Scholarlink

The incidence of Diabetes tends to increase each year. Neuropathy in the foot causes several problems that are not detectable in the early stages and, with scarce treatment, can lead to major risk complications, including ulcers and, in severe cases, amputation. Infrared thermography emerges as a useful non-invasive method for detecting temperature differences in the plantar area, which helps in identifying foot risk. Thermographic images can be analyzed through image classification, allowing them to divide different areas of the foot based on their risk. One of the most used architectures for this kind of task is the Convolutional Neural Network (CNN). This paper presents a CNN model ResNet-18 adapted to binary classification. In addition, data augmentation has been applied to the dataset to increase the number of images, since the original dataset was limited. The model uses transfer learning and achieves a precision of 83.02%, a recall of 90.00%, and specificity of 80.85%. These results are viable for clinical applications and require low computational cost.

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

Thermal Analysis for Diabetic Foot Image Classification Using ResNet-18: An Approach for Limited Data Scenarios

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jesús Antonio Garduño-Aguirre,
  • América Itzel Montaño-Mata,
  • Rafael Bayareh-Mancilla,
  • Lorenzo Leija-Salas,
  • Arturo Vera-Hernandez

摘要

The incidence of Diabetes tends to increase each year. Neuropathy in the foot causes several problems that are not detectable in the early stages and, with scarce treatment, can lead to major risk complications, including ulcers and, in severe cases, amputation. Infrared thermography emerges as a useful non-invasive method for detecting temperature differences in the plantar area, which helps in identifying foot risk. Thermographic images can be analyzed through image classification, allowing them to divide different areas of the foot based on their risk. One of the most used architectures for this kind of task is the Convolutional Neural Network (CNN). This paper presents a CNN model ResNet-18 adapted to binary classification. In addition, data augmentation has been applied to the dataset to increase the number of images, since the original dataset was limited. The model uses transfer learning and achieves a precision of 83.02%, a recall of 90.00%, and specificity of 80.85%. These results are viable for clinical applications and require low computational cost.