Using Deep Learning and Transfer Learning to Accurately Diagnose Early-Onset Glaucoma From Macular Optical Coherence Tomography Images

Asaoka, Ryo; Murata, Hiroshi; Hirasawa, Kazunori; Fujino, Yuri; Matsuura, Masato; Miki, Atsuya; Kanamoto, Takashi; Ikeda, Yoko; Mori, Kazuhiko; Iwase, Aiko; Shoji, Nobuyuki; Inoue, Kenji; Yamagami, Junkichi; Araie, Makoto

doi:10.1016/j.ajo.2018.10.007

Cited by 189 publications

(155 citation statements)

References 37 publications

Supporting

Mentioning

138

Contrasting

Unclassified

Order By: Relevance

“…In ophthalmology, AI is now about to enter into the clinical phase for the diagnosis and prognosis of diseases. [1][2][3][4] In the field of AI in ophthalmology, there are some new findings assumed to be not possible before AI. One of the most unexpected findings was the ability of AI to identify the sex of an individual based on the characteristics of the ocular color fundus photographs (CFPs) of the individual.…”

Section: Introductionmentioning

confidence: 99%

Factors in Color Fundus Photographs That Can Be Used by Humans to Determine Sex of Individuals

Yamashita

Asaoka

Terasaki

et al. 2020

Trans. Vis. Sci. Tech.

Self Cite

View full text Add to dashboard Cite

Purpose: Artificial intelligence (AI) can identify the sex of an individual from color fundus photographs (CFPs). However, the mechanism(s) involved in this identification has not been determined. This study was conducted to determine the information in CFPs that can be used to determine the sex of an individual. Methods:Prospective observational cross-sectional study of 112 eyes of 112 healthy volunteers. The following characteristics of CFPs were analyzed: the color of peripapillary area expressed by the mean values of red, green, and blue intensities, and the tessellation expressed by the tessellation fundus index (TFI). The optic disc ovality ratio, papillomacular angle, retinal artery trajectory, and retinal vessel angles were also quantified. Their differences between the sexes were assessed by Mann-Whitney U tests. Regularized binomial logistic regression was used to select the decisive factors. In addition, its discriminative performance was evaluated through the leave-one-out cross validation. Results:The mean age of 76 men and 36 women was 25.8 years. The regularized binomial logistic regression delivered the optimal model for sex selected variables of peripapillary temporal green and blue intensities, temporal TFI, supratemporal TFI, optic disc ovality ratio, artery trajectory, and supratemporal retinal artery angle. With this approach, the discrimination accuracy rate was 77.9%. Conclusions:Human-assessed characteristics of CFPs are useful in investigating the new theme proposed by AI, the sex of an individual.Translational Relevance: This is the first report to approach the thinking process of AI by humans and can be a new approach to medical AI research.

show abstract

Section: Introductionmentioning

confidence: 99%

Factors in Color Fundus Photographs That Can Be Used by Humans to Determine Sex of Individuals

Yamashita

Asaoka

Terasaki

et al. 2020

Trans. Vis. Sci. Tech.

Self Cite

View full text Add to dashboard Cite

show abstract

“…A recent study using deep learning on macula scans using GCIPL thickness maps to detect glaucoma had an AUC of 0.9307. 28 The differences from our study were that they defined glaucoma based on the presence of glaucomatous disc changes with RNFL defects on clinical examination along with corresponding VF defects and did not use multimodal longitudinal imaging for ground truth definitions. They used segmented data to estimate GCIPL thickness, did not use raw 3D cube scans, and did not include glaucoma suspects.…”

Section: Discussionmentioning

confidence: 87%

A 3D Deep Learning System for Detecting Referable Glaucoma Using Full OCT Macular Cube Scans

Russakoff¹,

Mannil

Oakley³

et al. 2020

Trans. Vis. Sci. Tech.

View full text Add to dashboard Cite

Citation: Russakoff DB, Mannil SS, Oakley JD, Ran AR, Cheung CY, Dasari S, Riyazzuddin M, Nagaraj S, Rao HL, Chang D, Chang RT. A 3D deep learning system for detecting referable glaucoma using full OCT macular cube scans. Trans Vis Sci Tech. 2020;9(2):12, https://doi.org/10. 1167/tvst.9.2.12 Purpose: The purpose of this study was to develop a 3D deep learning system from spectral domain optical coherence tomography (SD-OCT) macular cubes to differentiate between referable and nonreferable cases for glaucoma applied to real-world datasets to understand how this would affect the performance.Methods: There were 2805 Cirrus optical coherence tomography (OCT) macula volumes (Macula protocol 512 × 128) of 1095 eyes from 586 patients at a single site that were used to train a fully 3D convolutional neural network (CNN). Referable glaucoma included true glaucoma, pre-perimetric glaucoma, and high-risk suspects, based on qualitative fundus photographs, visual fields, OCT reports, and clinical examinations, including intraocular pressure (IOP) and treatment history as the binary (two class) ground truth. The curated real-world dataset did not include eyes with retinal disease or nonglaucomatous optic neuropathies. The cubes were first homogenized using layer segmentation with the Orion Software (Voxeleron) to achieve standardization. The algorithm was tested on two separate external validation sets from different glaucoma studies, comprised of Cirrus macular cube scans of 505 and 336 eyes, respectively. Results:The area under the receiver operating characteristic (AUROC) curve for the development dataset for distinguishing referable glaucoma was 0.88 for our CNN using homogenization, 0.82 without homogenization, and 0.81 for a CNN architecture from the existing literature. For the external validation datasets, which had different glaucoma definitions, the AUCs were 0.78 and 0.95, respectively. The performance of the model across myopia severity distribution has been assessed in the dataset from the United States and was found to have an AUC of 0.85, 0.92, and 0.95 in the severe, moderate, and mild myopia, respectively. Conclusions:A 3D deep learning algorithm trained on macular OCT volumes without retinal disease to detect referable glaucoma performs better with retinal segmentation preprocessing and performs reasonably well across all levels of myopia.Translational Relevance: Interpretation of OCT macula volumes based on normative data color distributions is highly influenced by population demographics and characteristics, such as refractive error, as well as the size of the normative database. Referable glaucoma, in this study, was chosen to include cases that should be seen by a specialist. This study is unique because it uses multimodal patient data for the glaucoma definition, and includes all severities of myopia as well as validates the algorithm with international data to understand generalizability potential.

show abstract

“…In the latest years, advanced methods based on the deep learning paradigm are gaining all the attention in the machine learning world. Deep learning strategies over OCT images are being applied with success on ophthalmology for automated lesion classification [39] [40]. No references on dermatology have been found at the time of writing this work.…”

Section: Resultsmentioning

confidence: 99%

Automatic pigmented lesion segmentation through a dermoscopy-guided OCT approach for early diagnosis

Sarachaga

Lage

Morales

et al. 2019

Photonics in Dermatology and Plastic Surgery 2019

View full text Add to dashboard Cite

show abstract

Using Deep Learning and Transfer Learning to Accurately Diagnose Early-Onset Glaucoma From Macular Optical Coherence Tomography Images

Cited by 189 publications

References 37 publications

Factors in Color Fundus Photographs That Can Be Used by Humans to Determine Sex of Individuals

Factors in Color Fundus Photographs That Can Be Used by Humans to Determine Sex of Individuals

A 3D Deep Learning System for Detecting Referable Glaucoma Using Full OCT Macular Cube Scans

Automatic pigmented lesion segmentation through a dermoscopy-guided OCT approach for early diagnosis

Contact Info

Product

Resources

About