Developing Imaging and Analysis Tools for Lymphoscintigraphy in Lymphedema
openNIBIB - National Institute of Biomedical Imaging and Bioengineering
Project Summary
Lymphedema occurs when a part of the lymphatic system fails to remove lymph fluid resulting in edema and impaired
immune function. Lymphedema impacts more than 10 million individuals in the United States (US) and is frequently
reported as a complication in cancer-related surgery, affecting up to 30% of breast cancer survivors, as well as patients
treated for prostate, ovarian, head and neck cancers, and melanoma. Nevertheless, lymphedema remains incurable. Despite
conservative treatments like compression therapy and physiotherapy, disease progression occurs, which reduces patient
quality of life and increases healthcare costs. The current standard imaging for lymphedema diagnosis is lymphoscintigraphy
(LSG). LSG (2D planar imaging) shows the distribution of counts detected by gamma cameras at two fixed angles
(anterior/posterior positions). LSG is widely available, easy to perform, and has long been considered the gold standard
owing to its high sensitivity for lymphedema diagnosis. Since advanced surgical treatments such as lymphatic venous
anastomosis are emerging, enhancing LSG is essential to guide lymphedema care and innovation. The primary limitation of
2D LSG is noise due to low photon counts. Lymphatic vessels are relatively small, lymph flow is generally much slower
than blood flow, and gamma cameras capture only about 0.01% of emitted photons fromlymph fluid due to their low
sensitivity. Consequently, LSG may require extended scan durations to collect sufficient photons, often resulting in noisy,
low-quality images. In addition, attenuation correction (AC) is not available in 2D planar imaging, leaving LSG non-
quantitative without calculating absolute activity concentrations (Bq/mL) at each voxel. Currently, visual assessment by
physicians (qualitative analysis) remains the clinical standard for evaluating lymphedema. Therefore, to address the unmet
needs of classical 2D LSG, we will develop a denoising solution (Aim 1) and quantitative imaging and analysis tools (Aim
2). In Aim 1, we will develop a deep learning (DL)-based denoising solution for LSG. A clinical dataset will be created by
prospectively acquiring 150 studies as well as recruiting 10 subjects undergoing lymphatic venous anastomosis (Task 1.1).
To address clinical data diversity, an anthropomorphic digital phantom dataset (1,000 subjects) will be created using the
XCAT phantom that includes organs, muscles, bones, soft tissues, and a scalable lymphatic system (Task 1.2). Using these
datasets, two state-of-the-art networks for PET denoising will be adapted for LSG denoising (Task 1.3). In Aim 2, we will
develop DL-based attenuation correction (DL-AC) methods for quantitative LSG (qLSG) and establish quantitative analysis
metrics for objective evaluation. For qLSG, a convolutional neural network (CNN) will be trained to translate CT scout
images (2D topograms) to 2D μ-maps for AC in LSG (scout-2-μ-map translation). In addition, a direct DL-AC model will
be developed to translate 2D anterior/posterior planar images (input) to 2D true activity maps (output) (NC-2-AC
translation) (Task 2.1). We will develop quantitative imaging biomarkers using qLSG and evaluate their correlation with
visual assessment by physicians (Task 2.2). The proposed project will enhance LSG by developing quantitative imaging and
analysis tools readily translatable to clinical practice, which will improve care for patients with lymphedema in various
clinics across the Plastic Surgery Department, the Lymphedema Center, and Children’s Medical Center.
Up to $643K
health research