Scholarlink

The automatic detection and classification of cervical cells is a crucial task before evaluating the samples in search for signs of cancer development. The paper describes a detection system based on the concepts of concurrency and specialization, that includes pre-processing of Papanicolaou smear images into overlapping patches of a fixed size, training separate CNNs for squamous and glandular cells on such patches and resolving potential conflicts by comparing the networks’ output values and their confidence in the classification. All parts of the system are flexible and scalable, from the dimensions of the input images to the number and type of modules that can analyze the patches concurrently. The method has been trained and tested with encouraging results on a database collected in the past couple of years, containing 6624 patches and including 99 squamous cells and 395 glandular cells. Individual CNNs and the full system obtain a total success rate above 89.5%.

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

Detection of Squamous and Glandular Cervical Cells Using Concurrent Convolutional Neural Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Dumitru-Viorel Zăbavă,
  • Rareş Ştefan Teodorescu,
  • Cătălina Neghină,
  • Mihai Neghină,
  • Florian Vintilă Armăşescu

摘要

The automatic detection and classification of cervical cells is a crucial task before evaluating the samples in search for signs of cancer development. The paper describes a detection system based on the concepts of concurrency and specialization, that includes pre-processing of Papanicolaou smear images into overlapping patches of a fixed size, training separate CNNs for squamous and glandular cells on such patches and resolving potential conflicts by comparing the networks’ output values and their confidence in the classification. All parts of the system are flexible and scalable, from the dimensions of the input images to the number and type of modules that can analyze the patches concurrently. The method has been trained and tested with encouraging results on a database collected in the past couple of years, containing 6624 patches and including 99 squamous cells and 395 glandular cells. Individual CNNs and the full system obtain a total success rate above 89.5%.