CC-GBLS: Collaborative-Competitive Representation-Based Graph Regularized Broad Learning System for Osteoporosis Diagnosis

Osteoporosis as a prevalent disease among middle-aged and elderly people has drawn more attention in recent years. Though new techniques such as deep learning has been applied in automatic osteoporosis diagnosis, it is still unsatisfied for the performance of clinical diagnosis because feature representation is not good enough in the process of automatic diagnosis and the inner relationship among the extracted features is also not be considered. To address these issues, a novel broad neural network named collaborative-competitive representation-based graph regularized broad learning system (CC-GBLS) is proposed in this paper and CC-GBLS based approach is then given to automatically diagnose the osteoporosis. In our proposed approach, we firstly extract the LBP and GLCM features of the original images and construct a relationship graph to obtain more information. Secondly, collaborative–competitive representation-based autoencoder (CCRAE) is integrated with GBLS to obtain better feature representation. Finally, we apply the CC-GBLS method to the realm of automatic osteoporosis detection. The effectiveness of our model is verified by comparing our proposed model with several other state-of-art models in two osteoporosis datasets.