Scholarlink

Kidney stones are a common and often painful health issue that can lead to serious complications if not identified early. This study proposes a novel method for detecting kidney stones by combining deep learning and machine learning techniques to improve diagnostic accuracy. ResNet18, a convolutional neural network, was used to analyze CT scan images and achieved an impressive test accuracy of 99% and a recall of 0.97. Additionally, a Random Forest model was applied to analyze urinary parameters such as specific gravity, pH, osmolarity, conductivity, urea, and calcium levels. This model achieved a test accuracy of 85% and a recall of 0.82, demonstrating its potential for non-invasive detection. By integrating the strengths of these models using ensemble learning with a 60:40 weight ratio, the results were further refined, significantly enhancing both precision and recall. This approach reduces the need for invasive procedures, offering a safer and more effective way to diagnose kidney stones early. The findings highlight the potential of combining imaging and biochemical analyses to improve diagnostic methods and enhance patient care. Future enhancements could involve the development of a unified dataset that integrates both CT images and urinary parameters, further improving diagnostic accuracy and scalability.

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

Hybrid Deep Learning and Machine Learning Approach for Non-invasive Kidney Stone Diagnosis: Integrating CT Imaging and Urinary Biochemical Analysis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Om S. Poonjani,
  • Om U. Ramani,
  • Sunil K. Vithalani,
  • Sandeep R. Suthar

摘要

Kidney stones are a common and often painful health issue that can lead to serious complications if not identified early. This study proposes a novel method for detecting kidney stones by combining deep learning and machine learning techniques to improve diagnostic accuracy. ResNet18, a convolutional neural network, was used to analyze CT scan images and achieved an impressive test accuracy of 99% and a recall of 0.97. Additionally, a Random Forest model was applied to analyze urinary parameters such as specific gravity, pH, osmolarity, conductivity, urea, and calcium levels. This model achieved a test accuracy of 85% and a recall of 0.82, demonstrating its potential for non-invasive detection. By integrating the strengths of these models using ensemble learning with a 60:40 weight ratio, the results were further refined, significantly enhancing both precision and recall. This approach reduces the need for invasive procedures, offering a safer and more effective way to diagnose kidney stones early. The findings highlight the potential of combining imaging and biochemical analyses to improve diagnostic methods and enhance patient care. Future enhancements could involve the development of a unified dataset that integrates both CT images and urinary parameters, further improving diagnostic accuracy and scalability.