Dual-energy X-ray absorptiometry (DXA) is the primary modality for bone mineral density (BMD), offering a reliable clinical indicator for orthopedic diagnosis and treatment. In practical applications, there are several factors that could lead to variations in the grayscale distribution of dual-energy X-ray images. The domain shift of the dual-energy X-ray images often makes neural networks used for ulna and radius segmentation ineffective, significantly impacting bone density detection. In this paper, we propose an unsupervised Global Contextual Enhanced Attention-guided Domain Adaptation (GCEADA) framework to enhance the performance of the network in the ulna and radius segmentation task of the target domain. The proposed method extracts global context information to enhance attention expression, obtaining more representative domain invariant features. We evaluate the GCEADA in two cross-domain tasks and conduct ablation experiments to assess the performance of each component. The results indicate that the proposed attention module effectively improves the feature extraction capability of the discriminator and the discriminability and transferability of the framework.
The accurate assessment of sinus invasion plays a crucial role in determining the surgical approach and prognosis evaluation in renal cell carcinoma. Clinical practice often relies on histopathological examination, which is timeconsuming and invasive for patients. Deep learning is frequently employed to enhance physicians' efficiency and assist clinical diagnosis and treatment from an imaging perspective. However, data imbalance and inter-patient variations pose challenges to neural network feature extraction and generalization. In this study, we propose a proxy task based 3D ResNet network for predicting sinus involvement in renal cell carcinoma to effectively aid physicians in diagnosing and treating sinus invasion. The proposed network leverages proxy tasks and a 3D attention mechanism to extract finer feature representations from CT images, thus improving the accuracy of sinus involvement prediction. We evaluate the performance of the proposed network on multi-center data. The results demonstrate that our approach achieves state-ofthe-art performance compared to other competitive methods.
Fourier ptychographic microscopy (FPM) undergoes fast development since its proposal. In FPM, a large number of small images using small numerical aperture (NA) objective lens is generally required for the process of high-resolution image reconstruction. Although various methods have been proposed to shorten the acquisition time or reconstruct the image using sparse sample of full data, the scientific question is still there, i.e. where is the boundary of extreme sparse sampling for FPM, either from the theory or experimental perspective. In this paper, based on the in-house Incremental Inverse Dynamical Photon scattering (IIDPS) framework, we report that this artificial neural network based method, utilizing least absolute deviations error metric instead of the commonly used mean square error metric, is able to reconstruct image at very low sampling rate the simulation data for FPM, i.e. 5.67%, that is much lower than the reported results while other traditional method, like Gerchberg-Saxton-type method or other similar method, could not perform successful image reconstruction at such low sampling rate.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.