Temperature prediction based on ANN linear regression with an LWIR sensor for the study of diabetic foot

• Main Authors: Mancilla, Rafael Bayareh, Daul, Christian, Martinez, Josefina Gutierrez, Salas, Lorenzo Leija, Wolf, Didier, Hernandez, Arturo Vera
• Format: Conference Proceeding
• Language: English
• Subjects: Artificial Neural Network; Cameras; Diabetic foot; IR medical thermography; Linear regression; Radiometry; Temperature; Temperature measurement; Temperature sensors; Training
• ISSN: 2642-3766
• DOI: 10.1109/CCE53527.2021.9633057

This work presents an improvement to a linear model obtained by Bayareh et al. in 2017, by training a single layer Perceptronand optimized by Gradient Descent. The objective of characterizing an IR radiometric sensor with a mathematical model is to predict the surface temperature of the body under study (e.g. diabetic foot) to perform quantitative thermography studies related to early detection through temperature difference. The data was obtained by measuring IR radiometric information on a phantom under a controlled environment. Subsequently, the predicted model was compared to the first characteristic equation. The model had an error of 0.14°C while the original model had an error of 1.28°C regard an industrial purpose camera, previously calibrated. The updated model could support the study of quantitative thermography with embedded systems whose sensors are not able to interpret temperature from factory settings, especially in studies focused on the difference of temperatures that support the diagnosis from the quantitative point of view.