Routine H&E histopathology provides spatial snapshots of tumor ecosystems, but standard computational pathology often treats whole-slide images as texture rather than explicitly linking them to underlying tumor dynamics. The authors present an approach that infers patient-specific parameters of a mechanistic reaction-diffusion model from a single routine H&E biopsy. By converting nuclei locations into point patterns, the method seeks to recover biologically interpretable parameters—notably proliferation and diffusion rates—directly from histology.
Cell-nucleus point patterns extracted from whole-slide images are fitted jointly to two analytic descriptors: the two-point correlation function and the power spectral density. Fitting these descriptors to analytical expressions derived from the reaction-diffusion model yields estimates of patient-specific mechanistic rates. The fitting approach was applied across eleven multicenter cohorts, demonstrating that the method can recover individualized proliferation and diffusion parameters from a single biopsy sample.
From the inferred mechanistic parameters, the authors derive secondary metrics including dispersion length and front velocity, which summarize spatial spread and invasion speed in mechanistic terms. These derived metrics are used to define mechanistic phenotypes that capture aspects of tumor growth and dispersal that are not directly observable from a single static image but are implied by the fitted reaction-diffusion dynamics.
Mechanistic phenotypes stratified both progression-free survival and overall survival in the analyzed cohorts. The authors report that dispersion length and front velocity add prognostic information that is independent of routine covariates commonly available in clinical data. Specific statistical details, model coefficients, and cohort-by-cohort hazard estimates were presented in the source but are not reproduced here beyond what the authors reported at a high level.
The study illustrates application of histology-calibrated mechanistic parameters to conduct counterfactual radiotherapy simulations in glioblastoma. Using inferred growth kinetics, tumors were classified by growth speed, and a virtual clinical trial simulated allocation as follows: deescalating faster-growing tumors to hypofractionation and dose-intensifying slower-growing tumors. In this in silico trial, the histology-guided allocation produced an increase in restricted mean progression-free survival of 96.7 days relative to standard-of-care allocation. The improvement was statistically significant by permutation testing (permutation p < 0.001) when compared with random allocation to the same arms. These results are presented as a proof of concept for model-based counterfactual evaluation of radiotherapy strategies using routine histology.
This work is a secondary analysis of existing, de-identified human data. The whole-slide images and clinical records analyzed include publicly available datasets from the Clinical Proteomic Tumor Analysis Consortium (CPTAC) via The Cancer Imaging Archive (TCIA) and other published cohorts such as UPenn-GBM and the ovarian cohort cited from Boehm et al. De-identified institutional cohorts (melanoma, cervical cancer, esophageal cancer) were analyzed under MD Anderson IRB protocols 2023-0971, 2025-0243 and PA17-0191, which included waivers of informed consent for retrospective use. The authors state no new human data or samples were collected for this study, and data availability is governed by the original dataset releases and institutional agreements.
The authors disclosed that a patent application has been filed covering aspects of the method. Funding sources named in the publication include The University of Texas MD Anderson Cancer Center, Austrian Academy of Sciences, Swiss National Science Foundation, Jayne Koskinas Ted Giovanis Foundation for Health and Policy, NIH, Department of Defense, National Cancer Institute, Leukemia & Lymphoma Society and others.
The reported method provides a pathway to translate routine histopathology into mechanistically interpretable, forward simulations that can generate counterfactual scenarios for treatment strategies, exemplified here by radiotherapy scheduling in glioblastoma. The authors frame this as a low-barrier transition from standard histology to individualized model-based simulations that can complement existing covariates.
Limitations and specifics not fully detailed in this summary include cohort-level numerical results beyond the headline effect in the virtual glioblastoma trial; readers should consult the full preprint for full methods, cohort descriptions, statistical models, and supplementary analyses. The authors emphasize that all datasets used were previously collected and released under appropriate approvals and that no new prospective clinical trial was performed as part of this work.