A Deep Learning Approach for Meibomian Gland Atrophy Evaluation in Meibography Images

Wang, Jiayun; Yeh, Thao N.; Chakraborty, Rudrasis; Yu, Stella X.; Lin, Meng C.

doi:10.1167/tvst.8.6.37

Cited by 53 publications

(58 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In Figure 5 , the NPID approach was compared against the clinical team (clinical meiboscore), the lead clinical investigator (LCI), and a supervised learning approach. 14 The NPID approach achieved 80.9% overall grading accuracy with ImageNet pretrained model, which outperformed the clinical team grading by 25.9% and lead clinical investigator by 1.3%. The NPID approach accuracy without using ImageNet pretrained model was also provided to show that the pretrained model benefited the performance by gaining around 14% accuracy.…”

Section: Methodsmentioning

confidence: 91%

“…Based on a previous study, 14 University of California, Berkeley Clinical Research Center recruited adult human subjects for a single-visit ocular surface evaluation, which included MG imaging for gland atrophy assessment, during the period from 2012 to 2017. Clinicians used the OCULUS Keratograph 5M (OCULUS, Arlington, WA), a clinical instrument that uses infrared light with wavelength 880 nm for MG imaging 21 to capture MG images of patients’ upper and lower eyelids for both eyes.…”

Section: Methodsmentioning

confidence: 99%

“… Meiboscore grading performance of clinicians and algorithm (%). The NPID approach was compared against the clinical team (clinical meiboscore), 14 the lead clinical investigator, 14 and a supervised learning approach. 14 The NPID approach achieves 80.9% overall grading accuracy with ImageNet pretrained model.…”

Section: Methodsmentioning

confidence: 99%

“… 12 , 13 AI has also benefited the MG atrophy evaluation from meibography images and have shown significantly improved performance. 14 However, it is costly, or sometimes even impossible for training CNNs on large labeled data sets because most of them have imbalanced label classes (i.e. one class accounts for almost 90% of the data, whereas other classes have far fewer samples).…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Meibography Phenotyping and Classification From Unsupervised Discriminative Feature Learning

Yeh

Lin

2021

Trans. Vis. Sci. Tech.

Self Cite

View full text Add to dashboard Cite

Purpose The purpose of this study was to develop an unsupervised feature learning approach that automatically measures Meibomian gland (MG) atrophy severity from meibography images and discovers subtle relationships between meibography images according to visual similarity. Methods One of the latest unsupervised learning approaches is to apply feature learning based on nonparametric instance discrimination (NPID), a convolutional neural network (CNN) backbone model trained to encode meibography images into 128-dimensional feature vectors. The network aims to learn a similarity metric across all instances (e.g. meibography images) and groups visually similar instances together. A total of 706 meibography images with corresponding meiboscores were collected and annotated for the use of network learning and performance evaluation. Results Four hundred ninety-seven meibography images were used for network learning and tuning, whereas the remaining 209 images were used for network model evaluations. The proposed nonparametric instance discrimination approach achieved 80.9% meiboscore grading accuracy on average, outperforming the clinical team by 25.9%. Additionally, a 3D feature visualization and agglomerative hierarchical clustering algorithms were used to discover the relationship between meibography images. Conclusions The proposed NPID approach automatically analyses MG atrophy severity from meibography images without prior image annotations, and categorizes the gland characteristics through hierarchical clustering. This method provides quantitative information on the MG atrophy severity based on the analysis of phenotypes. Translational Relevance The study presents a Meibomian gland atrophy evaluation method for meibography images based on unsupervised learning. This method may be used to aid diagnosis and management of Meibomian gland dysfunction without prior image annotations, which require time and resources.

show abstract

Section: Methodsmentioning

confidence: 91%

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Meibography Phenotyping and Classification From Unsupervised Discriminative Feature Learning

Yeh

Lin

2021

Trans. Vis. Sci. Tech.

Self Cite

View full text Add to dashboard Cite

show abstract

“…They evaluate their performance against various clinically relevant metrics and concluded that the automatic segmentation of MGs is very close to the results derived from the ground truth. 16 Developed an algorithm based on the pyramid scene parsing network 17 to segment the MG atrophy regions and eyelid from meibographic images. A dataset composed of 706 meibography images annotated with atrophy regions and eyelid, were used for this study.…”

Section: Introductionmentioning

confidence: 99%

Image based analysis of meibomian gland dysfunction using conditional generative adversarial neural network

et al. 2021

View full text Add to dashboard Cite

ObjectiveMeibomian gland dysfunction (MGD) is a primary cause of dry eye disease. Analysis of MGD, its severity, shapes and variation in the acini of the meibomian glands (MGs) is receiving much attention in ophthalmology clinics. Existing methods for diagnosing, detection and analysing meibomianitis are not capable to quantify the irregularities to IR (infrared) images of MG area such as light reflection, interglands and intraglands boundaries, the improper focus of the light and positioning, and eyelid eversion.Methods and analysisWe proposed a model that is based on adversarial learning that is, conditional generative adversarial network that can overcome these blatant challenges. The generator of the model learns the mapping from the IR images of the MG to a confidence map specifying the probabilities of being a pixel of MG. The discriminative part of the model is responsible to penalise the mismatch between the IR images of the MG and confidence map. Furthermore, the adversarial learning assists the generator to produce a qualitative confidence map which is transformed into binary images with the help of fixed thresholding to fulfil the segmentation of MG. We identified MGs and interglands boundaries from IR images.ResultsThis method is evaluated by meiboscoring, grading, Pearson correlation and Bland-Altman analysis. We also judged the quality of our method through average Pompeiu-Hausdorff distance, and Aggregated Jaccard Index.ConclusionsThis technique provides a significant improvement in the quantification of the irregularities to IR. This technique has outperformed the state-of-art results for the detection and analysis of the dropout area of MGD.

show abstract

Strip and boundary detection multi‐task learning network for segmentation of meibomian glands

Zhu,

Liu,

Zhuang

et al. 2024

Medical Physics

View full text Add to dashboard Cite

BackgroundAutomatic segmentation of meibomian glands in near‐infrared meibography images is basis of morphological parameter analysis, which plays a crucial role in facilitating the diagnosis of meibomian gland dysfunction (MGD). The special strip shape and the adhesion between glands make the automatic segmentation of meibomian glands very challenging.PurposeA strip and boundary detection multi‐task learning network (SBD‐MTLNet) based on encoder‐decoder structure is proposed to realize the automatic segmentation of meibomian glands.MethodsA strip mixed attention module (SMAM) is proposed to enhance the network's ability to recognize the strip shape of glands. To alleviate the problem of adhesion between glands, a boundary detection auxiliary network (BDA‐Net) is proposed, which introduces boundary features to assist gland segmentation. A self‐adaptive interactive information fusion module (SIIFM) based on reverse attention mechanism is proposed to realize information complementation between meibomian gland segmentation and boundary detection tasks. The proposed SBD‐MTLNet has been evaluated on an in‐house dataset (453 images) and a public dataset MGD‐1K (1000 images). Due to the limited number of images, a five‐fold cross validation strategy is adopted.ResultsAverage dice coefficient of the proposed SBD‐MTLNet reaches 81.08% and 84.32% on the in‐house dataset and the public one, respectively. Comprehensive experimental results demonstrate the effectiveness the proposed SBD‐MTLNet, outperforming other state‐of‐the‐art methods.ConclusionsThe proposed SBD‐MTLNet can focus more on the shape characteristics of the meibomian glands and the boundary contour information between the adjacent glands via multi‐task learning strategy. The segmentation results of the proposed method can be used for the quantitative morphological characteristics analysis of meibomian glands, which has potential for the auxiliary diagnosis of MGD in clinic.

show abstract

A Deep Learning Approach for Meibomian Gland Atrophy Evaluation in Meibography Images

Cited by 53 publications

References 31 publications

Meibography Phenotyping and Classification From Unsupervised Discriminative Feature Learning

Meibography Phenotyping and Classification From Unsupervised Discriminative Feature Learning

Image based analysis of meibomian gland dysfunction using conditional generative adversarial neural network

Strip and boundary detection multi‐task learning network for segmentation of meibomian glands

Contact Info

Product

Resources

About