Shearlet-based texture analysis and deep learning for osteoporosis classification in lumbar vertebrae

Abstract

Osteoporosis is a bone disorder characterized by reduced bone density and increased fracture risk. It challenges society's health, remarkably among the elderly population. This research proposed an innovative method by combining Shearlet-transform (ST) spectral analysis with a deep learning neural network (DLNN) and a convolutional neural network (CNN), for osteoporosis classification in lumbar vertebrae (LV) L1-L4 of spine X-ray images. The ST enables precise extraction of texture features from images by capturing significant information regarding trabecular bone micro-architecture and bone mineral density (BMD) variations revealing in osteoporosis regions. These extracted features serve as input to a DLNN for automated classification of osteoporotic and non-osteoporotic vertebrae. Similarly, without extracting any features from ST image is directly used as an input to the CNN to classify the images. The experimental results highlight the framework's effectiveness, achieving 96% accuracy in osteoporosis image classification using CNN. Early and precise detection of osteoporosis, particularly in the lumbar vertebrae, is vital for effective treatment and fracture prevention. This study particularly emphasizes the potential and effectiveness of integrating image spectral analysis technique with NN, to improving diagnostic accuracy and clinical decision-making in osteoporosis management.