The Next Frontier in Quantitative Co-Clinical Imaging to Advance Functional Precision Oncology

Abstract

Precision oncology seeks to transform cancer treatment into one tailored for individual patients. Genomic alterations have been the dominant drivers of patient-specific therapeutic strategies. While genomic insights can guide treatment planning, genotype-directed therapies do not guarantee response to therapy. Functional confirmation of target expression, target engagement, and downstream biological effects is essential to defining therapeutic endpoints. Translational imaging can play a central role in closing the loop between genome and function to operationalize functional precision medicine (FPM). Quantitative co-clinical imaging provides a powerful framework to link genotype with function by confirming drug delivery, target engagement, pharmacodynamic effects, tumor function and heterogeneity across animal models and patients to advance FPM. Over the past decade, the Co-Clinical Imaging Research Resource Program (CIRP) of the National Cancer Institute (NCI) has made considerable progress in advancing quantitative imaging biomarker development. However, critical gaps remain in the integration and standardization of co-clinical imaging biomarkers. The next frontier requires moving beyond quantitative endpoints toward predictive oncology. This includes validating imaging-aware models, developing quantitative imaging strategies to support new precision medicine trials, and embedding co-clinical data into computational and in silico frameworks such as radiomics, cross-species mathematical modeling, digital twins, and virtual imaging trials. Equally important is the development of a domain-specific repository that vertically integrates imaging with multi-modal data to enable reproducible, artificial intelligence-ready pipelines. These advances will accelerate the translation of co-clinical imaging biomarkers into decision models that complement genomics, improve patient stratification, and guide selection and personalization of therapies—advancing the promise of FPM.

Introduction

Precision oncology seeks to transform cancer research and treatment, moving beyond the traditional one-size-fits-all paradigm toward individualized care guided by tumor biology, molecular features, and patient context (1). Current approaches to precision oncology have been dominated by static genomic aberrations in initiatives led by the National Cancer Institute (NCI). The NCI MATCH (Molecular Analysis for Therapy CHoice) (2,3) and Pediatric MATCH (4) trials demonstrated the feasibility of large-scale genomic screening but also underscored the limited efficacy of single-agent therapies targeting single mutations, with resistance emerging through multiple mechanisms. To address these shortcomings, the NCI recently launched three new Precision Medicine Trials (PMTs) (5): ComboMATCH, which is designed to test specific molecularly-targeted combinations aimed at overcoming primary and adaptive resistance pathways; ImmunoMATCH, which enhances immunotherapy efficacy through biomarker-driven stratification rather than the conventional all-comers approach; and MyeloMATCH, which guides patients with myeloid malignancies through sequential, stage-specific trials with integrated molecular profiling.

The MATCH trials, and other documented clinical trials, highlight that the presence of mutation(s) does not guarantee therapeutic success. Functional confirmation of target expression, engagement, and biological effects is essential in defining therapeutic endpoints. Letai (6) suggested a functional precision medicine (FPM) paradigm with a focus on tumor activity and function to address limitations in using static genomic approaches as drivers of precision oncology (6–9). Operationally, FPM entails exposing patient-derived tumors or tissues ex vivo to one or more therapeutic strategies and using a measured biological response (e.g., viability, tumor growth rate, tumor killing) to identify actionable vulnerabilities that can guide treatment. Translational imaging technologies can overcome key limitations of genomically-targeted therapeutic studies by operationalizing FPM in vivo, linking genotype and function. Translational imaging can enable in vivo detection of tumors as well as in vivo assessment of tumor function and heterogeneity to confirm target activity and engagement, capture drug pharmacodynamic effects, monitor response to therapy, and characterize the tumor microenvironment (TME) (Table 1)—functions not achievable with genomics alone.

Table 1.

Key parameters and attributes/capabilities of modalities currently supported within the CIRP network

Modality	Sensitivity Range	Resolution Range	Attributes/capabilities
PET	pM–nM	Preclinical* 0.6–2 mm Clinical* 3.5–6 mm	Molecular and functional imaging; pharmacokinetics/pharmacodynamics; blood flow/perfusion; receptor density/occupancy, metabolism, tumor texture, whole-body imaging; treatment planning (evaluation of chemotherapy, targeted therapy, immunotherapy; and radiopharmaceuticals)
SPECT	High pM to mid nM)	Preclinical* 0.3-1mm typical for pin-hole collimator Clinical* 7-10 mm typical using low energy high resolution collimator	Molecular and functional imaging; pharmacokinetics/pharmacodynamics; cardiac blood flow/perfusion; cardiac function and ejection fraction; lung perfusion; bone imaging (fracture, infection, metastasis); receptor density/occupancy; treatment planning; quantitate uptake of radiopharmaceutical therapies.
CT	Generally, requires mM concentrations of x-ray contrast agents (iodine/barium); preclinical nanoparticles can boost effective local attenuation, but the detection is still mM.	Preclinical 10μm–0.15 mm Clinical 0.5–1 mm	Skeletal/lung imaging (clinical), tumor heterogeneity (shape/texture), vascular imaging, quantitative perfusion, spectral/dual-energy and photon-counting applications for material decomposition, radiotherapy planning and evaluation of theranostic agents
MRI	μM-mM depending on contrast mechanism	Preclinical 0.05–0.1 mm Clinical 0.5–1 mm	Array of contrast mechanisms to image soft tissue and bone marrow, tumor heterogeneity (texture/shape), perfusion, hypoxia, diffusion/cellularity, fat content
MRS	μM–mM	Preclinical 2–6 mm Clinical 4–8 mm	Proton or Deuterium metabolic imaging Hyperpolarized Carbon-13 metabolic, perfusion and pH imaging

PET, positron emission tomography; CT, computed tomography; MRI, magnetic resonance imaging; MRS, magnetic resonance spectroscopy;

^*Dependent on image reconstruction and isotope used

Over the past decade, the NCI’s Co-Clinical Imaging Research Resource Program (CIRP) (Table 2) has played a pivotal role in advancing translational cancer imaging through a coordinated multi-institutional effort. A major effort of the network to date has been to develop and harmonize imaging strategies in preclinical and clinical settings across numerous cancers, and to employ them to detect and predict response to therapy (10), similar to the efforts of the Quantitative Imaging Network (QIN) in standardization of human imaging studies and analysis (11). CIRP sites disseminated resources, including protocols, software, standard operating procedures (SOPs), and image datasets through web-resources (see Table 2). These efforts facilitate broader adoption of co-clinical imaging, enhance collaboration, promote reproducibility, and support the NCI’s Precision Medicine Initiative (PMI) (12). Despite these advances, critical gaps remain in the integration and standardization of co-clinical imaging biomarkers, especially considering NCI’s new PMTs (5). The objective of this perspective is to highlight unmet needs in co-clinical imaging in the context of NCI’s new PMTs and to highlight new opportunities for co-clinical imaging at the interface of predictive modeling, digital twins, and vertically integrated federated platforms to advance functional precision oncology.

Table 2.

CIRP Network

Site/trial information	Modalities and linked biology data	Co-Clinical Therapeutic Design	Endpoints	Resources and website
CIRP Site: Molecular and Imaging Response Analysis of Co-Clinical Trials (MIRACCL) (Baylor College of Medicine/University of Texas at Austin/Stanford University) Clinical Trial: Correlation of Clinical Response to Pathologic Response in Patients with Early Breast Cancer (RESPONSE) (NCT05020860) Status: Recruiting	MRI WES RNAseq Mass spectrometry proteomics Pathology	Preclinical: PDX treated with combination carboplatin + paclitaxel imaged and biopsied prior to treatment, after the first cycle of treatment, and end-of-treatment. Tumor samples were characterized by WES, RNAseq, and proteomics. Clinical trial: Patients treated with combination carboplatin + paclitaxel imaged and biopsied prior to treatment, after the first cycle of treatment, and end-of-treatment. Biopsies were then characterized by WES, RNAseq, and proteomics. # of subjects: 30 estimated (none imaged) # of models: 9 PDX	Primary (Clinical): Change in Clinical Tumor Measurement vs. Pathologic Response Secondary (Clinical): Pathologic Complete Response Rate; Predictive value of clinical response following 1 cycle of chemotherapy to predict pathologic complete response. Preclinical: Integration of omics and quantitative imaging data in co-clinical trials to predict treatment response in triple negative breast cancer.	Imaging and treatment protocols, software (MIRACCL) https://miraccl.research.bcm.edu/miraccl/
CIRP Site: The Duke Preclinical Research Resources for Quantitative Imaging Biomarkers (Duke University) Clinical Trial: SU2C-SARC032: A randomized trial of neoadjuvant RT and surgery with or without pembrolizumab for soft tissue sarcoma (NCT03092323) Status: Completed	Preclinical: MRI (primary tumor), gated micro-CT (lung metastases), pathology, WES. Clinical: MRI (primary tumor), chest CT (lung metastases), pathology, WES, RNAseq.	Preclinical: Immunocompetent GEMM (p53/MCA sarcoma) with spontaneous lung metastases; randomized to isotype control, anti–PD-1, RT, or RT+anti–PD-1; imaging pre- and post-treatment to assess local response/recurrence and metastatic burden. Clinical: Phase II SU2C-SARC032 served as the clinical framework for the co-clinical design; imaging in both arms used MRI and chest CT to support aligned assessments of local response/recurrence and metastatic burden. # of subjects: 127 patients # of models: 1 GEMM (p53/MCA sarcoma)	Primary (clinical): Disease-free survival (DFS). Primary (preclinical): Imaging-based local response/recurrence and lung metastasis burden (e.g., lesion number/volume). Secondary: Metastatic progression, imaging biomarker performance for early response prediction, and concordance between preclinical and clinical imaging endpoints.	Imaging protocols, analysis software, phantoms, imaging datasets, CIRP-shared resources. https://sites.duke.edu/pcqiba
CIRP Site: Co-Clinical Research Resource for Imaging Tumor Associated Macrophages (Stanford University) Clinical Trial: Testing the Combination of Two Immunotherapy Drugs (Magrolimab and Dinutuximab) in patients with relapsed or refractory neuroblastoma or relapsed osteosarcoma (NCT04751383) Status: Completed	FDG PET/MR and ferumoxytol-enhanced MRI and MegaPro nanoparticle enhanced MRI	Human osteosarcoma xenografts and murine osteosarcomas in mouse models. Co-clinical studies in patients with neuroblastoma and osteosarcoma # of subjects: 12 # of models: 6 mouse models of osteosarcomas	Primary endpoint: Phase I - Percentage of evaluable patients experiencing grade 3 or higher adverse events (AEs). Secondary endpoint: Assess the utility of ferumoxytol-MRI and FDG-PET/MR imaging in predicting response to therapy.	Imaging protocols, software, PET and MR imaging datasets https://radweb.su.domains/cirp/
CIRP Site: Co-Clinical Quantitative Imaging of Small Cell Neuroendocrine Prostate Cancer (SCNC) Using Hyperpolarized 13C MRI (University of California San Francisco) Clinical Trial: Magnetic Resonance (MR) Imaging with Hyperpolarized 13C Pyruvate as a Diagnostic and Response Monitoring Imaging Tool in Metastatic Prostate Cancer (NCT04346225) Status: Completed	Hyperpolarized (HP) 13C MRI/anatomic, perfusion and diffusion MRI Pathology IHC RNAseq	Metastatic prostate cancer PDX subtypes matched to patient population; Co-clinical study design includes pretherapy and on-treatment HP 13C MRI/MRI to predict/assess response to carboplatin treatment. # of subjects: 24 patients with SCNC metastases to the bone (N=13) and liver (N=11) # of models: 6 SCNC PDX models (LuCap93, LTL352 and LTL610) implanted in bone (N=3) and liver (N=3)	Primary endpoint: Determine whether the metabolic conversion of HP 13C pyruvate to 13C lactate measured by HP 13C MRI can accurately detect liver and bone prostate cancer metastases and provide a measure of response to carboplatin therapy. Secondary endpoint: Determine whether the addition of HP 13C MRI combined with standard of care 1H MRI can improve the assessment of therapeutic response of metastatic prostate cancer to carboplatin therapy.	Optimized liver and bone metastatic PDX models. Imaging protocols, software, phantoms, imaging data sets https://coclinicalimaging.ucsf.edu/
CIRP Site: University of Michigan Quantitative Co-Clinical Imaging Research Resource (University of Michigan) Clinical Trial: Quantitative MRI for myelofibrosis (NCT01973881) Status: Recruiting	MRI including water diffusion, magnetization transfer, fat content and R2* Genomics Pathology	Myelofibrosis (MF) JAK2V617F and MPLW515L subtype transplant models matched to patient mutational allele population; Co-clinical study design includes pretherapy and on-treatment MRI scans during JAK2 inhibitor treatment to predict/assess response to therapy using bone marrow renormalization as a quantitative imaging metric. # of subjects: 60 # of models: 2 transplant models corresponding to MF subtypes	Primary endpoint: Determine whether standard-of-care treatment using a JAK2 inhibitor will increase the fat content of the bone marrow and alter diffusion values as a readout of treatment response or resistance to treatment. Secondary endpoint: MRI-derived spleen volume change during treatment to assess this imaging metric for detection of treatment response.	Imaging protocols, SOP’s software, phantoms, imaging datasets https://umu24cirp.med.umich.edu/
CIRP Site: Penn Quantitative MRI Resource for Pancreatic Cancer (University of Pennsylvania) Clinical Trial: A Randomized Pilot Study of Perioperative Nivolumab and Paricalcitol to Target the Microenvironment in Resectable Pancreatic Cancer (NCT03519308). Additional surgically resected PDAC samples were obtained from biobank (IRB 853345) Status: Completed.	Diffusion-weighted MRI, Dynamic contrast enhanced -MRI, IHC for cell density in the tumor, microvascular density, matrix collagen and fibroblast activation protein for both human and murine PDAC. Single cell RNA sequencing of murine PDAC.	Co-clinical trial design using a GEMM model spontaneous PDAC (KPC = KrasG12D:Trp53R172H:Pdx1-Cre) with matched treatment to clinical trial and additional treatment and control groups for mechanistic studies. MRI protocols and IHC analyses are concordant between clinical and preclinical trial studies with additional single cell RNA sequencing conducted in murine PDAC. # of patients: 3 (from 9 evaluated) # of surgical PDAC specimens from biobank: 20 # of models: 1 (KPC mice)	Primary endpoint: Determine the feasibility and safety of weekly infusion of synthetic vitamin D with chemotherapy and immune therapy. Secondary endpoints: Utility of quantitative MRI markers for evaluating PDAC stromal changes. IHC (both human and mouse PDAC) and single cell RNA sequencing (mouse PDAC only) for mechanistic study.	Respiratory motion-robust diffusion-weighted MRI pulse sequence programs for murine abdomen tumors, dynamic contrast enhanced MRI protocols and analysis software for DW- and DCE-MRI. CIRP-shared resources: https://pennpancreaticcancerimagingresource.github.io/
CIRP Site: PET Imaging Resource to Enhance Delivery of Individualized Cancer Therapeutics (PREDICT) (University of Texas, MD Anderson Cancer Center) Clinical Trial: Novel PET/CT Imaging Biomarkers of CB-839 in Combination with Panitumumab and Irinotecan in Patients with Metastatic and Refractory RAS Wildtype Colorectal Cancer (NCT03263429) Status: Completed	18F-FSPG PET/CT (Primary), FDG-PET/CT (subset), Tumor glutamine metabolism gene signatures, Plasma exosomal biomarkers, CB-839 Pharmacokinetics (PK)	RAS wild-type metastatic colorectal cancer models matched to the clinical study population. Co-clinical studies incorporated baseline and early on-treatment 18F-FSPG PET/CT to assess tumor glutamine metabolism and predict response to combined glutaminase and EGFR inhibition. Imaging findings were correlated with tumor molecular features and plasma exosomal biomarkers. # of subjects: 29 patients # of models: 6 CRC patient-derived xenograft (PDX) models	Primary endpoint: Determine the safety, tolerability, and recommended phase II dose of CB-839 in combination with panitumumab and irinotecan, and assess the objective response rate of the combination in patients with RAS wild-type metastatic colorectal cancer. Secondary endpoints: Assess disease control rate, progression-free survival, and overall survival, and evaluate the utility of 18F-FSPG PET/CT as an imaging biomarker for early prediction of therapeutic response. Correlate imaging findings with tumor glutamine metabolism and plasma exosomal biomarkers.	Imaging protocols, phantoms, CIRP-associated translational imaging resources https://www.mdanderson.org/research/departments-labs-institutes/programs-centers/predict.html
CIRP Site: A Quantitative PET/CT Research Resource for Co-Clinical Imaging of Lung Cancer Therapies (University of Washington / Fred Hutchinson Cancer Center) Clinical Trial: SX-682 with Pembrolizumab for the Treatment of Metastatic or Recurrent Stage IIIC or IV Non-Small Cell Lung Cancer (NCT05570825) Status: Recruiting	PET/CT 18F-FDG for both patients and GEMM models to image tumor metabolism before and after immune checkpoint inhibitor therapies	The UW-FHCC Research Resource for Co-Clinical Imaging of Lung Cancer Therapies (Co-CIRP); SX-682 With Pembrolizumab for the Treatment of Metastatic or Recurrent Stage IIIC or IV Non-Small Cell Lung Cancer # of subjects: 13 # of models: 2 Adenocarcinoma GEMM and 1 Squamous cell carcinoma GEMM model	Develop predictive rules using quantitative pre-clinical PET imaging for immune checkpoint inhibitor therapies in non-small cell lung cancer. This is leveraging an existing co-clinical trial using genetically-engineered mouse model of lung adenocarcinoma.	Imaging protocols, software, phantoms, imaging datasets. Preclinical images, annotations and data are available in the IDC. In addition, there is an IDC Claude skill (120), which enables asking questions about the collection. https://sites.uw.edu/cocirp/
Site: Washington University Co-Clinical Imaging Research Resource (Washington University in St. Louis) Title: Genoproteomic discovery of response to therapy in breast cancer in TNBC (NCT02124902) Status: Completed	FDG-PET/MR (TNBC) Pathology	TNBC PDX subtypes matched to patient population; Co-clinical study design includes pretherapy and on-treatment FDG-PET/MR imaging to predict/assess response to therapy # of subjects: 35 # of models: 6 PDX models corresponding to TNBC subtypes	Primary endpoint: Determine whether neoadjuvant docetaxel and carboplatin will increase the pCR rate in TNBC. Secondary endpoint: Assess the utility of multi-parametric FDG-PET/MR imaging in predicting response to therapy after one cycle of therapy. Preclinical: Assess the utility of multi-parametric FDG-PET/MR imaging in predicting response to therapy after one cycle of therapy.	Imaging protocols, software, phantoms, imaging datasets. https://c2ir2.wustl.edu/
Site: Washington University Co-Clinical Imaging Research Resource (Washington University in St. Louis) Title: Functional Imaging in Prediction of Response to Abemaciclib for Advanced Hormone Receptor-Positive, HER2-Negative Breast Cancer (NCT06179303) Status: Recruiting	FFNP-PET/CT FES-PET (preclinical only) RNAseq ctDNA Pathology	Co-clinical study design includes pretherapy and on-treatment FFNP-PET imaging to predict/assess response to therapy. ER+ PDX with varying mutations to predict response to endocrine therapy alone and in combination with targeted therapy. # of subjects: (recruiting) # of models: ER+ PDX with varying mutation load (ongoing)	Primary endpoint: Assess the accuracy (sensitivity, specificity, PPV, NPV) of ΔFFNP-PET in predicting response (progression-free at 24 weeks) to Abemaciclib + endocrine therapy. Secondary endpoint: To assess circulating dynamic markers (ctDNA) of response and acquired resistance mechanisms; To determine breast cancer intrinsic subtypes with functional vs. nonfunctional ER by ΔFFNP-PET Preclinical: Assess the utility of FES-PET and ΔFFNP-PET in predicting response to therapy alone or in combination with genomic and pathologic biomarkers.	Imaging protocols, software, phantoms, imaging datasets. https://c2ir2.wustl.edu/

Next Generation Quantitative Co-Clinical Imaging Research

At its core, co-clinical imaging research facilitates the development and translation of imaging biomarkers to advance precision oncology across phases of therapeutic clinical trials (10). In the preclinical phase, imaging strategies are developed and validated in co-clinical animal models and used to assess the efficacy of therapeutic interventions. Co-clinical animal models include cell line-derived xenografts (CDX), patient-derived xenografts (PDX), xenografts of patient-derived organoids (PDO), genetically engineered murine models (GEMMs), and chemically-induced murine tumors (13–15). The development and validation of imaging biomarkers has been the subject of numerous publications for PET, MR, and CT (16–24).

Beyond preclinical validation of an imaging biomarker, advancing toward clinical use requires standardization to establish repeatability within a site and reproducibility across sites (25). A critical component of standardization is harmonization of acquisition protocols, reconstruction methods, analysis, and reporting. Once standardization is established, imaging biomarkers are tested prospectively in early-phase clinical trials using standardized analysis pipelines for a predefined context of use, such as patient stratification, target engagement, dose optimization, or early response assessment. Formal Quantitative Imaging Biomarkers Alliance (QIBA)-type profiles, which codify the technical performance claims into a citable, reusable standard (26,27), are typically pursued longer-term once sufficient multi-site evidence has accumulated to support community-wide adoption. A detailed overview of the imaging biomarker qualification pathway is provided by Buckler et al. (28).

Exogenously administered imaging agents or contrast agents can seek approval through the FDA’s New Drug Application (NDA) pathway. More broadly, imaging biomarkers enable patient stratification, optimization of image-guided therapeutic strategies including targeted radiopharmaceutical therapy (TRT), and assessment or prediction of therapeutic response in multi-site settings to support drug development (Figure 1) (11,23,29–32). The use of imaging biomarkers in drug development and in clinical research/practice, whether derived from endogenous or exogenous contrast, is generally consensus-based and evidence-graded rather than dependent on formal FDA biomarker qualification. When imaging is integrated into both arms of a co-clinical trial, this bidirectional design allows insights from the therapeutic clinical trial to further refine the imaging biomarker itself (Figure 1). Table 3 highlights successful examples of co-clinical imaging trials, not necessarily the full distribution of published outcomes, within and outside the CIRP network to predict response to therapy, investigate tumor biology and heterogeneity, and develop biomarkers that reflect human disease heterogeneity.

Figure 1. Co-clinical imaging trials to support functional precision oncology.

The interaction points of preclinical imaging trials with therapeutic clinical trials. In the preclinical phase of drug discovery, co-clinical animal models are used in research, development, and validation of imaging biomarkers to assess therapeutic efficacy using co-clinical animal models subtype-matched to the target patient population. In subsequent phases of drug discovery, the optimized preclinical imaging biomarkers are implemented in a corresponding therapeutic clinical trial integrated with imaging and other modalities to stratify patients and to predict/assess response to therapy. Bi-directional information exchange between the therapeutic clinical trial and the preclinical therapeutic imaging trial could include: Clinical to preclinical—molecular target and drug mechanism of action, pharmacokinetics data, patient populations, clinical endpoints, safety concerns, efficacy outcomes, treatment failures and adverse events, dose-response relationships, treatment schedules; Preclinical to clinical—evidence of target engagement, biomarker status, early response prediction, stratification data based on expanded genomics profiling, optimal dosing information, disease heterogeneity insights, surrogate endpoints, pharmacodynamic measures, and resistance mechanisms. The bi-directional translation in co-clinical imaging trials affords the opportunity to further refine imaging biomarkers based on insights from the therapeutic clinical trial. See references (11,23,29–32) for proposed imaging endpoints that could be incorporated into trial designs and how regulatory and logistical barriers might be addressed.

Table 3.

Representative co-clinical imaging trials demonstrating successful translation (note: does not represent the full distribution of published outcomes)

Brief description	Cancer	Co-Clinical Model	Imaging Modality	Refs.
[18F]FTT, a PET imaging agent targeting PARP-1, was evaluated in ovarian cancer using both preclinical models and a phase I clinical trial to assess PARP-1 expression. [18F]FTT was subsequently used to predict response to therapy in a pilot study.	Ovarian cancer	PDX	PET/CT	(121,122)
Co-clinical FDG-PET radiomic signature in predicting response to neoadjuvant chemotherapy in triple-negative breast cancer	Breast cancer	PDX	PET/CT	(88)
This co-clinical study tested trastuzumab deruxtecan (T-DXd) in HER2-expressing uterine carcinosarcoma PDX models derived from participants in the STATICE trial, showing tumor shrinkage correlating with clinical outcomes.	Uterine carcinosarcoma	PDX	CT	(123)
The study introduced a CD70-targeted PET/CT imaging strategy for clear cell renal cell carcinoma using novel radiotracers in PDX models and validated the approach in a pilot clinical trial.	Clear cell renal cell carcinoma	PDX	PET/CT	(124)
Evaluation of CD46 as a novel therapeutic target in prostate cancer. An ImmunoPET imaging agent was developed and validated using PDX; a clinical trial is underway (NCT05245006). A companion radiotheranostic agent targeting CD46 with Ac-225 in disseminated prostate cancer is in development.	Metastatic Castration-Resistant Prostate Cancer	PDX	PET/CT	(125,126)
Development of quantitative MRI parameters for assessing bone marrow composition and fibrosis in individuals with myelofibrosis (MF), as a noninvasive alternative to biopsy. The quantitative parameters were implemented in a prospective clinical trial (NCT01973881) and development of therapeutic strategies.	Myelofibrosis	Cell transplant models (CTM)	MRI	(52,53,127)
PET imaging with (4S)-4-(3-[18F]fluoropropyl)-L-glutamic acid 18FSPG in treatment-sensitive and treatment-resistant WT KRAS CRC patient-derived xenografts (PDXs) and subsequent Phase I/II clinical study	Colorectal cancer	PDX	PET	(128,129)
A randomized phase II study of neoadjuvant radiotherapy ± anti–PD-1 in extremity sarcomas. In the GEMM arm, mice with carcinogen-induced p53-mutant sarcomas were treated with the same regimens and imaged by micro-MRI/micro-CT to track tumor and metastasis response.	Sarcoma	GEMMs	CT/MRI	(130,131)
A comparative study of preclinical and clinical molecular imaging response to EGFR inhibition using osimertinib in glioblastoma.	Glioblastoma	PDOX	PET/MR	(132)
Co-clinical trial to determine if the MEK inhibitor selumetinib (AZD6244) increases the efficacy of docetaxel. PET/CT imaging was used to identify biological markers in mice and patients that provide a rationale for the differential efficacy of these therapies in the different genotypes.	Lung cancer	GEMMs	PET/CT	(133)

Numerous studies have highlighted the intratumoral heterogeneity of human cancers, demonstrated by multi-region biopsies from a single tumor, including in breast cancer (33,34), lung cancer (35), renal cell carcinoma (36), and glioblastoma (37). NCI’s Human Tumor Atlas Network (HTAN) has been tasked with mapping the heterogeneity of human cancers across phases. Integration of spatial biology information across the co-clinical paradigm can be transformative. PDX recapitulate the heterogeneity of human tumors (38–41), with PDX being limited in their capacity to recapitulate the immune component in the TME. Availability of humanized PDX, however, can partly address this limitation. Given limitations in tissue sampling and study design in clinical trials, tumor tissues derived from co-clinical trials can be used to integrate imaging, multi-omics, pathology, and multi-scale spatial biology to better characterize differences between imaging biomarkers across species, identify spatial characteristics in human tumors that can be interrogated in mouse tumors, and identify and validate imaging biomarkers predictive of response to therapy. Advancing these applications in the next generation quantitative co-clinical imaging research will require parallel progress in developing and validating quantitative imaging strategies, animal models, informatics, and integration of computational oncology and artificial intelligence (AI).

Unmet Needs to Advance NCI’s Precision Oncology Initiative

Over the past decade, the CIRP network has delivered the critical framework that makes quantitative co-clinical imaging possible: multi-site phantom harmonization studies (42), vendor-neutral pulse sequences (43,44), shared code (45,46), and the first FAIR (findable, accessible, interoperable, reusable) metadata schema (CIMI) that links mouse and human images (47). The next generation of PMTs, however, will demand more than harmonization and will require imaging strategies that address trial-specific challenges such as prediction of response to combination therapy, immune stratification, and longitudinal disease monitoring.

In the ComboMATCH trials, the number of potential drug combinations is exponentially greater, and the number and complexity of targets is substantially enhanced (5). While genomic information is to be used to match therapeutic drug combinations, demonstration of combinatorial effect and regression or sustained stabilization is required in at least two relevant in vivo models to support the drug combination. Importantly, genomic data will not provide guidance on optimal dosing strategies. Imaging approaches that predict therapeutic response—alone or integrated with other modalities (ctDNA, multi-omics, pathology)—are thus a critical unmet need. Optimizing imaging signatures in co-clinical animal models in the setting of drug combinations and integrating them with other modalities can facilitate the development of co-clinical decision support signatures to guide therapies. A recent article by Jane et al., Imageable Genome (48), may provide a framework to identify biomarkers in the development of PET imaging strategies to support ComboMATCH trials.

In the context of ImmunoMATCH trials, most clinical studies of immunotherapy combinations are conducted in “all-comers” due to lack of patient selection markers, and the results are often negative or noninformative. The potential for imaging to guide immunotherapy trials was recently highlighted (49). ImmunoPET approaches are ideally suited to facilitate this paradigm. Indeed, existing ImmunoPET imaging agents (50) and others to-be-developed can support ImmunoMATCH clinical trials to both stratify patients and predict response to therapy (51). Co-clinical imaging can play a central role in identifying, developing, and validating clinical-grade imaging biomarkers for biological stratification of patients to propel ImmunoMATCH trials.

MyeloMATCH is the first NCI PMI trial to treat patients with myeloid cancers across the course of disease. Novel PET, MR, or CT imaging strategies are needed to support detection of myeloid cancer (e.g., AML) and prediction of therapeutic response. As an example of such an approach, members of the CIRP network have developed a quantitative MRI methodology to assess the impact of treatment on diseased bone marrow in both patients and a model of myelofibrosis (52–54), which could be translated to other myeloid cancers.

Beyond diagnosis and response assessment, co-clinical imaging can directly impact and accelerate the growing field of theranostics which combines targeted diagnostic imaging with a therapeutic regimen (for example, radiopharmaceutical therapy, RPT), enabling both visualization and treatment of cancer using the same molecular target (55). Image-guided interventions, including radiotherapy planning, invasive procedures, and delivery of targeted radiopharmaceuticals, depend on quantitative imaging biomarkers to define targets and to monitor the efficacy of treatments and associated toxicities (i.e., kidneys, bone marrow). Co-clinical platforms provide a unique setting to prototype theranostic strategies, where imaging agents are paired with therapeutic isotopes or drug conjugates to enable both selection and treatment of responsive subgroups. Embedding these approaches within computational frameworks such as digital twins or virtual imaging trials (VITs, described below) could accelerate the translation of theranostics by simulating delivery, dosimetry, and off-target effects before first-in-human trials.

Co-Clinical Animal Models and New Approach Methodologies (NAMs)

The selection of appropriate co-clinical animal models for therapeutic, including theranostic assessment, typically focuses on replicating the genetic or molecular subtype of the targeted cancer (56) and site of tumor (orthotopic or metastatic) (57). As imaging features predictive of response to therapy are dependent on the TME with documented differences in response to therapy between the primary and metastatic sites (58), next-generation co-clinical trials will require models that more faithfully recapitulate the TME of both the primary and the metastatic sites with the metastasis preferably generated in a disseminated manner following the mechanical aspects of the seed (cancer cells) seeking the optimal soil (conducive microenvironment) as seen in patients. The use of patient-derived orthotopic xenograft mouse models is therefore highly desired.

An intact immune system in co-clinical animal models is central to recapitulating the TME. NCI’s ImmunoMATCH initiative calls for clinical-grade biomarkers which characterize elements of patients adaptive immunity for biological stratification. Co-clinical immunotherapy imaging studies, however, are hampered by the paucity of appropriate syngeneic murine models and the expense and complexity of humanized models with intact immune systems, highlighting an unmet need to significantly expand the availability of such models. While some advances have been made in generating allogenic humanized PDX mouse models using cord blood CD34+ cells, new next generation autologous humanized PDX models, either from self-bone marrow cells or iPS-induced hematopoietic stem cells, should better replicate human immunity in a personalized status. Similarly, MyeloMATCH will require additional models to support longitudinal assessment of AML and myelodysplastic syndromes. Uniquely, these trials will enroll patients from the time of initial diagnosis and throughout the course of disease. The public repository of leukemia and lymphoma PDX will facilitate preclinical studies (59), however, there remains a need for animal models that span the developmental stages of AML and MDS.

The new ComboMATCH PMT will pose new complexities for the selection of animal models. ComboMATCH aims to increase therapeutic efficacy by developing genomically targeted combination therapies (60). The selection of models is likely to proceed the same way as ComboMATCH matches treatments to patients and the trial should be conducted under conditions as close to human-equivalent as possible. Thus, the therapeutic outcome is unknown, and expansive studies will be needed to optimize dosing strategies for combination therapies. Furthermore, therapeutics for co-clinical imaging studies, including targeted protein degraders, variant-specific inhibitors, and antibody-based or cellular immunotherapies (61), will expand and present new challenges for administration and delivery in animals. A critical unmet need is testing and centrally disseminating results from combination therapies to support the development and validation of quantitative co-clinical imaging strategies to predict response to combination therapies. The NCI’s Patient-Derived Models Repository (PDMR) and Patient-Derived Xenograft Development and Trials Center Research Network (PDXNet) can play a central role in this effort.

The Pediatric MATCH trial includes 13 subprotocols that enrolled pediatric patients with advanced solid tumors that have either recurred or failed to respond to standard therapy (4). While PMTs in children are currently feasible, drug development remains a slow process. Importantly, immune checkpoint inhibitors, specifically PD-1 inhibitors, have demonstrated minimal activity across most pediatric cancers, with rare exceptions noted only in select lymphomas and a small subset of hypermutant brain tumors (62–64). The relative rarity of pediatric cancers precludes extensive Phase 1 trials, common in adult oncology, highlighting the need for robust preclinical models for validation. Areas of significant challenge include assessing response to therapy in bone-forming metastases (for osteosarcoma (65)) or marrow-based recurrence (for neuroblastoma (66)). For both diseases standard anatomical imaging is inadequate and there is an unmet need for translational functional imaging strategies that can be developed in the context of co-clinical trials. The uncommon nature of pediatric cancer further underscores the need for a multi-institutional network to gather enough patients for feasibility studies. To address these gaps, the Childhood Cancer Data Initiative (CCDI) provides a centralized catalog of resources, including a portfolio of PDXs and sequencing profiles, intended to assist researchers in identifying promising therapeutic agents and imaging strategies thereof for clinical transition (67).

The National Institutes of Health (NIH) recently announced that it is prioritizing human-focused research with an emphasis on new approach methodologies (NAMs) as complementary approaches to the use of animals in research (68). Co-clinical animal models are, by design, human-focused research since tumors are derived from human tumors. Examples of NAMs include ex vivo models such as organoids, microphysiologic systems and computational models. Of note, the NIH recently established the Standardized Organoid Modeling (SOM) Center which aims to serve as a neutral scientific hub for standardization, developing organoids that are reproducible, reliable, and accessible for medical research (69). Development and validation of functional (PET, SPECT, MRS) imaging strategies are ideally suited for organoids and microphysiologic systems. An example of how clinically relevant imaging studies might be conducted with NAMs is that described by Nasiri et al. where 3D lung tumor-on-a-chip methodology was combined with novel high-resolution radioluminescence microscopy using FDG (70). This model captured expected effects of hypoxia on radiotherapy response and FDG-PET imaging, validating the ability of this NAM to provide a realistic platform for co-clinical imaging studies.

Computational models of NAMs are highly relevant across all imaging modalities to develop imaging biomarkers and identify therapeutic endpoints, and several are exhibiting promising results in clinical applications (see examples (71–73)). At this stage, however, the complexity and dynamics of intact biological systems and response to therapy cannot be fully replicated in organoids or computational models. The incorporation of NAMs into the translational pipeline will require innovative approaches, extensive validation, and thoughtful study design. NAMs will likely extend the traditional translational pipeline from in vitro studies to include NAMs followed by preclinical to clinical translation with NAMs providing an early-stage platform to assess feasibility, thus minimizing the use of animal models. To that end, new NAM technologies that support early-stage validation of imaging biomarkers are warranted. These may include imaging of organoids to aid in development and validation of imaging and theranostic strategies or computational co-clinical models which can aid in harmonization, development, and validation of quantitative imaging biomarkers.

Opportunities in Co-Clinical Computational Imaging and Oncology

Quantitative imaging is highly susceptible to acquisition protocols in both preclinical and clinical settings. Ensuring that metrics such as apparent diffusion coefficient (ADC) or perfusion are comparable between mouse and human requires explicit metrology beyond nominal protocol matching. Human consensus efforts (QIBA DWI/DCE guidance (74); DSC-MRI recommendations (75)) show that protocol and processing variability can alter quantitative interpretation, including in post-therapy settings (76). A practical cross-species approach is to define common physiological targets and analytic models, qualify performance with phantoms/vendor-neutral procedures where feasible, and correct or report resolution/motion/partial-volume effects. Normal-tissue benchmark ROIs and uncertainty reporting provide stability checks before quantitative biomarkers are used as decision endpoints. Future developments in digital twins of imaging instruments (77,78) may provide an in-silico approach to optimizing co-clinical acquisition protocols.

Radiomic and deep learning (DL) features can play a central role in precision oncology by extracting high-dimensional quantitative features from standard medical images, thus enabling noninvasive characterization of tumor biology, predictive of patient stratification or therapy (79–82). These features can reveal patterns linked to prognosis, treatment response, and underlying molecular signatures that are not discernible to the human eye (see examples in Ref. (17,83–91)). Co-clinical imaging can significantly advance the application of radiomics and DL features. In the clinical setting, a key limitation is the availability of tissue as a reference standard, typically needle-core biopsies, to identify and validate imaging features. In contrast, in co-clinical imaging, whole tumors are available as a reference standard for validation of radiomic and DL pipelines, which can provide a platform for discovery and explainability. Importantly, traditional radiomics/DL pipelines are trained in isolation for mice and patients, neglecting abundant longitudinal data that animal studies generate. Thus, a co-clinical imaging framework could instead support shared representation learning across species, in which models are trained to identify common biological phenotypes despite differences in scale, acquisition, or species.

Such a workflow could proceed through five stages: (i) harmonized preprocessing and metadata capture across mouse and human datasets; (ii) self-supervised or contrastive pretraining to learn shared representations; (iii) task-specific fine-tuning for applications such as segmentation or response prediction; (iv) multi-site external validation with calibration and uncertainty assessment; and (v) deployment with monitoring for drift and periodic recalibration as imaging protocols evolve. Because longitudinal mouse studies can provide dense imaging and biologically anchored labels, they may support pretraining strategies that make more efficient use of limited human datasets, after which smaller human cohorts can be used for task-specific fine-tuning. With appropriate external validation, calibration, and monitoring for domain shift, such shared-feature approaches could help imaging biomarkers evolve from exploratory endpoints into more robust stratification tools for identifying likely non-responders.

A critical consideration in deploying radiomic and DL pipelines, however, is that these features are sensitive to acquisition/reconstruction settings (resolution, noise, slice thickness/scan plane) and segmentation variability (46,89,92–94); therefore, co-clinical radiomics should include harmonized preprocessing (common voxel size and discretization), perturbation-based sensitivity analyses, and locked multi-site validation with calibration and uncertainty reporting. Feature sensitivity to tumor volume/noise/resolution has been demonstrated in co-clinical radiomics studies (89). Standardized feature definitions (defined by the Image Biomarker Standardization Initiative (95)) can reduce software-induced variability. Imaging features, such as entropy, should be treated as quantitative surrogates of heterogeneity whose values depend on imaging conditions and discretization, and interpreted through linked biology and normal-tissue benchmarks rather than assumed to map directly to a single tissue mechanism (82).

Virtual imaging trials (VITs) can help characterize imaging features and other task-based quantitative metrics as endpoint biomarkers in clinical trials (96,97). VITs simulate imaging studies in silico, integrating patient anatomy, scanner physics, and image reconstruction; thus, enabling systematic evaluation of acquisition protocols and quantitative pipelines. They rely on anatomically realistic computational phantoms, such as XCAT digital phantoms for human subjects (98,99) and the MOBY digital phantoms for preclinical models (100), coupled with forward models of CT, MRI, or PET acquisition and reconstruction to produce synthetic images with fully known ground truth. Applications of VITs are broad, including preclinical (101), clinical (101–105) and their use in regulatory qualification of imaging-derived endpoints (106). However, co-clinical VITs for harmonizing acquisition protocols and quantitative imaging biomarkers across species—linking preclinical and clinical arms under a shared context of use—remain underexplored and require prospective validation, uncertainty quantification, and multi-site testing to characterize impact on downstream therapy response endpoints.

Digital twins are increasingly being adapted to support precision medicine (107–109). Digital twins represent dynamic, subject-specific computational models that integrate multimodal data to simulate disease trajectory and treatment response. Co-clinical imaging is uniquely positioned to support their development and implementation through a bidirectional pipeline: (i) preclinical imaging informs model priors through mechanistic modeling and validation, and through empirical characterization of treatment response and resistance pathways; and (ii) clinical imaging data updates the digital twin, enabling adaptive therapy simulations and outcome prediction. Imaging-informed mathematical models and early digital twin approaches have been demonstrated in clinical, preclinical, and technical pilot studies (71–73), including examples in which longitudinal imaging is used to calibrate patient-specific models of tumor growth or response (110). However, prospective co-clinical deployment, shared cross-species latent spaces, and co-clinical digital twins for adaptive therapy remain underexplored. The integration of digital twins, as virtual patients/subjects incorporating mechanistic clinical endpoints, with VITs presents a further opportunity for investigation.

It is useful to distinguish between imaging-aware and imaging-validated models in predictive oncology. Imaging-aware models incorporate imaging-derived features, priors, or constraints into computational or biological models of treatment response, whereas imaging-validated models are additionally evaluated against independent imaging endpoints and clinically relevant outcome-linked measures. In practice, benchmarking of such models should include agreement with response distributions across cohorts, concordance with pathology or molecular correlates, fidelity in capturing the timing and magnitude of pharmacodynamic effects, and reproducibility across sites and protocols with uncertainty reporting. Standardization across sites will require pre-specified endpoints, harmonized metadata and provenance, common evaluation metrics, and external multi-site validation.

Clinical relevance.

The clinical relevance of computational technologies hinges on whether the features or computational pipelines can reliably identify a patient who is unlikely to benefit from a therapy early enough to change management. Co-clinical imaging offers a pragmatic path forward by using longitudinal mouse studies as a source of biologically anchored labels for co-clinical representation learning, allowing models to learn shared phenotypes of growth, response, resistance, and phenotypic drift across species despite differences in scale, acquisition, and TME. If preclinical imaging can define priors for treatment response and patient imaging can iteratively update those priors, then radiomics and digital twin frameworks may begin to function less as retrospective biomarker discovery tools and more as prospective instruments for risk stratification, adaptive therapy simulation, and early recognition of non-response. Virtual imaging trials using paired mouse and human computational phantoms may further reduce a persistent source of translational failure by stress-testing acquisition parameters, motion, spatial resolution, dose constraints, and endpoint robustness before a biomarker is carried into clinical trials. For current oncology practice, the message is appropriately cautious but actionable: radiomics will not influence treatment decisions simply because it is high-dimensional or AI-enabled; it will become clinically useful only when embedded in standardized, externally validated, uncertainty-aware co-clinical workflows that can show, prospectively and reproducibly, that an imaging signature changes therapeutic choice, timing, or confidence at the point of care.

Vertical Integration of Imaging and Linked-Biology Data Through Machine Learning (ML)/AI

NCI’s precision medicine initiative emphasizes the use of translational oncology models to elucidate the biological and molecular basis of cancer prevention and treatment. However, many preclinical imaging laboratories are not equipped to capture the associated metadata or support the level of annotation required for these studies. Preclinical imaging workflows are inherently complex and often involve multiple downstream pipelines to validate in vivo imaging measurements. As a result, these workflows generate large, heterogeneous datasets that are difficult to track, manage, and preserve over time, thereby limiting open science and reproducibility. To partly address these challenges, CIRP sites were required to share all data and protocols through dedicated web-based resources (see Table 2). Several sites developed imaging repositories (BiRAT (111); PIXI (112); MIRACCL (113)) to support their CIRP activities; others uploaded datasets (114) to the Imaging Data Commons (IDC) (115).

These efforts exposed persistent challenges, including heterogeneity in file formats, inconsistent metadata collection, and lack of standardized ontologies. They also highlighted problems with interoperability and data integration, which have slowed standards adoption and limited cross-site reuse of data. As oncologic preclinical imaging workflows continue to increase in complexity, scale, and dependence on richly linked biological data, there is a growing need for a federated environment built on a unified platform for preclinical imaging and associated biology experiments. Such an environment would promote interoperability, streamline data integration, and support long-term reuse of complex multimodal datasets. A related and critical unmet need is the curation and quantification of images in ways that enable querying based on quantitative imaging metrics. Equally important is the need for standardized, reproducible analysis workflows, as emphasized in numerous publications (116,117). Containerization within a unified computational environment can support computational reproducibility across sites, facilitate sharing of validated analysis pipelines, and enable scalable image processing across high-performance computing environments.

To address these needs, the open-source Preclinical Imaging XNAT-enabled Informatics (PIXI) platform was developed with support from NCI’s Informatics Technology for Cancer Research (ITCR) program as the first domain-specific repository for preclinical imaging and linked biology experiments (https://www.PIXI.org). A domain-specific preclinical imaging environment such as PIXI provides a unified framework for deploying both data-driven approaches, including machine learning and AI, and mechanism-driven mathematical modeling. By vertically integrating curated images, quantitative imaging features, and linked biology data from multiple sources, such a platform can accelerate discovery of novel therapeutic and predictive strategies. Ultimately, the availability of this type of platform could transform cancer detection, diagnosis, and treatment by accelerating the translation of co-clinical research findings into predictive decision models, consistent with the goals of NCI’s precision medicine initiative.

Long-term success of co-clinical imaging data infrastructures will require clear governance and sustainability with appropriate funding models. Effective governance is needed to define roles, responsibilities, and decision-making processes for data submission, access, stewardship, and sharing across institutions. Closely related are issues relating to data ownership and quality control. Long-term sustainability may be achieved through public-private partnership (PPP). The Alzheimer’s Disease Neuroimaging Initiative (ADNI) is frequently cited as the gold standard for PPP sustainability in medical imaging (118,119). Other options for sustainability include tiered access and subscription models. Interoperability with existing repositories (e.g., The Cancer Imaging Archive, TCIA; IDC; The Cancer Genome Atlas, TCGA), analysis tools, and community standards also remain a critical challenge, as fragmented architectures and nonstandard data models hinder data exchange, integration, and reuse. Addressing these issues will require coordinated policies, standards-based system design, and durable operational support to ensure that co-clinical imaging resources remain usable and broadly adoptable over time.

Clinical relevance.

Vertically integrating preclinical imaging with biological data in a unified environment makes co-clinical research more reproducible, scalable, and translatable to patient care. Harmonizing imaging datasets, quantitative features, molecular profiles, pathology, and treatment response data within vertically integrated platforms enables development and validation of ML/AI and mechanistic models which are particularly valuable for linking tumor biology to imaging phenotype and treatment response to develop predictive biomarkers and therapeutic decision models. Durable governance, quality control, interoperability standards, and sustainable funding are essential for platform viability. These platforms have the potential to transform complex co-clinical imaging data into actionable clinical imaging paradigms for cancer detection and diagnosis, patient stratification, therapy selection, and monitoring response to therapy.

Limitations and Barriers to Translation

Despite its promise, broad adoption of quantitative co-clinical imaging remains constrained by several practical barriers. As highlighted above, the availability of preclinical models to support ImmunoMATCH trials, and scarcity of pediatric animal models, remain major barriers. Advanced imaging workflows are resource-intensive, requiring specialized instrumentation, maintenance, scan time, trained personnel, and substantial computational infrastructure for reconstruction, harmonization, and AI-enabled analysis. Access to state-of-the-art platforms and analytic pipelines also varies considerably across institutions, which may introduce capability gaps and complicate multi-site reproducibility. Regulatory and governance challenges remain important, including data sharing agreements, provenance tracking, and standards needed for qualification of imaging-derived endpoints. In addition, radiomics requires standardization, and AI methods remain vulnerable to overfitting, domain shift, and site-specific confounding introduced by differences in acquisition, reconstruction, and annotation practices, limiting generalizability if not rigorously externally validated. In this setting, simulation frameworks and virtual imaging trials may reduce exploratory burden and improve benchmarking and protocol planning across sites, but they should be viewed as complementary tools rather than substitutes for prospective biological and clinical validation.

Summary and Conclusions

Co-clinical imaging has matured from proof-of-concept studies into a translational framework with the potential to directly impact NCI’s next-generation precision medicine initiatives. The foundation established by CIRP includes harmonized protocols, quantitative pipelines, and informatics resources which have enabled rigorous cross-species comparisons and reproducible imaging biomarkers. Yet, the next frontier requires moving beyond descriptive endpoints toward predictive and mechanistic integration. To advance FPM in oncology over the next five to seven years, we outline five coordinated priorities sequenced by clinical urgency and translational impact (Figure 2):

Figure 2.

Proposed five coordinated priorities sequenced by clinical urgency and translational impact to advance functional precision oncology over the next five to seven years. (Figure images generated by ChatGPT, May 15, 2026, OpenAI, https://chat.openai.com.)

• Develop, validate, and harmonize PMT-specific co-clinical quantitative imaging biomarkers to guide patient stratification, optimize dosing, and predict response where genomics alone is insufficient.

• Expand the availability of PDX and develop next-generation autologous humanized PDX and pediatric-specific models to achieve the translational fidelity required by ImmunoMATCH and Pediatric MATCH trials.

• Develop a co-clinical infrastructure to validate AI models and establish shared representation learning pipelines and imaging-constrained digital twins to transition imaging biomarkers from descriptive endpoints into prospective tools for adaptive therapy in clinical trials.

• Establish a sustainable vertically integrated platform that unifies co-clinical imaging, multi-omics, and pathology under standardized ontologies, reproducible workflows, and durable governance to develop multi-modal biomarkers predictive of clinical outcome.

• Develop and integrate organoids, microphysiologic systems, and computational models into a tiered translational pipeline coupled with co-clinical models to accelerate early-stage biomarker validation and translation of clinical imaging endpoints to enable FPM.

These advances in co-clinical imaging will complement genomic insights to advance functional precision oncology—providing dynamic, noninvasive, and phenotypically-grounded biomarkers that improve patient stratification, predict/assess response to therapy, and guide adaptive therapeutic strategies, accelerating the translation of co-clinical discoveries into meaningful clinical impact.