Optimizing Field-of-View for Type-1 Retinopathy of Prematurity via Multiple Instance Learning.

IF 2.7 3区医学 Q2 OPHTHALMOLOGY

Translational Vision Science & Technology Pub Date : 2026-08-03 DOI:10.1167/tvst.15.8.13

Wen-Hsi Lan, Chia-Ling Tsai, Wei-Chi Wu

{"title":"Optimizing Field-of-View for Type-1 Retinopathy of Prematurity via Multiple Instance Learning.","authors":"Wen-Hsi Lan, Chia-Ling Tsai, Wei-Chi Wu","doi":"10.1167/tvst.15.8.13","DOIUrl":null,"url":null,"abstract":"Purpose: The purpose of this study was to cut down on screening time and costs, this study optimizes field-of-view (FOV) combinations for classifying type-1 retinopathy of prematurity (ROP) through multiple-instance learning (MIL) that models clinical decision making.Methods: The dataset included 1420 photographs of 284 eyes (204 eyes from Chang Gung Memorial Hospital [CGMH] and 80 eyes from Osaka University), each eye containing five FOVs (temporal, nasal, central, superior, and inferior). We evaluated various FOV combinations and compared two MIL fusion strategies, feature-level and outcome-level, to identify the most effective approach. Feature-level learns a joint representation across views, whereas outcome-level aggregates the predicted probabilities from each view at the decision stage. Model performance was evaluated using five-fold cross-validation, with metrics including accuracy, precision, recall, F1-score, and area under the curve (AUC).Results: The feature-level MIL significantly outperformed the outcome-level MIL in all combinations and metrics except the superior view (P < 0.05). Among FOV combinations, the temporal, nasal, and central set achieved the highest performance (accuracy = 0.892 ± 0.043, F1-score = 0.831 ± 0.097). Horizontal (temporal and nasal) and central views individually showed stronger diagnostic power, comparable to multi-view settings, whereas vertical (superior and inferior) views performed significantly worse.Conclusions: The feature-level MIL model effectively simulates clinical decision making for type-1 ROP classification. Temporal, nasal, and central views, as well as their combination, achieved superior performance over vertical views or their combination. Notably, feature-level MIL outperformed outcome-level MIL, underscoring the critical role of information fusion strategy.Translational relevance: Clinically aligned MIL with optimized FOV selection enables more efficient artificial intelligence (AI)-assisted telemedicine screening for retinopathy of prematurity.","PeriodicalId":23322,"journal":{"name":"Translational Vision Science & Technology","volume":"15 8","pages":"13"},"PeriodicalIF":2.7000,"publicationDate":"2026-08-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13496267/pdf/","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Translational Vision Science & Technology","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1167/tvst.15.8.13","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"OPHTHALMOLOGY","Score":null,"Total":0}

引用次数: 0

Abstract

Purpose: The purpose of this study was to cut down on screening time and costs, this study optimizes field-of-view (FOV) combinations for classifying type-1 retinopathy of prematurity (ROP) through multiple-instance learning (MIL) that models clinical decision making.

Methods: The dataset included 1420 photographs of 284 eyes (204 eyes from Chang Gung Memorial Hospital [CGMH] and 80 eyes from Osaka University), each eye containing five FOVs (temporal, nasal, central, superior, and inferior). We evaluated various FOV combinations and compared two MIL fusion strategies, feature-level and outcome-level, to identify the most effective approach. Feature-level learns a joint representation across views, whereas outcome-level aggregates the predicted probabilities from each view at the decision stage. Model performance was evaluated using five-fold cross-validation, with metrics including accuracy, precision, recall, F1-score, and area under the curve (AUC).

Results: The feature-level MIL significantly outperformed the outcome-level MIL in all combinations and metrics except the superior view (P < 0.05). Among FOV combinations, the temporal, nasal, and central set achieved the highest performance (accuracy = 0.892 ± 0.043, F1-score = 0.831 ± 0.097). Horizontal (temporal and nasal) and central views individually showed stronger diagnostic power, comparable to multi-view settings, whereas vertical (superior and inferior) views performed significantly worse.

Conclusions: The feature-level MIL model effectively simulates clinical decision making for type-1 ROP classification. Temporal, nasal, and central views, as well as their combination, achieved superior performance over vertical views or their combination. Notably, feature-level MIL outperformed outcome-level MIL, underscoring the critical role of information fusion strategy.

Translational relevance: Clinically aligned MIL with optimized FOV selection enables more efficient artificial intelligence (AI)-assisted telemedicine screening for retinopathy of prematurity.

查看原文本刊更多论文

基于多实例学习的早产儿1型视网膜病变视野优化研究。

目的：为了减少筛查时间和成本，本研究通过模拟临床决策的多实例学习（MIL）优化视场（FOV）组合对1型早产儿视网膜病变（ROP）进行分类。方法：数据集包括284只眼睛的1420张照片（204只眼睛来自长庚纪念医院[CGMH]， 80只眼睛来自大阪大学），每只眼睛包含5个fov（颞、鼻、中央、上、下）。我们评估了各种FOV组合，并比较了两种MIL融合策略，特征级和结果级，以确定最有效的方法。特征级学习跨视图的联合表示，而结果级在决策阶段汇总来自每个视图的预测概率。采用五重交叉验证对模型性能进行评估，指标包括准确性、精密度、召回率、f1评分和曲线下面积（AUC）。结果：除优势视图外，特征水平MIL在所有组合和指标上均显著优于结果水平MIL （P < 0.05）。在FOV组合中，颞部、鼻部和中央组的准确率最高（准确率= 0.892±0.043,F1-score = 0.831±0.097）。水平视图（颞部和鼻部）和中央视图单独显示出较强的诊断能力，与多视图设置相当，而垂直视图（上视图和下视图）表现明显较差。结论：特征级MIL模型可以有效地模拟1型ROP分类的临床决策。时间视图、鼻视图和中心视图以及它们的组合比垂直视图或它们的组合获得了更好的性能。值得注意的是，特征级MIL优于结果级MIL，强调了信息融合策略的关键作用。翻译相关性：临床对齐MIL与优化的视场选择使人工智能（AI）辅助的早产儿视网膜病变远程医疗筛查更有效。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Translational Vision Science & Technology Engineering-Biomedical Engineering

CiteScore

5.70

自引率

3.30%

发文量

346

审稿时长

25 weeks

期刊介绍： Translational Vision Science & Technology (TVST), an official journal of the Association for Research in Vision and Ophthalmology (ARVO), an international organization whose purpose is to advance research worldwide into understanding the visual system and preventing, treating and curing its disorders, is an online, open access, peer-reviewed journal emphasizing multidisciplinary research that bridges the gap between basic research and clinical care. A highly qualified and diverse group of Associate Editors and Editorial Board Members is led by Editor-in-Chief Marco Zarbin, MD, PhD, FARVO. The journal covers a broad spectrum of work, including but not limited to: Applications of stem cell technology for regenerative medicine, Development of new animal models of human diseases, Tissue bioengineering, Chemical engineering to improve virus-based gene delivery, Nanotechnology for drug delivery, Design and synthesis of artificial extracellular matrices, Development of a true microsurgical operating environment, Refining data analysis algorithms to improve in vivo imaging technology, Results of Phase 1 clinical trials, Reverse translational ("bedside to bench") research. TVST seeks manuscripts from scientists and clinicians with diverse backgrounds ranging from basic chemistry to ophthalmic surgery that will advance or change the way we understand and/or treat vision-threatening diseases. TVST encourages the use of color, multimedia, hyperlinks, program code and other digital enhancements.