TY - JOUR
AU - Bagherifard, Abolfazl
AB - ObjectiveToday, lifestyle changes cause spine abnormalities. Radiographic images are used to diagnose scoliosis, a common spine abnormality. Existing machine-learning algorithms are essential to assist doctors in diagnosis, treatment planning, and interventional guidance. As a subset of machine learning algorithms, deep neural networks can automatically extract features from images to segment and classify medical problems. To enhance the interpretability of radiographic images, this study used a deep, improved U-Net neural network to automatically segment spinal vertebrae.MethodBy modifying the architecture and loss function, the improved U-Net network could be more accurate. To enhance the interpretability of radiographic images and expedite evaluation, this study focuses on the automated segmentation of spinal vertebrae using the improved U-Net neural network. In this machine-learning algorithm, for every deviation between the target label and the network output, the penalty of the loss function is taken into account during the back-propagation of the partial derivative of the loss function.ResultsFinally, the new architecture of the improved U-Net network and combining the weighted cross-entropy loss and dice loss, focusing them on imbalanced and overlapping data sets will increase the accuracy of the improved U-Net network by 4% compared to the conventional U-Net network with binary cross-entropy loss function for the segmentation of the columns’ spine vertebrae, making it a better and more accurate interpretation of the spine by the doctors.ConclusionThis improves the interpretation of the vertebral column for medical professionals and has a positive impact on orthopedists’ assessments of patients with spinal deformities.
TI - Rethinking U-Net Deep Neural Network for Spine Radiographic Images-Based Spine Vertebrae Segmentation
JF - Journal of Medical and Biological Engineering
DO - 10.1007/s40846-023-00828-6
DA - 2023-10-01
UR - https://www.deepdyve.com/lp/springer-journals/rethinking-u-net-deep-neural-network-for-spine-radiographic-images-ip5n90OuoT
SP - 574
EP - 584
VL - 43
IS - 5
DP - DeepDyve
ER -