Scholarlink

Medical record summarization would be a potentially viable intervention to improve accessibility and make informed decisions in the context of low-resource care setting. In this paper, I have outlined a hybrid way of summarizing, which combines the two BERTs in order to gain intricate insight into the contextual language and VGG-19 to gain visual characteristics, which are premised on the scanned or picture-based articles on medicine. The optical character recognition (OCR) is used to read the textual contents and a layout structure and visual cues of the document can be read using VGG-19. A small portion of tagged medical text-summaries is analyzed to process and filter the retrieved information with BERT. This unavailability of annotated datasets has been addressed by learning to generate pseudo-labels on large volumes of unlabeled data extracted by OCR on a semi-supervised learning scheme. The suggested system reflects an effective solution to the way the access to the clinical information limited by the resources is facilitated to the point of providing the concise, contextual and medically important summaries.

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

OCR-Augmented Document Summarization with Semi Supervised BERT in Low-Resource Settings

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • K. Madhu Suganya,
  • P. M. Thirishala,
  • S. Janani Sri,
  • M. Naga Soundari,
  • V. Sukanth

摘要

Medical record summarization would be a potentially viable intervention to improve accessibility and make informed decisions in the context of low-resource care setting. In this paper, I have outlined a hybrid way of summarizing, which combines the two BERTs in order to gain intricate insight into the contextual language and VGG-19 to gain visual characteristics, which are premised on the scanned or picture-based articles on medicine. The optical character recognition (OCR) is used to read the textual contents and a layout structure and visual cues of the document can be read using VGG-19. A small portion of tagged medical text-summaries is analyzed to process and filter the retrieved information with BERT. This unavailability of annotated datasets has been addressed by learning to generate pseudo-labels on large volumes of unlabeled data extracted by OCR on a semi-supervised learning scheme. The suggested system reflects an effective solution to the way the access to the clinical information limited by the resources is facilitated to the point of providing the concise, contextual and medically important summaries.