LMGU-NET: methodological intervention for prediction of bone health for clinical recommendations

Osteoporosis (OP) is a bone-related ailment that aggravates owing to the decline in bone mineral density (BMD) or during deviations in the structure or quality of bone that may surge to fractures. The low BMD can be recognized from computed tomography (CT), X-ray, or Dual Energy X-ray absorptiometry (DXA/DEXA). Texture analysis is the most significant and distinguishing image feature. An enhanced discrimination power texture feature extraction system is developed for volumetric images by combining two complementary types of information: local binary patterns (LBP) and normalized grey-level co-occurrence matrix-based (nGLCM) techniques to extract features and U-Net for classification. The developed algorithm was validated on a Kaggle dataset comprising X-ray images acquired from patients suffering from osteoporosis. The modified U-Net (ModU-Net) semantic segmentation classifier is used for segmenting the low bone mass sections in the processed image. The developed LGMU-Net algorithm outperforms conventional feature extraction approaches and neural networks with a Dice Score of 88.82%, Tanimoto Co-efficient index of 71.74%, MSE of 0.0321, and PSNR of 65.74 dB. This method assists physicians in making early diagnoses and also protects patients from bone fraility and eventual fractures by ensuring that they follow the medications/surgery options prescribed by the doctors.