Tsukazaki Optos Public Project

Hiroki Masumoto, Chief Artificial Intelligence Engineer and ophthalmologist of Tsukazaki Hospital
Last updated: August 26, 2019
The dataset is available from the following URL:
Entering your name, email address and affiliation and checking the Purpose checkbox are necessary on the following page to download the dataset.
If you succeed the submit your information, the email which contains the dataset URL and the password for download would get to your mail address entered.
If we found the direct URL to this dataset are disclosed, we would change the location of the dataset.

What is Tsukazaki Optos Public project?

We published 13,047 images of 8588 eyes (of a total of 5389 people), which were obtained between October 11, 2011 and September 6, 2018. All the images were obtained at Tsukazaki Hospital, Himeji, Japan, with the Optos® 200Tx (Optos®, Dunfermline, U.K.), which was the main device of our development and research.

Tsukazaki Hospital is the largest ophthalmology facility in Japan. In the past 15 years, we have built a database including 1 million or more images of 100,000 or more patients. In addition, various papers have been written about the data obtained in my and my colleagues’ daily practice. In fact, we are top planners in the development of artificial intelligence in the Japanese ophthalmology industry.

It is hoped that publishing this dataset will not only advance artificial intelligence research in ophthalmology but also help to educate personnel in human resources for ophthalmology and artificial intelligence.

Why is the Optos images dataset?

We place great importance on the national health insurance system. This system, which ensures the right to medical treatment for rich and poor people equally, is necessary for medical treatment to function as an infrastructure in Japan. On the other hand, it is clear that because of the declining birthrate and the aging of the population, the system cannot be sustained.

Therefore, there are only three methods left to address the declining birthrate and aging population: (1) increasing efficiency of medical practice by information technology, (2) inbound marketing for medical services, and (3) medical innovation in Japan.

My colleagues and I have published various papers about AI and ultra-wide-field fundus images taken with the Optos® 200Tx. These publications have enhanced the presence of the Japanese company's products throughout the world.

Our paper about the diagnosis of rhegmatogenous retinal detachment (RRD) with the use of neural networks[1] has been taken up by the American Academy of Ophthalmology. RRD is the leading cause of litigation-related ophthalmologic disease in the United States and is a very important disease in developed countries. In a non-mydriatic state, Images of RRD lesions can be obtained only with the Optos®.

A content of this dataset

The img zip file is the zipped folder that contains the images.
The sample folder contains some of the images.
The data.csv file contains the columns of filenames, randomized ID, patients’ sex, left or right (LR) eye, and tags of disease.

The abbreviations in the data.csv are as below.
LR : Left or Right of the eyes
L : Left
R : Right
AMD: Age-related Macular Degeneration
RVO: Retinal Vein Occlusion
Gla: Glaucoma
MH : Macular Hole
DR : Diabetic Retinopathy
RD : Retinal Detachment
RP : Retinitis Pigmentosa
AO : Artery Occlusion
DM : Diabetes Mellitus
The value of 1 of the tag means the presence of the disease and that of 0 means the absence of the disease.

If all of AMD, RVO, Gla, MH, DR, RD, RP and AO tags have values of 0, it means that the eye does not have fundus disease (normal eye). The value of the DM tag is determined only according to HbA1c level in blood tests.
The mean age of patients is 65.1 ± 12.9 years.
The number of normal images is 4894.
The numbers of images of each disease are as follows:

Disease Amount
AMD 413
RVO 778
Gla 2619
MH 222
DR 3323
RD 974
RP 258
AO 21
DM 3895


Please download the data set from the following URL:

The information included in this database can be used, free of charge, only for research and educational purposes. Copy, redistribution, and any unauthorized commercial use are prohibited. Any researcher reporting results that involved the use of this database must acknowledge the Tsukazaki Optos Public program by adding the following information: “Kindly provided by the Tsukazaki Optos Public program partners (see https://tsukazaki-ai.github.io/optos_dataset/).”
The study was approved by the Ethics Committee of Tsukazaki Hospital (Himeji, Japan) (No 191014)

In addition, my colleagues and I would appreciate hearing about any publication in which the Tsukazaki Optos Public database is used. Feedback about the database and this website is also welcome. The person to contact is Hiroki Masumoto.

We also have the undisclosed tags of these images. If you would like to use these tags, please contact us. h.masumoto@tsukazaki-eye.net


I thank the following people for confirmation of the tags: Daisuke Nagasato, Shunsuke Nakakura, Masahiro Kameoka, Hitoshi Tabuchi, Ryota Aoki, Takahiro Sogawa, Shinji Matsuba, Hirotaka Tanabe, Toshihiko Nagasawa, Yuki Yoshizumi, Tomoaki Sonobe, Tomofusa Yamauchi
I also thank all the staff of Tsukazaki Hospital.

Published paper about Optos with neural network

  1. Accuracy of deep learning, a machine-learning technology, using ultra–wide-field fundus ophthalmoscopy for detecting rhegmatogenous retinal detachment
    Ohsugi H, Tabuchi H, Enno H, Ishitobi N
    Scientific Reports 2017 Aug 25;7(1):9425. doi: 10.1038/s41598-017-09891-x.

  2. Accuracy of ultra–wide-field fundus ophthalmoscopy-assisted deep learning, a machine-learning technology, for detecting age related macular degeneration
    Matsuba S, Tabuchi H, Ohsugi H, Enno H, Ishitobi N, Masumoto H, Kiuchi Y
    International Ophthalmology.
    2018 May 9. doi: 10.1007/s10792-018-0940-0.

  3. Deep-learning classifier with an ultrawide-field scanning laser ophthalmoscope detects glaucoma visual field severity
    Masumoto H, Tabuchi H, Nakakura S, Ishitobi N, Miki M, Enno H
    Journal of Glaucoma.
    2018 Jul;27(7):647-652. doi: 10.1097/IJG.0000000000000988.

  4. Accuracy of Deep Learning, a Machine-Learning Technology, Using Ultra–Widefield Fundus Ophthalmoscopy for Detecting Idiopathic Macular Holes
    Nagasawa T, Tabuchi H, Masumoto H, Enno H, Niki M, Ohsugi H, Mitamura Y
    2018 Oct 22;6:e5696. doi: 10.7717/peerj.5696. eCollection 2018.

  5. Deep neural network-based method for detecting central retinal vein occlusion using ultrawide-field fundus ophthalmoscopy
    Nagasato D, Tabuchi H, Ohsugi H, Masumoto H, Enno H, Ishitobi N, Sonobe T, Kameoka M, Niki M, Hayashi K, Mitamura Y
    Journal of Ophthalmology
    2018 Nov 1;2018:1875431. doi: 10.1155/2018/1875431. eCollection 2018

  6. Discrimination ability of glaucoma via DCNNs models from ultra-wide angle fundus images comparing either full or confined to the optic disc
    Tabuchi H, Masumoto H, Nakakura S, Noguchi A, Tanabe H
    Computer Vision – ACCV 2018 Workshops P229-234 doi:10.1007/978-3-030-21074-8_18

  7. Retinal Detachment Screening with Ensembles of Neural Network Models
    Masumoto H, Tabuchi H, Adachi S, Nakakura S, Ohsugi H, Nagasato D
    Computer Vision – ACCV 2018 Workshops P251-260 doi:10.1007/978-3-030-21074-8_20

  8. Deep-learning classifier with ultrawide-field fundus ophthalmoscopy for detecting branch retinal vein occlusion
    Nagasato D, Tabuchi H, Ohsugi H, Masumoto H, Enno H, Ishitobi N, Sonobe T, Kameoka M, Niki M, Mitamura Y
    International Journal of Ophthalmology.
    2019 Jan 18;12(1):94-99. doi: 10.18240/ijo.2019.01.15. eCollection 2019

  9. Accuracy of ultrawide-field fundus ophthalmoscopy-assisted deep learning for detecting treatment-naïve proliferative diabetic retinopathy
    Nagasawa T, Tabuchi H, Masumoto H, Enno H, Niki M, Ohara Z, Yoshizumi Y, Ohsugi H, Mitamura Y
    International Ophthalmology
    2019 Feb. doi: 10.1007/s10792-019-01074-z

  10. Accuracy of a deep convolutional neural network in detection of retinitis pigmentosa on ultrawide-field images
    Masumoto H, Tabuchi H, Nakakura S, Ohsugi H, Enno H, Ishitobi N, Ohsugi E, Mitamura Y
    PeerJ. 2019 May 7;7:e6900. doi: 10.7717/peerj.6900. eCollection 2019