Developing Imaging and Analysis Tools for Lymphoscintigraphy in Lymphedema

About This Grant

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.