[2024-04] Deep Learning by Vision Transformer to Classify Bacterial and Fungal Keratitis Using Different Types of Anterior Segment Images

홈 Program Working Paper

자격시험 응시 시 기본적인 논문 Draft 1편 이상 제출

[2024-04] Deep Learning by Vision Transformer to Classify Bacterial and Fungal Keratitis Using Different Types of Anterior Segment Images

운영자
날짜 2024.08.09
조회수 232

ChoongHan Kim

Moon Soul Graduate School of Future Strategy, KAIST

Yeo Kyoung Won

Department of Ophthalmology, Samsung Medical Center, Sungkyunkwan University School of Medicine

Jiho Cha
Moonsoul Graduate School of Future Strategy, KAIST

Jooyoung Jeon

Moon Soul Graduate School of Future Strategy, KAIST
Dong Hui Lim
Department of Ophthalmology, Samsung Medical Center, Sungkyunkwan University School of Medicine

Abstract
Background: Infectious keratitis, especially bacterial or fungal keratitis, is a common ocular emergency that can lead to permanent vision loss if not treated promptly.Bacterial and fungal keratitis are often mistaken for each other, especially in the early stages of the disease. Since fungal keratitis generally has a poor prognosis, it is crucial to develop a quick and accurate technique to discriminate between bacterial and fungal keratitis for appropriate treatment and prevention of further complications.
Methods: A Vision Transformer (ViT) was used to classify bacterial and fungal keratitis. We integrated one or more ViTs by adding a vector or by using self-attention to combine different types of anterior segment images (broad-beam, slit-beam, and blue-light). We compared the area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC) of the models. Cross-validation was performed thrice, and there was no overlap between the validation sets. The training/validation set was divided in an 8:2 ratio based on the number of individuals.
Findings: A total of 283 broad-beam, 610 slit-beam, and 342 blue-light images were obtained from 79 patients. 62 (78%) patients were assigned for training and 17 (22%) for validation. The AUROC of ViT with broad-beam images was 0·72. The top AUROC score (0·93) was attained by combining the outputs from two ViT models utilizing self-attention, incorporating both broad-beam and slit-beam images. Similarly, the highest AUPRC score (0·93) was reached by fusing the outputs from three ViTs with self-attention, involving broad-beam, slit-beam, and blue-light images.
Interpretation: Despite the limited dataset, we validated ViT with self-attention to learn different types of images to improve recognition accuracy in diagnosing bacterial and fungal keratitis. ViT with self-attention has a meaningful effect on enhancing the diagnostic performance of bacterial and fungal keratitis by combining two or more types of anterior segment images.
Funding: This work was supported by a National Research Foundation of Korea grant funded by the Korean government's Ministry of Education (NRF-2021R1C1C1007795; Seoul, Korea) and received by D. H. L.

Keywords: Deep learning, Bacterial keratitis, Fungal keratitis, Vision transformer, Anterior segment images, Self-attention

[2024-04] Deep Learning by Vision Transformer to Classify Bacterial and Fungal Keratitis Using Different Types of Anterior Segment Images.pdf

SNS Share

이전글[2024-05] Projection of Future Agricultural Land Supply: Integrating Workforce Potential and Climate Limitations

다음글[2023-09] Developments and challenges of foresight evaluation: Review of the past 30 years of research