Musashino University Undergraduate Student Publishes AI Glaucoma Diagnosis Paper in International Journal 'IEEE Access'

Ichigo Endo, a third-year student at Musashino University’s Faculty of Data Science (Koto-ku, Tokyo; President: Seiko Konishi), has published a research paper in 'IEEE Access,' an international academic journal of the Institute of Electrical and Electronics Engineers (IEEE), one of the world's largest technical professional organizations for engineering and information science. The paper is also indexed in 'IEEE Xplore,' making it accessible to researchers and engineers worldwide.

[Research Content] The published paper summarizes research on improving the accuracy of AI-driven glaucoma diagnosis. Specifically, it investigated whether adding information about the position and shape of the optic nerve truly enhances diagnostic precision. While previous studies often demonstrated the effectiveness of AI glaucoma diagnosis using about 1,000 images from a single hospital, this study analyzed over 12,000 fundus images from 19 different open-source datasets collected globally to verify effectiveness across diverse medical environments.

The results showed that while the optic nerve area could be analyzed with high precision, the improvement in diagnostic accuracy from adding this information was limited. Although improvements were seen in certain datasets, the AI failed to consistently deliver sufficient diagnostic performance in different imaging environments.

This research challenges the widely held belief in the medical AI field that 'adding anatomical information improves diagnostic performance' and highlights the importance of large-scale validation. The findings underscore that for medical AI to be used in actual clinical settings, it must function stably across diverse imaging environments, not just achieve high accuracy in a single controlled setting.

[Paper Overview] Title: "The Limited Utility of Segmentation Integration in Glaucoma Classification: A Large-Scale Diagnostic Evaluation" Authors: Ichigo Endo, Yoshihisa Fukuhara Journal: IEEE Access (Volume 14) Pages: 67738 – 67755 Electronic Publication: IEEE Xplore Issue Date: May 4, 2026

[Comments] ■ Ichigo Endo (Junior Student): 'This study questioned the generalization performance of current methods using diverse clinical data. By analyzing over 12,000 images, we identified the practical limitations of methods that only work in specific environments. I hope this empirical approach serves as a foundation for developing medical AI that truly works in the field.'

Lecturer Yoshihisa Fukuhara: 'This research demonstrates that undergraduate students can apply data science knowledge to fields like medicine. Mr. Endo utilized the university's computing resources with high technical skill to complete this rigorous study. I look forward to his future achievements.'