Skip Ribbon Commands
Skip to main content


  1. Distributed Parameter model for tracer kinetic modeling of DCE MRI

    Our group has invented a distributed parameter (DP) model for tracer kinetic modeling of DCE MRI. [1] [2] This model allows separate determination of blood flow and capillary permeability-surface area product.

    Flow calculated by the DP model has been validated by water PET. [5] The DP model was able to predict patient outcome in a cohort of colorectal primary but not the Generalized Kinetic model (ktrans, ve) nor the adiabetic tissue homogeneity model (St Lawrence and Lee). [3] The DP model is also able to predict patient outcome in a Phase II trial of Pazopanib (multiple tyrosine kinase inhibitor, antiangiogenic) and Nasopharyngeal carcinoma. [4]

    Our group used the DP model together with a dual arterial input function (hepatic artery and portal vein) to better study focal liver lesions. [6] Studying human neuroendocrine liver metastases, we found that a wash-in wash-out pattern is explained by high intravascular space and low interstitial space; whereas a progressively enhancing pattern is explained by low intravascular space and high interstitial space. [7] We are able to separately calculate the % contribution of splenic flow and splanchnic flow to total portal blood flow by deconvolution. [10]

    Studying SCID mice with human tumor xenografts, we found tracer kinetics in the “non-enhancing” center has no convective transport described by traditional DCE MRI modeling. The transport process is predominantly related to diffusion. We describe this by measuring the diffusivity of gadolinium in DCE MRI studies. [8]

    We postulate that this is related to high interstitial fluid pressure in the tumor core and show a relationship between low gadolinium diffusion and the presence of necrosis. [9] We have publications in Independent Component Analysis, IVIM, and are interested in studying the relationship of the liver interstitial space with liver fibrosis and cirrhosis.

    We are currently working on using IVIM and tracer kinetic modeling to study renal fibrosis and to estimate GFR in animals. [11] We are currently working on modeling dynamic PET to yield binding affinity and volume of distribution. Our current work focuses on macrophage imaging with microPET and immune cell imaging with microMRI and microPET.

    We have also evaluated IVIM and diffusion tensor imaging (DTI) of early renal fibrosis induced in a murine model of streptozotocin induced diabetes. [12] We are currently performing a clinical study evaluating the presence of metastatic pelvic nodes using IVIM in patients with newly diagnosed rectal, endometrial and cervical carcinomas.

    We are also currently working on modeling dynamic PET to yield binding affinity and volume of distribution. Our current work focuses on macrophage imaging with microPET in a murine model as well as T cell imaging with microMRI and microPET using a novel approach wherein immune cells can be imaged non-invasively in a murine model post immunotherapy exposure. 

  2. RAPIER – RAdiology Pathology Information Exchange Resource

    We are also a collaborating partner in Project RAPIER which aims to create a Rad-Path datalake on liver lesions and several deployable AI-driven applications for clinical decision support.

    The clinically-useful applications we aim to develop include an AI-assisted clinical decision support system (CDSS) that suggests diagnoses and calls up a montage of similar images and diagnoses. Another AI application could help doctors generate reports with “autocomplete” function for common lesions. The combined multi-disciplinary multimodal inputs from RadPath data will also enable development of AI solutions that can “see” patterns that a clinician may not easily visualize, such as detecting a cancerous nodule in scans which appear normal, or predicting when a nodule may become cancerous.

    Future AI projects we hope to enable include the ability to train with highly accurate data that to investigate how liver cancer evolves over time with distinct progression patterns. This may herald advanced AI solutions that allow doctors to “see the future” by unravelling hidden patterns in complicated data. This will catalyse next generation AI-enabled CDSS and predictive analytics, guiding precision medical care.

    This is a joint project between SGH Department of Diagnostic Radiology, Department of Pathology, NCCS Division of Oncologic Imaging and ASTAR Institute of High Performance Computing. It is supported by “AI SINGAPORE TECH CHALLENGE (OPEN-THEME) FUNDING SCHEME” with a total approved grant amount of approximately S$2.8 million over a course of 36 months.