Quantitative and qualitative evaluation of deep learning automatic segmentations of corneal endothelial cell images of reduced image quality obtained following cornea transplant

Joseph, Naomi; Kolluru, Chaitanya; Benetz, Beth Ann; Menegay, Harry; Lass, Jonathan H.; Wilson, David L.

doi:10.1117/1.jmi.7.1.014503

Cited by 21 publications

(35 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The precision of the refurbished automated analysis will be a subject for future studies. Recently, several studies have reported precise and accurate results using deep learning automatic segmentations of corneal endothelial cell images [10][11][12][13]. We expect the development of new software with precise automated analysis in the near future.…”

Section: Discussionmentioning

confidence: 99%

Overestimation of corneal endothelial cell density by automated method in glaucomatous eyes with impaired corneal endothelial cells

Minami

Chihara²

2021

Int Ophthalmol

View full text Add to dashboard Cite

Purpose To determine between-method differences in corneal endothelial cell parameters using center and automated methods of non-contact specular microscopy (CellCheck software of Konan, Inc.) in glaucomatous eyes. Methods We analyzed the central corneal endothelial cell density (ECD) of 245 glaucomatous eyes using center (ECD-Ce) and automated methods (ECD-Au). Based on the ECD-Ce, we allocated subjects to Groups 1 to 10 (at 250 cells/mm2 intervals) and evaluated the ECD, coefficient of variation in cell area (CV), and percentage of hexagonal cells (HEX). Results There was a close correlation (r = 0.91) between the ECD values measured using both methods. However, ECD-Au were significantly higher than those measured by the center method when ECD-Ce was less than 2500 (in Groups 1 to 8; P < 0.001 to P = 0.006). The regression equation of (ECD-Au—ECD-Ce) = 1028–0.397*ECD-Ce shows greater deviation in eyes with lower ECD, and this difference became 0 when ECD -Ce was 2593 cells/mm2. None of the 44 subjects with an ECD-Ce of < 1000 cells/mm2 recorded an ECD-Au < 1000 cells/mm2. Compared with the center method, the automated method had higher and lower median CV and HEX values, respectively (P < 0.001). The between-method differences in both CV and HEX were negatively correlated with ECD-Ce (r = −0.49, P < 0.001 and r = −0.25, P < 0.001, respectively). Conclusion The automated method of the CellCheck software overestimates ECD in eyes with lower ECD values and may overlook risk of corneal decompensation.

show abstract

Section: Discussionmentioning

confidence: 99%

Overestimation of corneal endothelial cell density by automated method in glaucomatous eyes with impaired corneal endothelial cells

Minami

Chihara²

2021

Int Ophthalmol

View full text Add to dashboard Cite

show abstract

“…Sensitivity to image quality variations will also need to be assessed on a larger dataset; further, methods to deal with images of reduced image quality can be implemented. 24 …”

Section: Discussionmentioning

confidence: 99%

“…The system shows human-comparable performance, i.e., 92% of cells from the test set were scored as acceptably segmented by a human evaluator, with a full annotation being able to be completed in under 5 min per image. 24 Auksorius et al 25 also recently proposed a similar pipeline for volumetric images obtained with Fourier-domain full-field optical coherence tomography (OCT). The proposed pipeline used a modified U-Net neural network for the segmentation, trained with two OCT images and four SM images.…”

Section: Introductionmentioning

confidence: 99%

Unbiased corneal tissue analysis using Gabor-domain optical coherence microscopy and machine learning for automatic segmentation of corneal endothelial cells

2020

View full text Add to dashboard Cite

. Significance: An accurate, automated, and unbiased cell counting procedure is needed for tissue selection for corneal transplantation. Aim: To improve accuracy and reduce bias in endothelial cell density (ECD) quantification by combining Gabor-domain optical coherence microscopy (GDOCM) for three-dimensional, wide field-of-view ( ) corneal imaging and machine learning for automatic delineation of endothelial cell boundaries. Approach: Human corneas stored in viewing chambers were imaged over a wide field-of-view with GDOCM without contacting the specimens. Numerical methods were applied to compensate for the natural curvature of the cornea and produce an image of the flattened endothelium. A convolutional neural network (CNN) was trained to automatically delineate the cell boundaries using 180 manually annotated images from six corneas. Ten additional corneas were imaged with GDOCM and compared with specular microscopy (SM) to determine performance of the combined GDOCM and CNN to achieve automated endothelial counts relative to current procedural standards. Results: Cells could be imaged over a larger area with GDOCM than SM, and more cells could be delineated via automatic cell segmentation than via manual methods. ECD obtained from automatic cell segmentation of GDOCM images yielded a correlation of 0.94 ( ) with the manual segmentation on the same images, and correlation of 0.91 ( ) with the corresponding manually counted SM results. Conclusions: Automated endothelial cell counting on GDOCM images with large field of view eliminates selection bias and reduces sampling error, which both affect the gold standard of manual counting on SM images.

show abstract

Section: Artificial Intelligencementioning

confidence: 99%

“…AI also allows rapid assessment of the corneal endothelium with good reliability [ 80 – 83 ]. A deep learning method called U -net was capable of substantially faster and more accurate segmentation compared to manual segmentation [ 80 , 81 ]. Heinzelmann et al [ 84 ] revealed that the endothelial cell counts measured using U-Net showed strong correlation with those obtained with the gold standard, suggesting the potential applicability of the AI model in the long-term assessment of corneal grafts.…”

Section: Artificial Intelligencementioning

confidence: 99%

Advances in Imaging Technology of Anterior Segment of the Eye

Han

Liu

Mohamed-Noriega

2021

Journal of Ophthalmology

View full text Add to dashboard Cite

Advances in imaging technology and computer science have allowed the development of newer assessment of the anterior segment, including Corvis ST, Brillouin microscopy, ultrahigh-resolution optical coherence tomography, and artificial intelligence. They enable accurate and precise assessment of structural and biomechanical alterations associated with anterior segment disorders. This review will focus on these 4 new techniques, and a brief overview of these modalities will be introduced. The authors will also discuss the recent advances in research regarding these techniques and potential application of these techniques in clinical practice. Many studies on these modalities have reported promising results, indicating the potential for more detailed comprehensive understanding of the anterior segment tissues.

show abstract

Quantitative and qualitative evaluation of deep learning automatic segmentations of corneal endothelial cell images of reduced image quality obtained following cornea transplant

Cited by 21 publications

References 19 publications

Overestimation of corneal endothelial cell density by automated method in glaucomatous eyes with impaired corneal endothelial cells

Overestimation of corneal endothelial cell density by automated method in glaucomatous eyes with impaired corneal endothelial cells

Unbiased corneal tissue analysis using Gabor-domain optical coherence microscopy and machine learning for automatic segmentation of corneal endothelial cells

Advances in Imaging Technology of Anterior Segment of the Eye

Contact Info

Product

Resources

About