Summary

This dataset provides paired ultra-widefield (UWF) fundus photographs and patient questionnaire responses collected at Tsukazaki Hospital, Japan, for developing and evaluating multimodal artificial intelligence models for retinal detachment (RD) diagnosis. The release includes per-subject questionnaire data (20 binary symptom items, sex, age, and label) and per-image JPEG files named to link images with the corresponding questionnaire ID.

Commercial use—including training or fine-tuning AI models intended for commercial products or services—is strictly prohibited.

Dataset overview

• Source: Tsukazaki Hospital, Japan
• Period: 2020–2024 for RD cases; 2020–2023 for controls
• Subjects: 98 RD patients and 164 non-RD controls (total 262 subjects)
• Images: JPEG format; multiple images per subject (see File naming)
• Questionnaires: CSV format; one row per subject (ID)

Packages

• RD.zip: 442 JPEG images from 98 subjects diagnosed with retinal detachment (preprocessed as described in the manuscript currently under submission)
• non-RD.zip: 421 JPEG images from 164 control subjects (preprocessed in the same manner as RD.zip)
• questionnaire.csv: provided as a separate file in this record (one row per subject)

File naming (images)

• Pattern: {ID}_{number}_{side}.jpg
  • ID: links the image to the questionnaire row
  • number: per-subject sequence, zero-padded (01, 02, …). If four images were acquired for a subject, numbers range 01–04.
  • side: "l" or "r" (left or right)

questionnaire.csv format

• One row per subject (ID). Suggested columns/order:
  • ID, Q1–Q20, Sex, Age, label
• Coding:
  • Symptom items (Q1–Q20): 1 = yes (symptom present), 0 = no (symptom absent)
  • Sex: 0 = male, 1 = female
  • Age: integer (years)
  • label: 1 = RD, 0 = non-RD

Symptom items (Q1–Q20)