Ex Vivo CT of GI and GU Tumors For Radiological and Pathological Correlation

The PI of this project is:

This project is funded by: Discretionary Research Funding

The term of this project is: January 2021 to August 2022

The number of subjects scanned during this project is: 100

Spatial heterogeneity is a common feature of many tumor types, with multiple studies demonstrating variability within tumors with respect to pathologic features, genomics, RNA expression, mutation analysis and protein expression. Radiologic evaluation allows analysis of a broad landscape of tumor features, and non-invasive imaging biomarkers may be useful in identifying tumors, as well as describing and characterizing them. Advanced CT techniques such as Dual energy CT (DECT) or CT post processing tools that allow high throughput extraction of a variety of imaging features (radiomics, e.g. CT Texture analysis or volumetric assessment) may be useful in evaluation of ex vivo tumors and could allow more direct radiologic-pathologic correlation. For example, CT texture analysis is a tool that allows quantification of tumor heterogeneity that may be challenging to identify and objectively evaluate visually. Although texture features have shown promising associations with pathologic features and clinical outcomes in a variety of tumor types, radiologic findings have only broadly been compared with pathologic findings from random tumor samples features previously and more precise radiologic-pathologic correlation is needed to evaluate what these imaging features represent in the tumor microenvironment. As another example, prostate cancers may have very low zinc levels compared to background prostate and DECT may be able to create a zinc-water filter on CT imaging which could highlight areas of lower zinc concentration and aid in identification of tumors. In addition, in the past, a classical approach to radiomics has been taken with more manual extraction of human selected features. There may be utility of application of machine learning analysis or deep learning models to this type of data