Efficient Diabetic Foot Ulcer Classification Using Residual Linearly Clamped Layers with Detachable Convolution Layer in DCNN Xception Model

Diabetes Foot Ulcer (DFU) is caused in the foot of diabetes patients and if the disease is not diagnosed within time, it may lead to various complications and severities. The major symptoms of DFU are foot swelling, dry cracks, leg pain, varied temperature of skin and changes in skin color. When DFU is not treated properly complications tend to remove limb or foot of a person. To overcome these limitations, automatic methods are required for detecting and classifying healthy skin and DFU images. However, the classification of DFU from existing research provides challenges like resemblance between healthy skin and DFU and has high intra-class and inter-class variations in accordance with the lighting conditions and patient’s ethnicity. The collection and labeling of DFU requires large time and it is complex to differentiate a DFU and a healthy skin, for instance when there are images with a toe, skin wrinkles and small sizes of DFU. Hence, the proposed system implemented a Deep Convolutional Neural Network (DCNN) based Xception model, to involve binary classification of healthy and DFU skin images. The DCNN is employed to handle large dataset, but produces over-fitting issues. So, the proposed DCNN-based Xception classifier is integrated with Residual Linearly Clamped Layers (RLCL) consisting of minimum detached convolution layers. It helps in reducing over-fitting problems and thus increases precision and computational speed of the proposed system. Further, the input images are optimized by using image enhancement techniques such as Histogram equalization, Adaptive filter and Gamma correlation. Then the efficiency of the proposed system is evaluated based on the performance metrics namely precision, F1-score, recall and accuracy to validate the performance of proposed model with existing algorithms.