||5U01CA220401-02 Interpret this number
||Informatics Tools for Quantitative Digital Pathology Profiling and Integrated Prognostic Modeling
Accurate biomarker-driven prognostic stratification, response prediction, and cohort enrichment are critical for
realizing precision treatment strategies and population health management approaches that optimize quality of
life and survival for cancer patients. Genomics holds promise for improving classification and prognostication of
malignancies, yet oncology practice continues to rely heavily on immunohistochemistry (IHC) as a fundamental
tool due to its practicality and ability to provide protein-level and subcellular localization information. The goal
of this proposal is to create an open-source software resource for the quantitative analysis of IHC stained
tissues and effective integration of IHC, genomic, and clinical features for cancer classification and
prognostication. This proposal builds on our collective experience in computer-assisted analysis of microscopic
images (including IHC images), development of machine-learning methods to address the challenges of
classification and prognostication with heterogeneous and high-dimensional data, and leadership in collection
and large-scale analysis of cancer outcomes involving collaboration with multiple medical centers. This effort
for the first time will create tools to integrate quantitative IHC imaging, clinical, and genomic information that
will in turn enable the research community to explore strategies for the classification of malignancies and
prediction of outcomes. The proposed tools will be developed and extensively validated in close collaboration
with clinical, genomic, and digital pathology data from the NCI-supported Lymphoma Epidemiology of
Outcomes (LEO) cohort study. The software tools produced by this proposal will enable the characterization of
subcellular protein expression in cell nuclei, membranes and cytoplasmic compartments. Spatial features of
protein expression heterogeneity, along with patient-level summaries of protein expression will be used to
develop machine-learning classifiers for cancer subtypes, using diffuse large b-cell lymphomas as a driving
application. Technology for automatic tuning of machine learning algorithms will enable a broad class of
clinically and biologically motivated users to utilize these tools in their investigations. We will also provide an
interactive dashboard that enables users to integrate genomic and IHC-based features to explore prognostic
models of patient survival. These tools will be released and documented under an open-source model,
integrated with HistomicsTK (https://histomicstk.readthedocs.io/en/latest/), and available to the broader cancer
A population-based multistate model for diffuse large B-cell lymphoma-specific mortality in older patients.
, Goldstein J.S.
, Ayer T.
, Rai A.
, Flowers C.R.
Cancer, 2019-06-01; 125(11), p. 1837-1847.
Informatics Approaches to Address New Challenges in the Classification of Lymphoid Malignancies.
, Goldstein J.S.
, Jaye D.L.
, Gurcan M.
, Flowers C.R.
, Cooper L.A.D.
JCO clinical cancer informatics, 2018; 2, .