Memorial Sloan Kettering Cancer Center
Physics and information based modeling, outcome prediction, and optimization of cancer treatment. Integrating dose, imaging, and molecular data to understand cancer and optimize treatments.
Why our work matters
Cancer varies — from patient to patient, tumor to tumor, and cell to cell. The Deasy Lab treats that variability as information: we build models that read tumor biology from routinely acquired images, genomes, and clinical data, and turn it into decisions that make treatment more precise and less toxic. The aim is not modeling for its own sake, but tools that reach the clinic and change how patients are cared for.
Research programs
The through-line across all of them: innovative mathematics used to understand patient and disease variability — and to turn it into a leverage point for cancer care.
Mathematical Oncology Initiative
Breast Cancer Research Foundation · Simons Foundation
Geometry- and topology-based methods to subtype patients, integrate multi-omic networks, and surface therapeutic targets — and to train the next generation who build them.
ECHO
2021 INFORMS Franz Edelman Laureate
A prioritized optimization that turns the hand-tuned trade-off between tumor coverage and organ sparing into a reproducible, mathematically optimal plan — now used to help plan roughly 23% of patients treated at MSK.
Proton & particle radiobiology
A thread since the lab's first NIH award
Mechanistic models of how dose, timing, and tumor biology combine to determine response and toxicity when the physics differs from conventional photons — reproducing clinical observations across photon and carbon-ion therapy.
Digital twins
Multimodality imaging + tumor biology
Combining FDG-PET and MRI with a mechanistic tumor-control model to forecast response and flag likely non-responders early — while treatment can still be adapted.
Tools for cancer research
Open platforms and methods from the lab — for images and dose, for the geometry of omics data, for AI segmentation in the clinic, and for keeping a research program organized.
A lightweight, browser-based tool for managing projects and getting things done. FocusNexus organizes projects and tasks by role — Chair, Science, Mentor, Personal — so the next action in each part of your life is always in view. Runs entirely in the browser, imports and exports its full state as JSON, and is free for personal and non-commercial use.
The lab's open-source platform for relating medical images to radiotherapy dose and outcomes — registration, dose calculation, contouring, and quantitative ("radiomic") image analysis. CERR has underpinned 70+ of the lab's studies and been cited over 800 times; pyCERR is its native Python successor, with a graphical viewer and AI-pipeline utilities. Development is led by Aditya Apte, PhD.
Apte et al., Medical Physics (2018), PMC6597320.
Deep-learning models for segmenting tumors and organs at risk on CT and MRI — heart and cardiac substructures, lung, head-and-neck, prostate, rectum, pancreas, and whole-body models — published with pyCERR and installable with one command via the CERR model installer. Models of this kind are in routine clinical use for radiotherapy planning at MSK.
Ollivier-Ricci Curvature-Omics: an open-source Python package that lays omics data over gene- or
protein-interaction networks and measures the curvature of each interaction — a geometric readout of
how robust or fragile a biological system is, which can surface hidden therapeutic targets and
regulatory modules. Install with pip install orcomics.
A framework for exploring topological representations of high-dimensional biomedical data. NetFlow turns a cohort of samples into an interpretable graph of sample relatedness — built on optimal transport, diffusion, and topological data analysis — to support precision oncology and hypothesis generation, with Jupyter tutorials that reproduce the multiple-myeloma analysis from the paper.
Virchow2-powered Spatial Characterization and feature Extraction. An end-to-end computational pathology pipeline for whole-slide images of lung and breast cancer: it pairs the Virchow2 pathology foundation model with a pixel-wise decoder to segment tissue into tumor, stroma, inflammatory cells (TILs), necrosis and other, then extracts quantitative spatial biomarkers of the tumor microenvironment — tumor–stroma ratio, stromal TIL scores, immune and necrosis proximity, and tumor morphology. Developed by Himangi Srivastava; open source (Apache 2.0), paper under submission.
Featured
Using optimal-transport-based clustering of molecular data, the lab isolated a poor-prognosis, hypoxia-enriched subgroup of breast cancers and characterized its biology — showing that geometry-based methods can define clinically meaningful subtypes that conventional analyses miss.
Oh et al., npj Breast Cancer (2025), doi:10.1038/s41523-025-00725-y.
Recent work
A few highlights — the full, searchable library has every paper with a plain-language summary.
An AI model estimates a head-and-neck tumor's oxygen-starved volume directly from routine FDG-PET — no specialized hypoxia scan required.
Physics and Imaging in Radiation Oncology, 2025A large language model trained on oncology records extracts signals of cancer progression from free-text reports at scale.
npj Digital Medicine, 2025Pooled imaging maps where brain metastases arise, informing precise detection and function-sparing radiotherapy.
Nature Communications, 2025The lab
Led by Joseph O. Deasy, PhD, Chair of Medical Physics at MSK, the lab brings together mathematicians, physicists, and computational scientists.
Join us · Collaborate
We welcome postdoctoral fellows, graduate students, and clinical fellows from mathematics, physics, computer science, engineering, statistics, and medicine — and research collaborations across imaging, genomics, and radiotherapy.
