Abstract Peripheral Arterial Disease (PAD) identification is a complex task as a set of different factors cause this disease such as: smoking, diabetes mellitus, old age, hypertension, renal insufficiency, among others. Recently, non-invasive methods based on Infrared Thermography (IRT) are effective for the detection of type-2 diabetes and diabetic foot ulcers from plantar thermograms. However, we have not found studies on the characterization of PAD from the top of the foot. In this work, it is presented a non-invasive methodology for this characterization. We are proposing the analysis of relevant features extracted from IRT images of the upper side of the foot and toes. With these features, we built a Support Vector Classification model that encompasses the data from two groups of Mexican participants one includes twenty-three diabetic patients and the control group has twenty non-diabetic. The average performance of the classification model was estimated under a rigorous bootstrapping method on 1000 randomized and independent runs of 5-fold cross-validations and reached 92.64% of accuracy, 91.80% of sensitivity, and specificity of 93.59%. The experimental data and the source code of the proposed methodology are publicly available; it allows an easy implementation as a supporting tool for physicians in the identification of PAD.

中文翻译：

---

2型糖尿病患者外周动脉疾病的红外热像和机器学习分类方法

摘要 外周动脉疾病 (PAD) 识别是一项复杂的任务，因为一系列不同的因素会导致这种疾病，例如：吸烟、糖尿病、老年、高血压、肾功能不全等。最近，基于红外热像仪 (IRT) 的非侵入性方法可有效地从足底温度图检测 2 型糖尿病和糖尿病足溃疡。然而，我们还没有发现从脚的顶部表征 PAD 的研究。在这项工作中，它提出了一种用于这种表征的非侵入性方法。我们建议对从脚和脚趾上侧的 IRT 图像中提取的相关特征进行分析。有了这些特点，我们建立了一个支持向量分类模型，该模型包含来自两组墨西哥参与者的数据，一组包括 23 名糖尿病患者，对照组有 20 名非糖尿病患者。分类模型的平均性能是在严格的自举方法下对 1000 次随机和独立运行的 5 倍交叉验证进行估计的，准确率达到 92.64%，灵敏度达到 91.80%，特异性达到 93.59%。所提出方法的实验数据和源代码是公开的；它允许作为医生识别 PAD 的支持工具轻松实施。分类模型的平均性能是在严格的自举方法下对 1000 次随机和独立运行的 5 倍交叉验证进行估计的，准确率达到 92.64%，灵敏度达到 91.80%，特异性达到 93.59%。所提出方法的实验数据和源代码是公开的；它允许作为医生识别 PAD 的支持工具轻松实施。分类模型的平均性能是在严格的自举方法下对 1000 次随机和独立运行的 5 倍交叉验证进行估计的，准确率达到 92.64%，灵敏度达到 91.80%，特异性达到 93.59%。所提出方法的实验数据和源代码是公开的；它允许作为医生识别 PAD 的支持工具轻松实施。