Scholarlink

This paper presents our approach to the MARIO (Monitoring Age-related Macular Degeneration Progression In Optical Coherence Tomography) challenge at MICCAI 2024, focusing on automated analysis of Age-related Macular Degeneration (AMD) evolution using Optical Coherence Tomography (OCT) images. For Task 1, we propose a novel method that fuses consecutive OCT scan pairs, utilizing a MaxVit Tiny architecture to classify AMD progression. This approach mimics clinicians’ side-by-side comparison technique. In the preliminary phase, our model ranked first among 20 teams across all evaluation metrics for Task 1, achieving an F1 score of 0.863, Matthews Rk-correlation coefficient (RkC) of 0.709, and Specificity of 0.913. For Task 2, predicting AMD progression over three months, we combined EfficientNet V2 S for OCT image processing with a Multi-Layer Perceptron for handling patient metadata. While our Task 2 solution ranked competitively (3rd in F1, 1st in QWK among 16 teams), overall scores were low (F1: 0.685, RkC: 0.111, QWK: 0.231, Specificity: 0.688), highlighting challenges in long-term progression prediction. Our Task 1 solution demonstrates significant potential for improving AMD monitoring efficiency in clinical settings, while Task 2 results underscore areas needing further research. This work contributes to enhancing AMD management, with implications for improved patient care in ophthalmology.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Leveraging MaxVit on Fused OCT Scan Pairs for Age-Related Macular Degeneration Evolution Assessment

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yosuke Yamagishi

摘要

This paper presents our approach to the MARIO (Monitoring Age-related Macular Degeneration Progression In Optical Coherence Tomography) challenge at MICCAI 2024, focusing on automated analysis of Age-related Macular Degeneration (AMD) evolution using Optical Coherence Tomography (OCT) images. For Task 1, we propose a novel method that fuses consecutive OCT scan pairs, utilizing a MaxVit Tiny architecture to classify AMD progression. This approach mimics clinicians’ side-by-side comparison technique. In the preliminary phase, our model ranked first among 20 teams across all evaluation metrics for Task 1, achieving an F1 score of 0.863, Matthews Rk-correlation coefficient (RkC) of 0.709, and Specificity of 0.913. For Task 2, predicting AMD progression over three months, we combined EfficientNet V2 S for OCT image processing with a Multi-Layer Perceptron for handling patient metadata. While our Task 2 solution ranked competitively (3rd in F1, 1st in QWK among 16 teams), overall scores were low (F1: 0.685, RkC: 0.111, QWK: 0.231, Specificity: 0.688), highlighting challenges in long-term progression prediction. Our Task 1 solution demonstrates significant potential for improving AMD monitoring efficiency in clinical settings, while Task 2 results underscore areas needing further research. This work contributes to enhancing AMD management, with implications for improved patient care in ophthalmology.