In this paper, we propose a co-localization metric derived from the G-function and show that this metric is significantly correlated with adverse tumor signatures, such as larger tumor size, increased vascular density, and greater depth of invasion. Pre-print to be uploaded soon! This work was done in collaboration with the wonderful Dr. Michael Tetzlaff from the Department of Pathology at the M.D Anderson Cancer Center.
I was chosen as one of 13 attendees worldwide for the inaugural PhD summit at EPFL, Lausanne, Switzerland. The summit was designed to bring together final year PhD students to present and discuss their work with EPFL professors and students. The theme of this year was ‘Data-driven engineering in Life sciences’ (https://phdsummit.epfl.ch/)
I presented my work on ‘Towards Personalized Treatment: Leveraging structure in cancer imaging to predict clinical outcomes’. A video of my talk can be found here: https://phdsummit.epfl.ch/presentations/
Very grateful to the EPFL PhD summit organizing committee, especially Drs. Ali Sayed for the initiative, Pascal Frossard and Jose Millan for the lab visits, and the tireless and amazing Sebastian Gautsch for managing anything and everything for this trip.
Our paper titled “A Functional Spatial Analysis Platform for Discovery of Immunological Interactions Predictive of Low-Grade to High-Grade Transition of Pancreatic Intraductal Papillary Mucinous Neoplasms” has been published in Cancer Informatics’ Special issue on Ensemble Learning and Deep Learning in Cancer Genomic and Imaging Data. In this work, we enhance our G-function based spatial analysis framework by incorporating ideas from functional data analysis, to build a tool more effective at representing the structure in the G-function. We show that spatial metrics derived from our functional spatial platform is able to predict the risk of progression in IPMN’s with a higher accuracy than simpler metrics such as counts or the G-function AUC alone. AN ensemble of models built using counts and the proposed G-function MFPCA metric performs the best. The paper can be found here.
Our paper on using the spatial G-function to quantify tumor:immune interactions in multiplexed Immunofluorescence images from Non-Small Cell Lung Cancer patients has been accepted to Lung Cancer! [Paper link] We made two contributions in this paper: 1) Repurposed the G-function used in ecology to study predator-prey relationships for studying tumor-immune cell interactions in NSCLC, and 2) That a simple area under the curve (AUC) metric derived from the G-function representing tumor cell- regulatory T cell interaction correlates with poor outcome for patients, independently of clinical variables that doctors currently use! (such as age, smoking history, number of positive lymph nodes, size of tumor etc.)
Since the G-function is a unique signature of immune infiltration, my hope is that doctors will now incorporate our fast and easy-to-use algorithm in clinical practice!
Our work with Dr. Phyu Aung, Dr. Michael Tetzlaff (Dept. of Pathology, M.D Anderson Cancer Center) and Dr. Ignacio Wistuba’s lab (Dept. of Translational Molecular Pathology) on studying the co-localization of B7-H3 expression in endothelial cells of Merkel Cell Carcinoma (MCC) patients, has been accepted as an abstract to the prestigious United States and Canadian Academy of Pathology (USCAP) Annual Meeting in March 2018. MCC is a very aggressive form of skin cancer, and patient response is closely linked to immune system integrity. The B7-H3 biomarker has been shown to be a potent inhibitor of the human body’s immune response. We used our novel spatial infiltration metric on multiplexed Immunofluorescence images of resected MCC tumors, to compute the extent of B7-H3 expression in endothelial cells. This quantification enables us to now directly study the impact of B7-H3 colocalization on patient outcome, potentially allowing us to design optimal immunotherapy regimens for individual patients with MCC!!
Our work with Dr. Hesham Elhalawani (Dept. of Radiation Oncology, M.D Anderson Cancer Center) on early prediction of Osteo-radio-necrosis (ORN) using radiomics has been accepted to the American Society for Radiation Oncology’s Head and Neck cancer (ASTRO-HNC) symposium in 2018. We show that a functional principal component analysis (FPCA) of radiomic features extracted at multiple time points before and after radiotherapy, can predict for ORN development in patients. We significantly outperform prediction models based on pre-radiotherapy images and delta radiomics (current practice). We envisage our FPCA screening tool can be used by radiologists to optimize radiation plan for patients undergoing radiotherapy!
Back in Houston after a refreshing trip to India, with a small escapade to Thailand in between. Great food, and great times with family!
Our invited book chapter on the advances made in radiology and histopathology analysis in glioblastoma using image processing and machine learning. Great job Michael Lehrer in integrating all our contributions to this chapter! Check it out here.
Our work on showing the spatial distribution of intratumoral T cells in the tumor has a significant effect on survival is now published in Nature Communications! Find the paper here. It was a great experience to combine our ideas on spatial statistics with the cancer biology expertise of Julienne Carstens and Pedro Correa at MD Anderson.