58: Spatial Characterization of the Tumor Microenvironment using Self Supervised Learning

Music City Center 

The immunological nature of the tumor microenvironment (TME) in cancer is strongly associated with treatment response and clinical outcomes. The dynamics driving these associations are believed to be spatially-dependent and occurring at various scales within the tissue. Thus, detection of the immunological features of the TME requires a holistic spatial analysis integrating information across scales. In this work we develop a computational pipeline automating the discovery and spatial localization of immunological characteristics of the tumor microenvironment in melanoma. We use a novel deep learning-based framework implementing a self-supervised spatial segmentation of multiplexed immunofluorescence tissue samples into interpretable categories representing distinct immunological features and cell types.

Deep Learning

Self Supervised Learning

Gigapixel Imaging

Cancer - Melanoma

Image Segmentation

Multiplexed Immunofluorescence 

Section on Statistical Learning and Data Science