Net Variants
title_full Comparative Analysis of Deep Learning Architectures for Macular Hole Segmentation in OCT Images: A Performance Evaluation of U-Net Variants
title_fullStr Comparative Analysis of Deep Learning Architectures for Macular Hole Segmentation in OCT Images: A Performance Evaluation of U-Net Variants
title_full_unstemmed Comparative Analysis of Deep Learning Architectures for Macular Hole Segmentation in OCT Images: A Performance Evaluation of U-Net Variants
title_short Comparative Analysis of Deep Learning Architectures for Macular Hole Segmentation in OCT Images: A Performance Evaluation of U-Net Variants
title_sort comparative analysis of deep learning architectures for macular hole segmentation in oct images a performance evaluation of u net variants
topic optical coherence tomography
segmentation
convolutional neural networks
macular hole
url https://www.mdpi.com/2313-433X/11/2/53
work_keys_str_mv AT hmssherath comparativeanalysisofdeeplearningarchitecturesformacularholesegmentationinoctimagesaperformanceevaluationofunetvariants
AT slpyasakethu comparativeanalysisofdeeplearningarchitecturesformacularholesegmentationinoctimagesaperformanceevaluationofunetvariants
AT nuwanmadusanka comparativeanalysisofdeeplearningarchitecturesformacularholesegmentationinoctimagesaperformanceevaluationofunetvariants
AT myunggiyi comparativeanalysisofdeeplearningarchitecturesformacularholesegmentationinoctimagesaperformanceevaluationofunetvariants
AT byeongillee comparativeanalysisofdeeplearningarchitecturesformacularholesegmentationinoctimagesaperformanceevaluationofunetvariants