Scholarlink

<p>Bacterial skin infections are common and often challenging to diagnose accurately and quickly. Current diagnostic methods such as clinical examination, microbiological culture, and histopathology are time-consuming, subjective, and sometimes lack sensitivity. To address these issues, we have developed an automated system that uses advanced image processing, skin lesion segmentation, feature extraction, and deep learning models to identify infectious skin conditions. In this work, the dataset includes not only bacterial but also fungal, parasitic, and viral skin infections, allowing the system to learn a broad spectrum of lesion variations and improve generalization across multiple disease types. The images are enhanced using the HSV color space and segmented using thresholding and morphological operations, followed by contour filtering to isolate the infected region. Specific features such as area, aspect ratio, mean intensity, and equivalent diameter are extracted to describe lesion geometry and improve classification accuracy. Ten advanced deep learning models were trained, with InceptionResNetV2 achieving the highest accuracy of 99.88% and a loss of 0.22. Additionally, EfficientNetB0 demonstrated the highest precision and F1 score (90.50%), while DenseNet201 achieved the best recall (93.50%). These results demonstrate the system’s potential to support dermatologists by making the diagnostic process faster, more reliable, and suitable for real-world clinical applications.</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

Automated Detection and Diagnosis of Bacterial Skin Infections using Deep Learning with Segmentation Techniques

  • 期刊论文

  • 发表日期：

    2025年12月26日

查看全文
  • Jasdeep Singh,
  • Jasmeen Gill,
  • Yogesh Kumar

摘要

Bacterial skin infections are common and often challenging to diagnose accurately and quickly. Current diagnostic methods such as clinical examination, microbiological culture, and histopathology are time-consuming, subjective, and sometimes lack sensitivity. To address these issues, we have developed an automated system that uses advanced image processing, skin lesion segmentation, feature extraction, and deep learning models to identify infectious skin conditions. In this work, the dataset includes not only bacterial but also fungal, parasitic, and viral skin infections, allowing the system to learn a broad spectrum of lesion variations and improve generalization across multiple disease types. The images are enhanced using the HSV color space and segmented using thresholding and morphological operations, followed by contour filtering to isolate the infected region. Specific features such as area, aspect ratio, mean intensity, and equivalent diameter are extracted to describe lesion geometry and improve classification accuracy. Ten advanced deep learning models were trained, with InceptionResNetV2 achieving the highest accuracy of 99.88% and a loss of 0.22. Additionally, EfficientNetB0 demonstrated the highest precision and F1 score (90.50%), while DenseNet201 achieved the best recall (93.50%). These results demonstrate the system’s potential to support dermatologists by making the diagnostic process faster, more reliable, and suitable for real-world clinical applications.