Automated and robust lesion segmentation in PET/CT imaging can be achieved by incorporating tracer-specific characteristics and anatomical knowledge into deep learning models.
A compact deep learning model, PocketNet, can accurately segment cervical tumors and gynecologic organs on T2-weighted MRI, enabling efficient and consistent radiotherapy planning for cervical cancer patients.
Synthetic data augmentation using class-specific Variational Autoencoders (VAEs) and latent space interpolation can significantly improve the classification accuracy of esophagogastroduodenoscopy (EGD) images, especially for underrepresented classes, by addressing data scarcity and imbalance.
Automated segmentation of skeletal muscle, visceral adipose tissue, and subcutaneous adipose tissue from 2D MRI scans at the L3 vertebral level using DAFS Express software shows high accuracy and reliability compared to manual segmentation.
A framework for generating a large and diverse library of highly detailed, patient-specific anatomical models, representing human digital twins for research in medical imaging.
Subvolume merging technique effectively mitigates stitching artifacts in synthetic computed tomography (sCT) generated from magnetic resonance imaging (MRI), leading to improved image quality.
Nailfold capillaroscopy can identify specific capillary abnormalities associated with various nail conditions, providing a potential non-invasive diagnostic tool.
Two data augmentation methods, synthetic data generation and traditional image transformations, can improve the accuracy of machine learning models in detecting hemarthrosis, a key symptom of hemophilia.
Fusion of complementary information from imperfectly registered multimodal MRI can improve pancreas segmentation, but the optimal fusion location is model-specific and gains are small.
False-positive mammogram results can lead to a significant decrease in the likelihood of women returning for future routine breast cancer screenings.