Diabetic foot ulcer classification of hybrid convolutional neural network on hyperspectral imaging

Diabetic foot ulcer is a chief distress of diabetes mellitus. The diabetic foot ulcer (DFU) is the extremely injurious dilemma associated to diabetes mellitus. DFU is a risky illness, and it desires customary cure or else it might direct to foot amputation. When not treated it leads to some health issues and hence a novel method is proposed for efficient classification of DFU images. The DFU in this research is categorized into four classes like normal foot, high risk foot, ulcerated foot, and infected foot. Initially, a DFU dataset is made utilizing hyperspectral DFU images and pre-processing is done with aid of adaptive median filter. Consequently, the image is segmented by improved fuzzy c-means - particle swarm optimization algorithm. Then, a count of second order statistical texture features comprising entropy, energy; correlation, homogeneity, and contrast are produced via Gray Level Co-occurrence Matrix (GLCM). Finally, images are classified with aid of novel hybrid convolution neural network along with support vector machine. Here the novelty is derived by use of a new regularizer. The experiment is done with a manually created dataset. The performance evaluation is done by computing recall, precision, F1-score and accuracy. The results are compared with existing algorithms that show that the proposed hybrid system gives high classification accuracy.