Assertion of Low Bone Mass in Osteoporotic X-ray images using Deep Learning Technique

Abstract

Osteoporosis is a disease that causes low bone mass and bone tissue deterioration. The application of medical image segmentation in identifying the anomalies present in the multimodal medical images is utilized in this work to predict and locate the low bone mass section. The X ray dataset is downloaded from the freely available Kaggle repository contains osteoporosis and normal bone images. This paper utilizes Gabor Filter for the extraction of features and the modified U-net algorithm derived from Deep Convolutional Neural Network for segmentation. For improvement over classical U-Net, we had gone for Parametric Relu as an activation function to overcome the disadvantages of Relu while keeping the advantages of Relu intact. The performance metric parameters including Mean Squared Error (MSE), Peak Signal to Noise Ratio (PSNR), Tanimoto Co-efficient index (TC), Dice Overlap Index (DOI) and computational time (T) are evaluated andthe outcomes attained are remarkable.Some fuzzy logic dependent rules could be applied to the final outcome of image processing and deep learning techniques for clinical decision making and handling treatmenttherapy of those suffering from any of these orthopedic ailments in the regular clinical practices. The proposed method performs better on many aspects than the existing deep learning algorithms like Convolutional Neural Networks (CNN) and Artificial Neural Networks (ANN).