Scholarlink

We propose an image retrieval system which reduces search space using the concept of selective classification. The proposed system can partially annotate a given medical radiology image, marking ambiguous parts as undefined. For illustration, a model was built using the publicly available IRMA dataset from ImageCLEF2009 medical annotation task, consisting of 14, 410 X-ray images. Mean average precision (MAP) and IRMA error score were used for model validation. The proposed model is compared with other state-of-the-art models reported in the literature. In a 193 class-code setup, it attains IRMA error score 101.36 and MAP 0.745, surpassing current state-of-the-art by 22.89 and 0.112, respectively. The results suggest that the proposed model is usable for fast and accurate retrieval of hierarchically labelled medical radiographs. Code required to recreate our experiments is available at https://github.com/AIlab-RITEH/radret/.

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

Selective Classifier Based Search Space Shrinking for Radiographs Retrieval

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Teo Manojlović,
  • Ivo Ipšić,
  • Ivan Štajduhar

摘要

We propose an image retrieval system which reduces search space using the concept of selective classification. The proposed system can partially annotate a given medical radiology image, marking ambiguous parts as undefined. For illustration, a model was built using the publicly available IRMA dataset from ImageCLEF2009 medical annotation task, consisting of 14, 410 X-ray images. Mean average precision (MAP) and IRMA error score were used for model validation. The proposed model is compared with other state-of-the-art models reported in the literature. In a 193 class-code setup, it attains IRMA error score 101.36 and MAP 0.745, surpassing current state-of-the-art by 22.89 and 0.112, respectively. The results suggest that the proposed model is usable for fast and accurate retrieval of hierarchically labelled medical radiographs. Code required to recreate our experiments is available at https://github.com/AIlab-RITEH/radret/.