Classification method of peripheral arterial disease in patients with type 2 diabetes mellitus by infrared thermography and machine learning

Padierna, Luis Carlos; Amador-Medina, Lauro Fabián; Murillo-Ortíz, Blanca; Villaseñor‐Mora, Carlos

doi:10.1016/j.infrared.2020.103531

Cited by 15 publications

(5 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In a similar way, these results were also found with the BNB model, which also achieved a very good result of 77.2% accuracy. However, this result is significantly different from that achieved in [19] since in that work, it was used to classify peripheral arterial disease in patients with type 2 diabetes using ML models, and the model achieved a performance of 92% accuracy and a sensitivity of 91.80%. The difference in accuracy results is mainly due to the dataset with which it is processed and the characteristics it includes.…”

Section: Discussioncontrasting

confidence: 69%

“…As a result, the XG boost classifier obtained the best metric with 0.9013 in cross-validation, and the model with the lowest performance was SVM with 61.17%. The work [19] proposed a non-invasive methodology for the analysis of features extracted from tomographic images, and they used SVM as an ML classification model. The model performance using cross-validation reached an accuracy of 92.64% and a sensitivity of 91%.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Application of Machine Learning Models for Early Detection and Accurate Classification of Type 2 Diabetes

Iparraguirre-Villanueva,

Espinola-Linares,

Flores Castañeda

et al. 2023

Diagnostics

View full text Add to dashboard Cite

Early detection of diabetes is essential to prevent serious complications in patients. The purpose of this work is to detect and classify type 2 diabetes in patients using machine learning (ML) models, and to select the most optimal model to predict the risk of diabetes. In this paper, five ML models, including K-nearest neighbor (K-NN), Bernoulli Naïve Bayes (BNB), decision tree (DT), logistic regression (LR), and support vector machine (SVM), are investigated to predict diabetic patients. A Kaggle-hosted Pima Indian dataset containing 768 patients with and without diabetes was used, including variables such as number of pregnancies the patient has had, blood glucose concentration, diastolic blood pressure, skinfold thickness, body insulin levels, body mass index (BMI), genetic background, diabetes in the family tree, age, and outcome (with/without diabetes). The results show that the K-NN and BNB models outperform the other models. The K-NN model obtained the best accuracy in detecting diabetes, with 79.6% accuracy, while the BNB model obtained 77.2% accuracy in detecting diabetes. Finally, it can be stated that the use of ML models for the early detection of diabetes is very promising.

show abstract

Section: Discussioncontrasting

confidence: 69%

Section: Related Workmentioning

confidence: 99%

Application of Machine Learning Models for Early Detection and Accurate Classification of Type 2 Diabetes

Iparraguirre-Villanueva,

Espinola-Linares,

Flores Castañeda

et al. 2023

Diagnostics

View full text Add to dashboard Cite

show abstract

“…Hüsers et al [ 17 ] presented the transfer learning method to detect the wound maceration and was able to achieve a recall of 0.69. Carlos Padierna et al [ 18 ] extracted features from infrared images of the upper side of the foot and toes to propose a classification approach for finding PAD and achieved 92.64% using SVM. Adam et al [ 19 ] developed density dual-tree complex wavelet transform in an automated detection method to locate diabetic feet with and without neuropathy.…”

Section: Related Workmentioning

confidence: 99%

Automated Detection of Infection in Diabetic Foot Ulcer Images Using Convolutional Neural Network

Yogapriya¹,

Elakkiya³

et al. 2022

Journal of Healthcare Engineering

View full text Add to dashboard Cite

A bacterial or bone infection in the feet causes diabetic foot infection (DFI), which results in reddish skin in the wound and surrounding area. DFI is the most prevalent and dangerous type of diabetic mellitus. It will mainly occur in people with heart disease, renal illness, or eye disease. The clinical signs and symptoms of local inflammation are used to diagnose diabetic foot infection. In assessing diabetic foot ulcers, the infection has significant clinical implications in predicting the likelihood of amputation. In this work, a diabetic foot infection network (DFINET) is proposed to assess infection and no infection from diabetic foot ulcer images. A DFINET consists of 22 layers with a unique parallel convolution layer with ReLU, a normalization layer, and a fully connected layer with a dropout connection. Experiments have shown that the DFINET, when combined with this technique and improved image augmentation, should yield promising results in infection recognition, with an accuracy of 91.98%, and a Matthews correlation coefficient of 0.84 on binary classification. Such enhancements to existing methods shows that the suggested approach can assist medical experts in automated detection of DFI.

show abstract

“…Several other studies also reported that the contralateral feet' 2.2 •C cutoff temperature difference indicates abnormality [82]- [83]. Peregrina-Barreto et al presented a methodology to detect ulceration risks by comparing temperature differences in four different regions between feet.…”

Section: Peripheral Arterial Diseasementioning

confidence: 99%

“…It is unclear why this specific ROI should be more clinically crucial than selecting the whole foot sole or regions within the sole for temperature evaluation. Padierna et al proposed a method to classify PAD and non-PAD patients based on 12 thermal features extracted from the upper side of the foot and toes [82]. Ilo et al, in another paper, showed that an increase in ABI and toe pressure (TP) was associated with an increase in the average temperature of the foot before and after revascularization [85].…”

Section: Peripheral Arterial Diseasementioning

confidence: 99%

Advances in medical infrared thermography

Al-Sadr¹

View full text Add to dashboard Cite

[EMBARGOED UNTIL 6/1/2023] The association between illness and body temperature dates back to the beginnings of medicine. Over the last few years, infrared thermography has attracted increased attention in various medical applications due to technological advancements that reduce manufacturing costs and enable high-resolution cameras with great sensitivity. The recent 2019 coronavirus disease (COVID-19) pandemic revealed that disease transmission control is crucial in preventing an outbreak. Thermography has been considered the gold standard for screening fevered individuals during pandemics since the 2003 SARS outbreak. With the increasing demands for objective and quantitative diagnostic tools in primary health care, thermography offers advantages over conventional approaches as a low-cost, non-invasive, radiation-free, and pain-free method for detecting changes in skin temperature with high spatial accuracy. Several topics are investigated in this dissertation to address the use of infrared imaging as a quantitative tool in medical applications. Peripheral arterial disease (PAD) is a common circulatory problem that affects 202 million patients globally and more than 20 percent of people over 70. The ankle-brachial index (ABI) measurement has been considered the most reliable diagnostic test for PAD. However, some limitations exist where ABI may give false normal values above 1.4 for patients with diabetes or chronic kidney disease with rigid and non-compressible arterial vessel walls. Additionally, ABI must be performed in a vascular laboratory by trained vascular staff to ensure the accuracy of the measurements. During the physical examination, one of the PAD signs is the reduction in the patient's feet and legs' skin temperature, which offers the potential of using infrared thermography in diagnosing PAD. Thus, we conducted a study based on 12 patients recruited by vascular surgeons at the University of Missouri Hospital outpatient clinic. These patients were diagnosed with PAD and scheduled for intervention after clinical evaluation. We developed a framework based on Gaussian Process Regression (GPR) with LOSO cross-validation to correlate the thermal features with the corresponding ABI. Additionally, we proposed a classification model based on the weighted K-Nearest neighbors (WKNN) algorithm to assess the severity of PAD based on infrared thermography. The regression results demonstrate a strong association between the thermal features and the ABI values. The proposed WKNN classification model shows thermography's capabilities in assessing the severity of PAD with high accuracy, sensitivity, and specificity. Since PAD patients are more likely to develop wounds and ulcers in the foot region than in the leg, we conducted another study with 10 PAD patients where the foot and ankle region was selected as the targeted ROI. Using SVM regression with LOSO cross-validation and Bayesian optimization, the proposed also shows a strong correlation between the ABI and the extracted thermal features. These studies demonstrate the potential for thermography to be used as a non-invasive supplementary screening tool for PAD. Endovascular treatments, such as peripheral angioplasty and conventional bypass surgery, have been extensively adopted to treat PAD. Despite promising initial results with surgical interventions, long-term durability leads to more surgeries and an increased risk of amputation, disability, and mortality. Following surgical treatments, the current standard of care is a clinic-based surveillance program consisting of office visits every 3 to 6 months during the patient's lifetime. Surgical intervention failures occur in up to 49 percent of patients after two years. However, any failure that is recognized early can be treated effectively with elective surgery, which significantly minimizes the risk of amputation. Therefore, we proposed a novel home-based, patient-operated early warning system to detect declining perfusion following surgical interventions. The study group consists of 10 PAD patients recruited by vascular surgeons at the University of Missouri Hospital outpatient clinic. We proposed a novel PAD Temperature Index Score (PTI) based on the FCM algorithm to interpret changes in lower limb temperature distribution patterns. The proposed PTI index shows stable and repeatable results in recognizing the impact of the revascularization procedure across different scenarios. Additionally, we examine the potential of deep learning-based thermal image analysis in determining the severity of PAD for our in-home surveillance system. ABI is utilized as ground truth in categorizing PAD severity into three classes (severe, moderate, and mild/normal). Based on 10-fold cross-validation, the model achieved high accuracy, sensitivity, specificity, and precision. As a result, the proposed framework provides frequent surveillance that can result in earlier failure detection, lower the risk of amputation, and improve survival in PAD patients. Detecting abnormal body temperature patterns can help predict sepsis before any other symptoms of infection. As a result, we propose using thermography as a non-invasive tool to measure body temperature patterns and detect abnormalities. We hypothesize that sepsis onset will lead to abnormal body temperature patterns, which can be detected by calculating the temperature difference (ETD) between the core and the body extremities. We use "cold stress" induced by cold water immersion (CWI) as a proxy for sepsis. Toward this goal, we proposed a fully automated methodology for calculating core vs. extremity ETD based on the frontal and lateral view of the face. We developed a framework based on FCM clustering for inner and outer ear localization for the frontal view of the face. While for lateral view, we implement an efficient approach for tracking the tip of the nose and the inner corner of the eyes by using a mixture of Viola-Jones, KLT, and superpixel algorithms. The results produce specific core versus body extremities temperature patterns that can be used to simulate body response to sepsis. Additionally, we proposed a deep learning classification procedure for sepsis detection. The study is based on 140 patients being evaluated for sepsis at the University of Missouri Emergency Department. The result demonstrates the viability of infrared imaging with deep learning models for future use in the emergency department setting to detect sepsis. As a result, these studies highlight the potential of using infrared thermography as a suitable screening approach in different medical applications.

show abstract

Classification method of peripheral arterial disease in patients with type 2 diabetes mellitus by infrared thermography and machine learning

Cited by 15 publications

References 14 publications

Application of Machine Learning Models for Early Detection and Accurate Classification of Type 2 Diabetes

Application of Machine Learning Models for Early Detection and Accurate Classification of Type 2 Diabetes

Automated Detection of Infection in Diabetic Foot Ulcer Images Using Convolutional Neural Network

Advances in medical infrared thermography

Contact Info

Product

Resources

About