Scholarlink

<p>Glaucoma is a group of optic neuropathy disorders related to the degeneration of retinal ganglion cells. It is complex to diagnose for the variety of clinical manifestations and the presence or absence of symptoms. Color Fundus Photographs (CFP) and Optic Coherence Tomography (OCT) are two of the principal imaging techniques used during evaluations by clinicians. An enriched and complementary view of the retina could be reached by using both modalities. In this work we propose a Convolutional Neural Network (CNN) ensemble consisting of two modules (by image type) and then their mixture into a single prediction via fusion techniques (intermediate fusion, late fusion) to classify glaucomatous cases. Our approach achieved acc = 91.88%, prec = 89.71%, recall = 92.84%, F1 = 90.93% and acc = 94.02%, prec = 94.06%, recall = 94.05%, F1 = 94.02% for the ensemble of each respective modality (CFP, OCT). The performance was of acc = 93.60%, prec = 83%, recall = 87.03%, F1 = 84.85% and acc = 86.84%, prec = 87.97%, recall = 86.39%, F1 = 86.61% for two testing databases from mexican and pakistan populations, respectively.</p>

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

Glaucoma classification using deep learning ensemble and multi-modal image modality

  • 期刊论文

  • 发表日期：

    2026年1月13日

查看全文
  • Rodrigo García-López,
  • Saúl Tovar-Arriaga,
  • Gendry Alfonso-Francia,
  • Mariana Badillo-Fernández

摘要

Glaucoma is a group of optic neuropathy disorders related to the degeneration of retinal ganglion cells. It is complex to diagnose for the variety of clinical manifestations and the presence or absence of symptoms. Color Fundus Photographs (CFP) and Optic Coherence Tomography (OCT) are two of the principal imaging techniques used during evaluations by clinicians. An enriched and complementary view of the retina could be reached by using both modalities. In this work we propose a Convolutional Neural Network (CNN) ensemble consisting of two modules (by image type) and then their mixture into a single prediction via fusion techniques (intermediate fusion, late fusion) to classify glaucomatous cases. Our approach achieved acc = 91.88%, prec = 89.71%, recall = 92.84%, F1 = 90.93% and acc = 94.02%, prec = 94.06%, recall = 94.05%, F1 = 94.02% for the ensemble of each respective modality (CFP, OCT). The performance was of acc = 93.60%, prec = 83%, recall = 87.03%, F1 = 84.85% and acc = 86.84%, prec = 87.97%, recall = 86.39%, F1 = 86.61% for two testing databases from mexican and pakistan populations, respectively.