Towards label-free 3D segmentation of optical coherence tomography images of the optic nerve head using deep learning

Devalla, Sripad Krishna; Pham, Tan Hung; Panda, Satish Kumar; Zhang, Liang; Subramanian, Giridhar; Swaminathan, Anirudh; Yun, Chin Zhi; Rajan, Mohan; Mohan, Sujatha; Krishnadas, R; Senthil, Vijayalakshmi; Leon, John Mark de; Tun, Tin A.; Cheng, Ching-Yu; Schmetterer, Leopold; Perera, Shamira; Aung, Tin; Thiéry, Alexandre H.; Girard, M

doi:10.1364/boe.395934

Cited by 27 publications

(22 citation statements)

References 67 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In this study, we obtained an averaged Dice coefficient of 0.93 ± 0.03, which is comparable with past ONH segmentation performance obtained with various networks: 0.93 ± 0.02 with ONH-Net, 23 0.91 ± 0.04 with Dilated-Residual U-NET, 24 and 0.90 ± 0.04 with U-Net. 14 The only difference being that ours added an additional class for ODD regions and conglomerates, which did not seem to decrease the overall performance.…”

Section: Discussionsupporting

confidence: 89%

See 1 more Smart Citation

Discriminating Between Papilledema and Optic Disc Drusen Using 3D Structural Analysis of the Optic Nerve Head

et al. 2023

Self Cite

View full text Add to dashboard Cite

Background and Objectives:The distinction of papilledema from other optic nerve head (ONH) lesions mimicking papilledema, such as optic disc drusen (ODD), can be difficult in clinical practice. We aimed: (1) To develop a deep learning algorithm to automatically identify major structures of the optic nerve head (ONH) in three dimensional (3D) optical coherence tomography (OCT) scans; (2) To exploit such information to robustly differentiate among ODD, papilledema, and healthy ONHs.Methods:This was a cross-sectional comparative study of patients from three sites (Singapore, Denmark, Australia) with confirmed ODD, papilledema due to raised intracranial pressure, and healthy controls. Raster scans of the ONH were acquired using OCT imaging, then processed to improve deep-tissue visibility. First, a deep learning algorithm was developed to identify major ONH tissues, and ODD regions. The performance of our algorithm was assessed using the Dice coefficient. Second, a classification algorithm (random forest) was designed to perform 3-class classifications (1: ODD, 2: papilledema, 3: healthy) strictly from their drusen and prelamina swelling scores (calculated from the segmentations). To assess performance, we reported the area under the receiver operating characteristic curve (AUC) for each class.Results:A total of 241 patients (256 imaged ONHs, including 105 ODD, 51 papilledema, and 100 healthy ONHs) were retrospectively included in this study. Using OCT images of the ONH, our segmentation algorithm was able to isolate neural and connective tissues, and ODD regions/conglomerates whenever present. This was confirmed by an averaged Dice coefficient of 0.93±0.03 on the test set, corresponding to good segmentation performance. Classification was achieved with high AUCs, i.e. 0.99±0.001 for the detection of ODD, 0.99±0.005 for the detection of papilledema, and 0.98±0.01 for the detection of healthy ONHs.Discussion:Our AI approach can discriminate ODD from papilledema, strictly using a single OCT scan of the ONH. Our classification performance was very good in the studied population, with the caveat that validation in a much larger population is warranted. Our approach may have the potential to establish OCT imaging as one of the mainstays of diagnostic imaging for ONH disorders in neuro-ophthalmology, in addition to fundus photography.

show abstract

Section: Discussionsupporting

confidence: 89%

“…While it was able to handle multiple scan sizes (all raster, but not diagonal), we have yet to make it device-agnostic because we proposed for both healthy and glaucoma eyes. 23…”

Section: Discussionmentioning

confidence: 99%

Discriminating Between Papilledema and Optic Disc Drusen Using 3D Structural Analysis of the Optic Nerve Head

et al. 2023

Self Cite

View full text Add to dashboard Cite

show abstract

“…Since the quality of OCT images is devicedependent, segmentation networks trained with data from one machine may not perform well on data from others [28]. A few recent studies have already demonstrated device-independent and label-free techniques for OCT image segmentation [29,57]. We aim to test our methodology with data from multiple devices in the near future.…”

Section: Discussionmentioning

confidence: 99%

“…An effective way to address this issue is to, in a first step, segment and thus identify or label all neural and connective tissue layers of the ONH (Fig. 1a, Bottom) using a deep learning network as was performed in our previous work [28,29,30]. Using segmentation as a first step will facilitate our understanding of complex structural differences between glaucoma and non-glaucoma eyes.…”

Section: Segmenting Neural and Connective Tissues In Oct Images Of Th...mentioning

confidence: 99%

Describing the Structural Phenotype of the Glaucomatous Optic Nerve Head Using Artificial Intelligence

Panda¹,

Cheong²,

Tun³

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

The central event in glaucoma is slow and irreversible damage of retinal ganglion cell axons [1, 2]. Retinal ganglion cells carry visual information from the retina to the brain. When damaged, they undergo programmed cell death (apoptosis) resulting in vision loss. The main site of retinal ganglion cell damage occurs within the optic nerve head (ONH) located at the back of the eye [3]. To date, it is widely accepted that clinical evaluation and documentation of the ONH is essential for the diagnosis and the monitoring of glaucoma [4, 5].However, providing an accurate glaucoma diagnosis is a complex endeavor that is time consuming for clinicians. It is subjective and heavily dependent on a clinician's experience/expertise, and in more complex cases, requires a battery of multiple clinical tests. Such tests often need to be repeated at multiple patients' visits to overcome their inherent subjectivity and patient variability to confirm a diagnosis. As stated by the World Glaucoma Association: "As yet there is no widely-accepted method of combining the results of several [glaucoma] tests [to provide an improved diagnosis]". In fact, the use of multiple glaucoma diagnostic tests has been found to increase the likelihood of false-positives yielding to over-treatment [6].Axonal damage in glaucoma is typically evaluated through two tests that assess ONH structure and visual function. In routine clinical practice, functional measures of axonal loss are frequently assessed with visual field testing. During testing, flashes of light reach various regions of the retina to generate a map of its sensitivity to light. Structural measures of axonal loss are assessed with optical coherence tomography (OCT) -a 3D imaging modality that allows fast, high-resolution and non-invasive visualization of the ONH [7]. Using OCT, studies have found that retinal nerve fiber layer (RNFL) thinning was strongly associated with glaucoma severity [8, 9,10,11]. Because RNFL thinning generally precedes detectable functional defects in glaucoma patients [12], RNFL thickness (assessed through OCT) has remained a gold-standard parameter for glaucoma.However, Li and Jampel concluded that: "OCT was not a sufficient stand-alone test for the detection of glaucoma or for triage use in primary care" [7]. We argue this is because the vast amount of information available in a 3D OCT scan of the ONH has not been fully exploited. RNFL thickness and the more recent minimum-rim-width are 1-dimensional (depth) parameters representative of neural tissues only. However, connective tissues also exhibit complex 3-dimensional changes during the development and progression of glaucoma, including, but not limited to: post-laminar deformations [13], changes in LC and choroidal thickness [14,15], peripapillary atrophy [16], scleral canal expansion [17], posterior migration of the LC [18], and peripapillary scleral bowing [19]. To date, no clinical solutions currently consider changes in both neural and connective tissues simultaneously, which may also explain the lack in ...

show abstract

“…Die automatische Segmentierung konnte robust sowohl in gesunden als auch in für Segmentierungsfehler anfälligen glaukomatös veränderten Sehnervenköpfen durchgeführt werden [ 24 ]. In einem weiteren Ansatz konnte die gleiche Arbeitsgruppe einen DL-Algorithmus entwickeln, der die Qualität von OCT-B-Scans unterschiedlicher Geräte so harmonisierte, dass in einem weiteren Schritt eine geräteunspezifische Segmentierung der Strukturen möglich war [ 25 ]. Dies macht eine einfache Implementierung der Segmentierung von OCT-Scans im klinischen Alltag auf unterschiedlichen Geräten möglich und erleichtert die Diagnostik und Verlaufsbeurteilung von Erkrankungen des Sehnervenkopfes wie dem Glaukom.…”

Section: Künstliche Intelligenzunclassified

Diagnostik von Erkrankungen des Sehnervenkopfes in Zeiten von künstlicher Intelligenz und Big Data

2021

View full text Add to dashboard Cite

Towards label-free 3D segmentation of optical coherence tomography images of the optic nerve head using deep learning

Cited by 27 publications

References 67 publications

Discriminating Between Papilledema and Optic Disc Drusen Using 3D Structural Analysis of the Optic Nerve Head

Discriminating Between Papilledema and Optic Disc Drusen Using 3D Structural Analysis of the Optic Nerve Head

Describing the Structural Phenotype of the Glaucomatous Optic Nerve Head Using Artificial Intelligence

Diagnostik von Erkrankungen des Sehnervenkopfes in Zeiten von künstlicher Intelligenz und Big Data

Contact Info

Product

Resources

About