Scholarlink

<p>Pollen allergy has become a significant public health concern. Although Artificial Intelligence (AI)-based automatic pollen identification has already demonstrated high accuracy with idealized test data, efforts toward practical application remain inadequate in Japan.&#xa0;In this study, we developed and implemented an AI-based airborne pollen monitoring system utilizing a Human-in-the-Loop (HITL) machine learning (ML) framework. A general pollen model trained on modern pollen images (Pre model) demonstrated high accuracy, precision, and recall scores during test data validation. However, the Pre model showed markedly different values compared with manual counting when deployed on actual airborne pollen slides collected at the National Hospital Organization Fukuoka National Hospital, especially for the main spring airborne pollen of Japanese cedar and cypresses. In contrast, fine-tuned models successfully distinguished these pollen taxa and resulted in substantially higher similarity indices between the manual pollen counts, which were attributed to active learning within the HITL ML framework.&#xa0;AI-based pollen monitoring shows promising potential for its practical implementation in the treatment of pollen allergies through effective and efficient strategies for active learning within the HITL ML process.</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

Practical implementation of Artificial Intelligence for airborne pollen monitoring in Japan through the Human-in-the-Loop machine learning

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • Ryoma Hayashi,
  • Takuya Itaki,
  • Ayumu Miyakawa,
  • Kazuhide Mimura,
  • Chie Oshikawa,
  • Eiko Koto,
  • Reiko Kishikawa

摘要

Pollen allergy has become a significant public health concern. Although Artificial Intelligence (AI)-based automatic pollen identification has already demonstrated high accuracy with idealized test data, efforts toward practical application remain inadequate in Japan. In this study, we developed and implemented an AI-based airborne pollen monitoring system utilizing a Human-in-the-Loop (HITL) machine learning (ML) framework. A general pollen model trained on modern pollen images (Pre model) demonstrated high accuracy, precision, and recall scores during test data validation. However, the Pre model showed markedly different values compared with manual counting when deployed on actual airborne pollen slides collected at the National Hospital Organization Fukuoka National Hospital, especially for the main spring airborne pollen of Japanese cedar and cypresses. In contrast, fine-tuned models successfully distinguished these pollen taxa and resulted in substantially higher similarity indices between the manual pollen counts, which were attributed to active learning within the HITL ML framework. AI-based pollen monitoring shows promising potential for its practical implementation in the treatment of pollen allergies through effective and efficient strategies for active learning within the HITL ML process.