Scholarlink

Sensors have emerged as pivotal tools in the early detection and monitoring of infections, including Urinary Tract Infections (UTIs), enabling timely interventions and reducing potential complications. However, rapid and accurate diagnosis of UTIs is often hindered by the limitations of traditional microscopic methods. In this study, we propose an automated UTI detection system that leverages sensor-captured high-resolution microscopic images. The system integrates two relevant deep learning models: a YOLOv8 module for real-time segmentation of pus cells and a Vision Transformer (ViT) module for precise classification of these cells. The integration of sensor data with deep learning enables the development of a fast reliable diagnostic tool adaptable to various clinical settings. In fact, YOLOv8 has exceled in real-time segmentation with its superior detection capabilities, achieving a mean Average Precision (mAP) of 87.5% and an Intersection over Union (IoU) of 0.82. Meanwhile, ViT significantly enhances classification accuracy through its ability to capture long-range dependencies, attaining an accuracy of 94.3%. The system achieved an overall accuracy of 91.2% while processing each image in just 45.6 milliseconds, demonstrating its suitability for real-time clinical applications.

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

UTI-Dx-ViT: Enhancing UTI Diagnosis with YOLOv8 Segmentation and Vision Transformer-Based Classification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Amal Jlassi,
  • Sami Hafsi,
  • Walid Barhoumi

摘要

Sensors have emerged as pivotal tools in the early detection and monitoring of infections, including Urinary Tract Infections (UTIs), enabling timely interventions and reducing potential complications. However, rapid and accurate diagnosis of UTIs is often hindered by the limitations of traditional microscopic methods. In this study, we propose an automated UTI detection system that leverages sensor-captured high-resolution microscopic images. The system integrates two relevant deep learning models: a YOLOv8 module for real-time segmentation of pus cells and a Vision Transformer (ViT) module for precise classification of these cells. The integration of sensor data with deep learning enables the development of a fast reliable diagnostic tool adaptable to various clinical settings. In fact, YOLOv8 has exceled in real-time segmentation with its superior detection capabilities, achieving a mean Average Precision (mAP) of 87.5% and an Intersection over Union (IoU) of 0.82. Meanwhile, ViT significantly enhances classification accuracy through its ability to capture long-range dependencies, attaining an accuracy of 94.3%. The system achieved an overall accuracy of 91.2% while processing each image in just 45.6 milliseconds, demonstrating its suitability for real-time clinical applications.