EAR-UNet: A deep learning-based approach for segmentation of tympanic membranes from otoscopic images

Pham, Van Truong; Tran, Thi Thao; Wang, Pa Chun; Chen, Po‐Yu; Lo, Men Tzung

doi:10.1016/j.artmed.2021.102065

Cited by 45 publications

(22 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…32 Focus on the TM and localising anatomical or pathological characteristics with segmentation models has achieved substantial improvements in accuracy while considering various middle ear conditions. 33 This review has several strengths. Firstly, broad inclusion criteria were used to explore articles clinically relevant to the study question.…”

Section: Aiversushumanclassificationmentioning

confidence: 99%

See 1 more Smart Citation

Artificial intelligence to classify ear disease from otoscopy: A systematic review and meta‐analysis

Habib

Kajbafzadeh

Hasan

et al. 2022

Clinical Otolaryngology

View full text Add to dashboard Cite

Objectives:To summarise the accuracy of artificial intelligence (AI) computer vision algorithms to classify ear disease from otoscopy.Design:Systematic review and meta-analysis.Methods:Using the PRISMA guidelines, nine online databases were searched for articles that used AI computer vision algorithms developed from various methods (convolutional neural networks, artificial neural networks, support vector machines, decision trees and k-nearest neighbours) to classify otoscopic images. Diagnostic classes of interest: normal tympanic membrane, acute otitis media (AOM), otitis media with effusion (OME), chronic otitis media (COM) with or without perforation, cholesteatoma and canal obstruction.Mainoutcomemeasures:Accuracy to correctly classify otoscopic images compared to otolaryngologists (ground truth). The Quality Assessment of Diagnostic Accuracy Studies Version 2 tool was used to assess the quality of methodology and risk of bias.Results: Thirty-nine articles were included. Algorithms achieved 90.7% (95%CI: 90.1-91.3%) accuracy to difference between normal or abnormal otoscopy images in 14 studies. The most common multiclassification algorithm (3 or more diagnostic classes) achieved 97.6% (95%CI: 97.3-97.9%) accuracy to differentiate between normal, AOM and OME in three studies. AI algorithms outperformed human assessors to classify otoscopy images achieving 93.4% (95%CI: 90.5-96.4%) versus 73.2% (95%CI: 67.9-78.5%) accuracy in three studies. Convolutional neural networks achieved the highest accuracy compared to other classification methods. Conclusion:AI can classify ear disease from otoscopy. A concerted effort is requiredto establish a comprehensive and reliable otoscopy database for algorithm training.An AI-supported otoscopy system may assist health care workers, trainees and primary care practitioners with less otology experience identify ear disease.

show abstract

Section: Aiversushumanclassificationmentioning

confidence: 99%

“…, achieving 96% accuracy to localise the TM in normal, AOM, OME and COM otoscopic images 33. Transitioning the AI-based computer vision algorithm for otoscopy from a virtual environment to the clinical frontline will depend on real-world test performance and applicability in daily clinical practice.…”

mentioning

confidence: 99%

Artificial intelligence to classify ear disease from otoscopy: A systematic review and meta‐analysis

Habib

Kajbafzadeh

Hasan

et al. 2022

Clinical Otolaryngology

View full text Add to dashboard Cite

show abstract

“…Subsequent researchers developed this UNet architecture with several modifications. UNet development with replacement of block contents is done at [10][11][12][13]. The replacements include residual block [11,13,14], two-path residual blocks and applying variations in the use of these blocks in the contracting and expanding sections [10], and dilated convolution [12,15].…”

Section: Related Workmentioning

confidence: 99%

“…UNet development with replacement of block contents is done at [10][11][12][13]. The replacements include residual block [11,13,14], two-path residual blocks and applying variations in the use of these blocks in the contracting and expanding sections [10], and dilated convolution [12,15]. In addition, the skip connection section has also been modified with an attention gate at [13].…”

Section: Related Workmentioning

confidence: 99%

SDA-UNET2.5D: Shallow Dilated with Attention Unet2.5D for Brain Tumor Segmentation

2022

IJIES

View full text Add to dashboard Cite

Many studies have been carried out to segmentation brain tumors on 3D Magnetic Resonance Imaging (MRI) images with 3D or 2D approaches. The 3D approach pays attention to the interrelationships between slices in a 3D image. However, this requires high resources, while the 2D approach requires lower resources but ignores the voxel relationship in 3D space. The 2.5D approach seeks to combine the lightness of the 2D approach and the voxel interconnection of the 3D approach. This article proposes SDA-UNet2.5D, a shallow UNet 2.5D architecture that pays attention to the interconnectedness of 3D images by involving five slices to get one slice of segmentation prediction results. The architecture is trained using the Brain Tumor Segmentation (BraTS) 2018, 2019, and 2020 datasets. Compared to other architectures, this proposed architecture has a high segmentation speed with 4.05-4.24 seconds to segment one patient data. Online validation resulted in superior average dice

show abstract

“…Among them, machine learning-based methods are considered as a promising approach since they use features extracted from MRI images, which can learn from images during the training stage. In recent years, thanks to the great successes of deep learning, convolutional neural network (CNN)-based approaches have shown outstanding performance in image segmentation [10]. In the field of medical image segmentation, the work proposed by Long et al namely the Fully Convolutional Network (FCN) [11] -a CNN based architecture, has attracted a lot of research [12].…”

Section: Introductionmentioning

confidence: 99%

Brain Tumor Segmentation Based on U-Net With Image Driven Level Set Loss

Pham

Tran

2021

Vietnam J. Sci. Technol.

View full text Add to dashboard Cite

This paper presents an approach for brain tumor segmentation based on deep neural networks. The paper proposes to utilize U-Net as an architecture of the approach to capture the fine and soars information from input images. Especially, to train the network, instead of using commonly used cross-entropy loss, dice loss or both, in this study, we propose to employ a new loss function including Level set loss and Dice loss function. The level set loss is inspired from Mumford-Shah functional for unsupervised task. Meanwhile, the Dice loss function measures the similarity between the predicted mask and desired mask. The proposed approach is then applied to segment brain tumor from MRI images as well as evaluated and compared with other approaches on a dataset of nearly 4000 brain MRI scans. Experiment results show that the proposed approach achieves high performance in terms of Dice coefficient and Intersection over Union (IoU) scores.

show abstract

EAR-UNet: A deep learning-based approach for segmentation of tympanic membranes from otoscopic images

Cited by 45 publications

References 24 publications

Artificial intelligence to classify ear disease from otoscopy: A systematic review and meta‐analysis

Artificial intelligence to classify ear disease from otoscopy: A systematic review and meta‐analysis

SDA-UNET2.5D: Shallow Dilated with Attention Unet2.5D for Brain Tumor Segmentation

Brain Tumor Segmentation Based on U-Net With Image Driven Level Set Loss

Contact Info

Product

Resources

About