Affine Registration of Plantar Foot Thermal Images with Deep Learning: Application to Diabetic Foot Diagnosis

Early prevention of diabetic foot ulceration is possible by using the plantar foot temperature that can be measured with a thermal camera. In this work, we performed the plantar foot registration using three Deep Learning methods. These methods include two parts: an affine registration module for estimating transformation parameters and a spatial transformer for getting the registered image. All three models performances were evaluated using the Dice similarity coefficient (DSC), Mean Square Error (MSE), and peak signal-to-noise ratio (PSNR). Our aim was to find an accurate, fully convolutional neural network suitable for our database of thermal images of diabetic feet. Results showed that Affine ConvNet and DLIR (affine part) models produce the best plantar foot affine registration results with a Dice score of 95%.