A Critical Review of PET/CT Assessment After CAR-T Cell Therapy in Large B-Cell Lymphoma, Limitations of Current Criteria, Emerging Biomarkers, and a Framework for Standardised Evaluation

Abstract

Chimeric antigen receptor (CAR) T-cell therapy has fundamentally altered the management of relapsed or refractory Large B-Cell Lymphoma (LBCL). Despite remarkable clinical efficacy, the radiologic evaluation of therapeutic response remains highly problematic. Standard assessment frameworks, notably the Lugano classification and the Deauville Five-Point Scale (D5PS), were calibrated for cytotoxic chemoimmunotherapy and demonstrate profound inadequacies when applied to the unique biological kinetics of cellular immunotherapy. This review critically examines the limitations of standard ¹⁸F-Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography (¹⁸F-FDG PET/CT) in the CAR-T setting through a multidisciplinary lens, incorporating perspectives from hematology, nuclear medicine, neuroradiology, and molecular pathology. Pre-infusion quantitative parameters, including total metabolic tumour volume and spatial dissemination, operate as robust biomarkers for risk stratification and predicting immune-mediated toxicities. Post-infusion, conventional visual grading yields unacceptable false-positive rates. Phenomena including pseudoprogression, delayed complete responses, and systemic inflammation secondary to Cytokine Release Syndrome (CRS) routinely mimic active malignancy. Furthermore, Immune Effector Cell-Associated Neurotoxicity Syndrome (ICANS) necessitates specialized neuroradiological evaluation. To resolve these diagnostic ambiguities, circulating tumour DNA (ctDNA) provides a precise, tumour-specific molecular signal capable of differentiating therapy-induced inflammation from genuine refractory disease. This review supports a structural transition from static visual scoring to a dynamic, multi-modal paradigm. Accordingly, the CAR-T Response Assessment Framework (CART-RAF), is proposed as an algorithmic approach that integrates continuous metabolic kinetics with serial molecular surveillance. Adoption of this multidisciplinary framework may reduce false-positive progression classifications, limit the premature discontinuation of effective cellular therapy, and improve clinical outcomes.

Introduction

The deployment of CD19-directed Chimeric Antigen Receptor (CAR) T-cell therapy represents a critical advancement in the treatment of relapsed or refractory Large B-Cell Lymphoma (LBCL). While these engineered cellular products can induce durable complete remissions in heavily pre-treated cohorts, identifying early therapeutic failure versus transient immune activation remains a persistent clinical challenge.¹

The evaluation of therapeutic response relies almost exclusively on ¹⁸F-Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography (¹⁸F-FDG PET/CT), assessed using the Lugano 2014 criteria and the Deauville Five-Point Scale (D5PS).^2,3 However, these legacy frameworks were designed to capture the linear apoptotic effects of conventional cytotoxic regimens. CAR-T cells, acting as living drugs, exhibit logarithmic in vivo expansion, precipitating intense, localized inflammatory responses and massive systemic cytokine release. Consequently, traditional metabolic imaging frequently mischaracterizes immune-mediated tumour clearance as progressive disease.

The mechanistic basis for this diagnostic failure is rooted in a fundamental mismatch between the biological assumptions of legacy criteria and the pharmacodynamics of cellular immunotherapy. The Lugano classification and D5PS were calibrated for cytotoxic regimens that produce linear, dose-dependent tumour cell apoptosis; residual FDG avidity after chemotherapy overwhelmingly reflects viable malignant cells.³ In stark contrast, CAR-T cells function as self-amplifying living drugs, undergoing logarithmic in vivo expansion at tumour sites. This expansion generates dense infiltrates of highly glycolytic immune effector cells that are metabolically indistinguishable from viable lymphoma on FDG-PET. The D5PS relies on visual comparison of lesion uptake against mediastinal blood pool (scores 2–3) and hepatic parenchyma (score 4) as internal reference standards. However, the massive cytokine release associated with CRS destabilises both reference organs: splenic and marrow FDG uptake increases due to reticuloendothelial activation, while hepatic metabolism fluctuates secondary to acute-phase protein synthesis and hepatocyte stress.⁴ Consequently, the very biological anchors upon which D5PS scoring depends become unreliable during the critical early post-infusion assessment window. Pseudoprogression is mechanistically driven by tumour-infiltrating CAR-T cell expansion, not by clonal tumour proliferation, a distinction that binary visual scoring fundamentally cannot capture.⁵ These limitations are not addressable through simple threshold adjustments to the existing Deauville scale; they require an entirely distinct assessment paradigm that integrates molecular confirmation of disease status.

Critically, the heterogeneity of CAR-T products themselves introduces further variability into imaging assessment. The three FDA-approved CD19-directed products—axicabtagene ciloleucel (axi-cel), tisagenlecleucel (tisa-cel), and lisocabtagene maraleucel (liso-cel)—differ fundamentally in their costimulatory domain architecture, manufacturing processes, and consequent in vivo expansion kinetics. Axi-cel incorporates a CD28 costimulatory domain that drives rapid, high-magnitude T-cell expansion with earlier and higher peak serum cytokine concentrations, resulting in more frequent and severe CRS (grade ≥3 CRS in 7–13%) and correspondingly more intense early post-infusion FDG uptake confounding.⁶ In contrast, tisa-cel and liso-cel employ 4–1BB (CD137) costimulatory domains that promote slower but more sustained T-cell expansion, generally producing lower-grade CRS and a different temporal window for immune-mediated metabolic flares.⁷ These product-specific pharmacokinetic differences mean that a single fixed assessment schedule—such as the standard day-28 PET/CT—may capture peak inflammatory confounding for one product but miss it for another.⁸ Furthermore, the emergence of next-generation constructs incorporating dual costimulatory signals, armoured CAR-T cells secreting checkpoint-blocking antibodies, and allogeneic off-the-shelf products will further diversify the imaging landscape. This pharmacodynamic diversity reinforces the central argument that modifications to existing criteria are fundamentally insufficient; a dedicated, product-aware assessment framework is required.⁹ A recent machine learning-driven informetrics analysis of over 2,300 CAR-T publications identified adverse event management and safety optimisation as the most emerging research cluster, reflecting a field-wide recognition that standardised assessment frameworks—including imaging response criteria—remain a critical unmet need.¹⁰

Addressing this diagnostic vulnerability requires a paradigm shift towards multidisciplinary healthcare delivery. The intersection of haematology, nuclear medicine, neuroradiology, and molecular pathology is essential to accurately decode the complex post-infusion landscape. This review critically evaluates the limitations of current imaging criteria, explores the prognostic utility of quantitative PET metrics and liquid biopsies, and proposes a novel, multidisciplinary assessment framework tailored specifically for cellular immunotherapies.

Pre-Infusion Imaging: Baseline Metabolic Tumour Burden as A Prognostic Biomarker

Total Metabolic Tumour Volume (TMTV) acts as a quantitative surrogate for global lymphoma burden. High baseline TMTV functions as a vast antigenic sink that can prematurely exhaust infused CAR T-cells, establishing an immunosuppressive microenvironment that impedes effector cell persistence. Elevated pre-treatment TMTV independently correlates with compromised therapeutic efficacy, inferior progression-free survival (PFS), and reduced overall survival (OS).^11,12 Volumetric burden also directly influences the incidence of severe systemic toxicities. Patients harbouring an elevated baseline metabolic volume (>80 mL) face significantly higher risks of developing high-grade cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS).^13,14

The manufacturing process for autologous CAR-T cell products introduces a compulsory delay of three to five weeks between leukapheresis (time of decision, TD) and reinfusion (time of transfusion, TT). Tumour growth during this bridging interval fundamentally alters baseline prognostic parameters and necessitates dual-timepoint baseline assessment. The pre-infusion tumour growth rate (TGR) operates as a highly sensitive, independent prognostic factor; rapid positive TGR indicates highly proliferative, therapy-resistant clones that outpace subsequent CAR-T cell expansion. Increases in MTV between the pre-leukapheresis and pre-lymphodepletion timepoints strongly associate with an increased risk of death, while absolute pre-lymphodepletion MTV dictates the risk of severe (grade 3+) ICANS.¹⁵ To synthesize these multidimensional data, the International Metabolic Prognostic Index (IMPI) integrates baseline TMTV, patient age, and advanced disease stage, significantly outperforming standard clinical indices in predicting survival trajectories.^16,17

Macroscopic volumetric indices fail to capture the internal architecture and spatial heterogeneity of malignant lesions. Computational radiomics extracts high-dimensional, voxel-level quantitative data from pre-infusion PET/CT images. Shape-based morphological features and textural complexity quantify geometric irregularity and the spatial arrangement of differing metabolic intensities. These computational metrics reflect underlying biological phenomena—such as neoangiogenesis, focal necrosis, and dense stromal infiltration—that physically impede the homing and penetration of circulating CAR-T cells. Consequently, incorporating novel radiomic signatures provides independent prognostic value that extends beyond basic volumetric metrics, accurately predicting overall treatment response, PFS, and OS.^18,19

Maximum tumour dissemination (Dmax), defined as the maximum Euclidean distance separating the two most remote hypermetabolic lesions, reflects the spatial distribution of the disease. Extensive baseline Dmax operates as a significant independent predictor of PFS, forcing the compulsory division of the in vivo effector pool and reliably predicting earlier progression.²⁰ Additionally, maximum standardized uptake value (SUVmax) serves as a marker of aggressive metabolic potential. High pre-therapeutic SUVmax (> 9.0) correlates with shorter PFS and OS, and baseline values at TD (> 17.1) or TT (> 12.1) identify patients at high risk of treatment failure and severe CRS.^21–23 See Table 1.

Table 1.

Summary of Pre-Infusion ¹⁸F-FDG PET/CT Parameters and Their Prognostic Value in CAR-T Therapy

PET/CT Parameter	First Author (Year)	N	Optimal Threshold	Primary Prognostic Finding (with HR, 95% CI, and p-value)	Associated Endpoints	Segmentation Method
TMTV / MTV	Dean (2020)13	96	>147.5 mL	Low MTV predicts superior OS (HR=0.25, 95% CI 0.10–0.66) and PFS (HR=0.40, 95% CI 0.18–0.89)	PFS, OS	Manual
	Galtier (2023)24	160	>80 mL	High TMTV predicts inferior OS (HR=4.52, p<0.001) and PFS (HR=2.05, p=0.009)	PFS, OS	41% SUVmax
	Breen (2023)15	69	Continuous	Greater pre-lymphodepletion MTV significantly correlates with severe (Grade 3+) ICANS (p=0.042)	ICANS	Absolute SUV 2.5
SUVmax	Cohen (2022)22	48	>17.1	High baseline SUVmax independently predicts shorter OS (HR=10.3, p<0.01)	OS, PFS	Variable
	Peters (2025)21	18	>9.0	SUVmax > 9.0 indicates shorter PFS (p=0.04) and OS (p<0.01); HR=7.0 for progressive disease	PFS, OS	Variable
	Guidetti (2023)23	38	Continuous	Pre-infusion SUVmax correlates strongly with the risk and severity of subsequent CRS (rs=0.806, p<0.001)	CRS	41% SUVmax
Dmax	Mirshahvalad (2025)20	51	>14 cm	Extensive baseline Dmax predicts inferior PFS (HR=13.08, p=0.013)	PFS, OS	41% SUVmax
Radiomics	Zhang (2022)25	152	Radiomic Signature	RS based on TMTV independently predicts PFS and OS; hybrid nomograms outperform IPI (p<0.05)	PFS, OS	41% SUVmax
IMPI	Winkelmann (2025)16	504	>1.35 (Median)	CAR-IMPI significantly discriminates risk categories for PFS (p<0.0001) and OS (p<0.0001)	PFS, OS, CRS	41% SUVmax

Post-Infusion Response Assessment: Limitations of Current Criteria

Radiologic monitoring post-infusion operates on a framework structured around PET/CT examinations at approximately one month (M1) and three months (M3). The M1 scan provides initial kinetic data, separating rapid responders from those exhibiting primary refractoriness.²⁶ The M3 evaluation serves as the confirmatory scan; by day 90, the overwhelming majority of therapy-induced systemic inflammatory responses have subsided, leaving a clear metabolic window to observe residual malignancy and confirm durable remission²⁷ see Figure 1.

Figure 1.

Mapping the CAR-T Journey: A PET/CT Assessment Timeline. This figure illustrates the sequential PET/CT evaluation time points across the CAR-T treatment continuum, from leukapheresis (time of decision) through bridging therapy, lymphodepletion, CAR-T infusion, and post-infusion landmark assessments at Months 1, 3, and 6. Key biological events (CRS window, CAR-T expansion peak, immune reconstitution) are overlaid to contextualise the imaging assessment windows.

The Lugano 2014 classification and D5PS were engineered for mid- and end-of-treatment assessment in chemoimmunotherapy. The fundamental limitation lies in its suboptimal positive predictive value (PPV), peaking at approximately 60% due to reactive post-treatment inflammation.¹⁷ The mechanism of engineered T-cells precipitates massive cytokine release and dense local inflammatory cell recruitment. Applying a rigid Deauville framework to an environment flooded with highly metabolic immune effectors virtually guarantees the mischaracterization of therapeutic inflammation as active, refractory disease.

The unsuitability of legacy criteria leads to severe clinical discordance. The exact same radiologic dataset can yield divergent response classifications depending on the framework applied. The Lugano criteria classify up to 32% of post-infusion patients as progressing, primarily driven by the rigid designation of “progressive metabolic disease” without accompanying volumetric expansion. In sharp contrast, RECIL classifies only 17% of the same cohort as progressing.²⁸ This discordance leaves clinicians navigating contradictory endpoints.

Diagnostic confusion is largely driven by pseudoprogression—an acute, transient increase in the physical size or metabolic avidity of FDG-avid lesions that subsequently undergoes spontaneous regression. Occurring in roughly 5% to 10% of lymphoma patients receiving cellular immunotherapy, this phenomenon typically emerges between weeks 2 and 6 post-infusion. It reflects the in vivo expansion of tumour-infiltrating CAR-T cells. Failure to recognize this inflammatory mimic during early M1 evaluations severely compromises accurate response assessment.⁴ The clinical consequence is catastrophic: premature classification as progressive disease frequently leads to the inappropriate initiation of salvage therapies, including high-dose corticosteroids, which rapidly ablate circulating CAR-T cells and destroy the potential for a delayed cure.

Engineered T-cells do not exert equivalent cytotoxic pressure uniformly across all anatomical compartments. Extranodal (EN) sites, particularly the lung, pleura, and gastrointestinal tract, consistently demonstrate the lowest local response rates due to hostile extracellular matrices and restricted vascular permeability. EN relapses carry a dramatically worse prognosis, yielding a 2-year OS rate of 23%, compared to 64% in patients whose relapses remain confined to nodal compartments.²⁹ Standard whole-body imaging metrics fail to account for this spatial heterogeneity. See Table 2.

Table 2.

Limitations of Standard Imaging Response Criteria and CAR-T-Specific Imaging Phenomena

Assessment Dimension	Lugano/Deauville Standard Approach	Limitation in CAR-T Context	CAR-T-Specific Phenomenon	Estimated Frequency/Magnitude	Clinical Consequence if Unaddressed	Key Reference(s)
A. Response Classification & Scoring
Interpretation of residual FDG uptake	Deauville scores 4 or 5 definitively denote partial or progressive disease (PD), mandating a change in therapy.	Binary visual thresholds yield a suboptimal positive predictive value, failing to differentiate active tumour from immune infiltration.	Early metabolic flare, localised immune activation, and histiocytic reactions mimicking viable tumour cells.	26% of Day-30 DS4 and 8% of DS5 patients spontaneously convert to complete response (CR).	Premature termination of curative therapy; unnecessary exposure to toxic salvage regimens or stem cell transplantation.	[24,30]
B. Assessment Timing & Kinetics
Static milestone-based evaluation	Fixed scans assess response at mid- or end-of-treatment; early lesion growth signifies definitive therapeutic failure.	Static snapshots miss dynamic kinetic evolution, misclassifying delayed tumour clearance as refractory disease.	CAR-T induced pseudoprogression (CARTiPP) and delayed metabolic clearance occurring weeks to months post-infusion.	~33% (one-third) of patients exhibiting only partial response at Day 30 eventually achieve complete remission.	Inappropriate initiation of immunosuppressive steroids or radiation, prematurely destroying the in vivo CAR-T effector pool.	[31,32]
C. Anatomical & Biological Confounders
Physiological background reference	Liver uptake serves as a static, reliable threshold to visually grade the intensity of residual lymphoma.	Hepatic metabolism fluctuates significantly due to systemic immune responses, destabilising the baseline reference.	Systemic inflammation and acute phase protein release elevate hepatic FDG uptake, altering Deauville grading.	Decreased Delta-Liver-SUVmean at 30 days yields an inferior median PFS (3.0 months vs Not Reached).	False-negative/positive interpretations due to shifting reference backgrounds, skewing survival prognostication and risk stratification.	[5,33]
D. Molecular–Imaging Discordance
Defining progressive disease (PD)	Progressive metabolic disease (increased FDG uptake) triggers failure classification, even without structural growth.	Strict metabolic criteria conflict with morphological frameworks (RECIL), creating profound inter-criteria discordance.	Massive localised cytokine release drives intense glycolysis that outpaces or occurs without actual clonal proliferation.	PD rates fluctuate drastically by criteria: 32% (Lugano), 27% (Cheson), 17% (RECIL), and 17% (LYRIC).	Contradictory trial endpoints; patients classified as progressing by Lugano may still benefit from ongoing CAR-T activity.	[16]
E. Neurotoxicity Assessment
Central nervous system evaluation	Brain FDG-PET is not routinely indicated or evaluated; focuses purely on lymphomatous compartments.	Existing frameworks ignore profound systemic immune-related neurotoxicities unique to cellular immunotherapies.	Immune effector cell-associated neurotoxicity syndrome (ICANS) causes extended bilateral frontolateral and orbitofrontal hypometabolism.	High-grade ICANS occurs in ~10–32% of patients, frequently demonstrating higher baseline metabolic tumour volumes.	Misdiagnosing ICANS encephalopathy as fatal CNS lymphoma relapse, delaying critical corticosteroid or targeted interventions.	[34,35]

Abbreviations: CAR-T, Chimeric Antigen Receptor T-cell; FDG, Fluorodeoxyglucose; DS, Deauville Score; PD, Progressive Disease; CR, Complete Response; PFS, Progression-Free Survival; SUV, Standardised Uptake Value; RECIL, Response Evaluation Criteria in Lymphoma; LYRIC, Lymphoma Response to Immunomodulatory Therapy Criteria; ICANS, Immune Effector Cell-Associated Neurotoxicity Syndrome; CNS, Central Nervous System; DLBCL, Diffuse Large B-Cell Lymphoma; PMBCL, Primary Mediastinal B-Cell Lymphoma.

Imaging Mimics: CRS and ICANS on PET and MRI

CRS precipitates a widespread inflammatory cascade that alters the physiological biodistribution of FDG. The massive release of pro-inflammatory cytokines drives marked metabolic shifts, revealing intense, symmetrical FDG uptake within the spleen and bone marrow compartments. These hypermetabolic, immune-mediated phenomena create severe diagnostic ambiguity, generating a radiologic signature functionally indistinguishable from residual lymphoma. Integrating continuous quantitative metabolic metrics alongside clinical inflammatory markers is vital to differentiate therapy-induced reticuloendothelial activation from malignant progression.⁴

Immune effector ICANS requires rigorous multidisciplinary evaluation. Standard structural neuroimaging frequently fails to correlate with clinical encephalopathy. Conventional Magnetic Resonance Imaging (MRI) remains normal in a majority of patients experiencing acute neurotoxicity. When structural MRI abnormalities manifest, they typically present as symmetrical T2/FLAIR hyperintensities within the periventricular white matter, or leptomeningeal enhancement.³⁶ Electroencephalography (EEG) operates as a more sensitive diagnostic tool in the acute phase, capturing functional derangements before tissue damage becomes visible.³⁷ Functional connectivity MRI can detect neural network disruptions directly correlating with observed neurotoxicity.³⁸ See Figure 2.

Figure 2.

Visualizing the “Living Drug”: Imaging Responses and Toxicities in CAR T-Cell Therapy. This figure presents representative schematic imaging patterns encountered during CAR-T therapy, including: (A) pseudoprogression at Month 1 with spontaneous resolution at Month 3; (B) CRS-associated diffuse FDG uptake in the spleen and bone marrow; (C) ICANS-related MRI and brain PET findings.

Given the limited sensitivity of structural MRI, ¹⁸F-FDG PET of the brain is emerging as a valuable diagnostic adjunct. Brain PET scans capture regional cortical metabolic derangements underpinning the encephalopathic state. Patients with advanced ICANS consistently demonstrate bilateral hypometabolism localizing to the frontolateral and orbitofrontal cortices. Establishing a dedicated neuro-PET assessment framework would allow clinicians to differentiate ICANS-induced encephalopathy from occult central nervous system lymphoma relapse.

Circulating Tumour DNA: A Molecular Complement to Imaging

Because FDG maps metabolic intensity rather than cellular lineage, standard imaging cannot reliably distinguish active inflammation from genuine residual malignancy early post-infusion. Circulating tumour DNA (ctDNA) provides a tumour-specific molecular signal independent of localized FDG avidity. By quantifying fragments of tumour-derived nucleic acids in the peripheral bloodstream, liquid biopsies bypass the inflammatory noise of the microenvironment.

Higher baseline ctDNA concentrations strictly correlate with inferior PFS and OS, alongside an elevated risk of severe CRS and ICANS. Frank et al (2021)³⁹ demonstrated that ctDNA tracking effectively captures disease kinetics, detecting failure at or before radiographic relapse in 94% of patients. Patients achieving undetectable ctDNA by day 28 experienced exceptional long-term outcomes (median PFS not reached), whereas those with detectable ctDNA faced an exceedingly dismal prognosis (median PFS 3 months).

The rate of ctDNA clearance at sequential milestones (days 7, 14, and 28) is an exceptionally accurate early predictor of therapeutic response. Cohorts achieving ctDNA-negative status by day 28 demonstrate a 1-year PFS of 90.9% versus 27.3% in persistently positive patients. Dynamic monitoring vastly outperforms conventional PET/CT in early risk stratification: among patients with a partial radiographic response (PR) at day 28, only 1 out of 10 with undetectable ctDNA relapsed, compared to 15 out of 17 with detectable ctDNA.⁴⁰

End-of-therapy ctDNA Minimal Residual Disease (MRD) assessment carries vastly superior prognostic utility compared to conventional PET imaging (HR 28.3 for detectable ctDNA vs HR 3.6 for positive PET).⁴¹ Molecular clearance operates as a leading indicator, predicting complete radiologic response approximately 97 days ahead of imaging confirmation.⁴² Combining genomic sequencing with spatial imaging solves the pseudoprogression dilemma: undetectable ctDNA alongside a simultaneously enlarging PET lesion heavily dictates a diagnosis of localized benign pseudoprogression. See Table 3.

Table 3.

Head-to-Head Comparison of ¹⁸F-FDG PET/CT and ctDNA for Response Assessment at Key Post-CAR-T Time Points

Time Point	PET/CT: What Is Measured	PET/CT: Diagnostic Performance	PET/CT: Key Limitation at This Time Point	ctDNA: What Is Measured	ctDNA: Diagnostic Performance	ctDNA: Key Limitation at This Time Point	Concordance/Discordance	Recommended Clinical Action Based on Combined Assessment	Key Reference(s)
A. Pre-Infusion Baseline
Time of Decision (TD, leukapheresis)	Absolute SUVmax, baseline TMTV, TLG, and Dmax.	High SUVmax (>17.1) independently predicts shorter OS (HR=10.3). High TMTV predicts inferior PFS.	Static snapshot; misses aggressive tumour growth dynamics during the manufacturing and bridging interval.	Baseline ctDNA concentration via tumour-informed NGS.	High baseline concentration predicts inferior PFS/OS and higher risk of severe CRS/ICANS.	Dependent on tumour shedding; poorly vascularised or CNS sanctuary sites may yield false negatives.	Good concordance; both modalities accurately reflect the total systemic antigenic load prior to therapy.	Identify high-risk patients; consider intensive bridging strategies and prepare proactive CRS/ICANS prophylaxis.	43 ★★;39 ★★★
Time of Transfusion (TT, pre-lymphodepletion)	Tumour growth rate (TGR) from TD; updated MTV/TLG.	Increased MTV from TD to TT strongly predicts progression, death, and severe ICANS.	Recent bridging chemo- or radiotherapy induces transient inflammation, confounding accurate metabolic interpretation.	Pre-lymphodepletion absolute ctDNA concentration.	Concentration correlates closely with MTV; detects disease resistant to recent bridging therapy.	Bridging therapy may transiently suppress shedding, causing false reassurance despite remaining chemoresistant clones.	Concordant in capturing bridging-resistant growth (positive TGR with rising or stable ctDNA).	Adjust lymphodepletion intensity if needed; anticipate severe toxicities if tumour burden has expanded.	15 ★★;28 ★★
B. Early Post-Infusion
Day 7 (D7)	Not routinely performed or indicated.	No published data.	Intense cytokine storm and CAR-T expansion create massive non-malignant FDG uptake (flare).	Early kinetic molecular clearance.	Rapid clearance by D7 strongly predicts eventual durable complete response.	Assay turnaround time currently precludes real-time clinical decision-making.	No published data.	Observation and toxicity management only; do not alter anti-lymphoma therapy.	39★★★
Day 14 (D14)	Not routinely performed or indicated.	No published data.	Peak systemic inflammation and macrophage activity maintain a high false-positive FDG signal.	Ongoing kinetic clearance rates.	Undetectable ctDNA at D14 accurately separates long-term responders from non-responders.	Turnaround delays results; universally accepted quantitative clearance thresholds remain undefined.	No published data.	Continue supportive observation; avoid premature corticosteroid administration unless severe ICANS/CRS dictates.	40 ★★;39★★★
Day 28/Month 1 (D28/M1)	Deauville Score (DS), ΔSUVmax, MTV.	DS 1–3 predicts durable CR. DS 5 indicates failure. ΔSUVmax >66% predicts better OS.	High false-positive rate (~40% PPV) for DS 4 due to immune infiltration (pseudoprogression).	Detection status (qualitative: undetectable vs detectable).	Highly specific (94%). Undetectable status yields excellent 1-year PFS (90.9%).	Antigen-loss escape (eg., CD19-negative clones) may bypass tumour-informed sequencing primers.	Highly discordant in PET PR/SD; many PET-positive cases are ctDNA-negative (benign pseudoprogression).	If PET PR/SD but ctDNA-negative: observe. If both tests are positive: high risk for relapse.	39★★★;44★★; 1 ★★
C. Late Post-Infusion
Month 3 (M3/D90)	Absolute SUVmax (>6.3), MTV (>120 mL), DS.	Superior PPV. SUVmax ≥6.3 yields 8-fold mortality risk; high MTV yields 10-fold risk.	Localised sarcoid-like granulomatous reactions may still occasionally mimic residual disease.	End-of-treatment MRD monitoring.	All durable responders are undetectable Detectable ctDNA highly specific for definitive failure.	Clonal evolution and genetic drift may render baseline-informed primers obsolete.	High concordance. Dual-positivity definitively confirms therapeutic failure.	If dual-positive, promptly initiate salvage therapy or enroll in clinical trials.	44★★; 39 ★★★
Month 6 (M6) and beyond	Routine surveillance scans (Lugano criteria).	Sensitive for macroscopic relapse, but yields false-positives from infections or secondary primary malignancies.	Routine imaging in asymptomatic patients lacks proven survival benefit and adds radiation exposure.	Serial ctDNA surveillance tracking.	Detects molecular relapse approximately 97 days before overt radiographic progression.	Extreme cumulative cost of serial NGS; ongoing logistical burden.	Discordant early in relapse (ctDNA-positive, PET-negative) until macroscopic structural growth occurs.	Use serial ctDNA to trigger early, targeted PET/CT; enables pre-emptive intervention before symptomatic relapse.	42 ★★;41 ★★★

Notes: Data Maturity & Evidence Base: Pre-infusion and Day 28/Month 3 PET/CT parameters have highly robust, mature evidence with prospective multicentre validation (★★★). Early ctDNA kinetics (Day 7/14) and Month 6+ surveillance currently rely on early prospective cohorts (★★) but lack global standardisation. PET/CT at Day 7/14 lacks published diagnostic data and is universally discouraged. Assay Types: ctDNA evidence predominantly reflects tumour-informed, capture-based NGS or multiplex PCR (mPCR) (eg., clonoSEQ/IgHTS, Signatera), requiring a baseline tissue sample. Emerging assays (PhasED-Seq) may offer higher sensitivity without baseline tissue. Logistics: ctDNA turnaround times (often 1–2 weeks) currently limit acute clinical utility compared to same-day PET/CT results. Cost for serial bespoke NGS assays remains substantially higher than standard functional imaging. Evidence Level: ★★★ = prospective multicentre; ★★ = retrospective multicentre / prospective single centre; ★ = case series/pilot.

Abbreviations: OS, Overall Survival; PFS, Progression-Free Survival; HR, Hazard Ratio; PPV/NPV, Positive/Negative Predictive Value; DS, Deauville Score; MTV, Metabolic Tumour Volume; TLG, Total Lesion Glycolysis; TGR, Tumour Growth Rate; CRS, Cytokine Release Syndrome; ICANS, Immune Effector Cell-Associated Neurotoxicity Syndrome.

Novel and Emerging Imaging Approaches

Because FDG maps global glucose consumption, it captures the intense glycolysis of reactive macrophages. Innovative research involves the direct visualisation of immune effector cell function. The radiotracer ⁶⁸Ga-Grazytracer facilitates the non-invasive assessment of granzyme B activity, a primary serine protease released by cytotoxic T-cells, serving as a highly specific surrogate for active immune-mediated tumour eradication.⁴⁵ CXCR4-directed radiotracers, specifically ⁶⁸Ga-Pentixafor, generate contrast in CXCR4-expressing lymphomas, providing a distinct molecular signature independent of glycolytic shifts.⁴⁶

Hybrid PET/MRI systems offer a multiparametric diagnostic framework. Diffusion-weighted imaging (DWI) via MRI complements PET by quantifying tissue microstructure. Highly cellular malignant tissues restrict water movement, yielding a low Apparent Diffusion Coefficient (ADC). Fusing these modalities allows clinicians to differentiate benign necrosis (low ADC, lacking FDG avidity) from active refractory lymphoma (low ADC, intensely high FDG uptake).

Automated tumour segmentation powered by deep learning architectures can rapidly and reproducibly quantify TMTV, eliminating human bias. By translating subtle voxel-level variations into validated predictive signatures, artificial intelligence (AI)-guided quantitative analysis transitions radiomics from an experimental exercise into a standardized clinical tool.⁴⁷

Proposed Framework: Towards Car-T-Specific Imaging Response Criteria

Legacy criteria lack the diagnostic precision required for cellular therapy. A CAR-T-specific multidisciplinary system must structurally accommodate delayed responses, pseudoprogression, site-specific response heterogeneity, and the critical integration of molecular biomarkers (ctDNA) to confirm true clonal eradication.

The CAR-T Response Assessment Framework (CART-RAF) mandates the strict multi-modal integration of PET/CT imaging with concurrent ctDNA quantification at D28, D90, and an extended D180 evaluation. The framework introduces a rigid pseudoprogression rule: patients demonstrating positive PET/CT findings (eg., D5PS 4 or 5) at D28, but concurrently exhibiting negative ctDNA, are classified into an “indeterminate, likely pseudoprogression” category, see Figure 3. This prohibits premature salvage regimens and mandates a repeat evaluation at day 60. The framework also replaces binary visual D5PS interpretation with continuous variables (∆SUV_max, ∆TMTV) and incorporates extranodal site risk stratification as an independent modifier. See Table 4.

Figure 3.

Navigating post-CAR-T recovery: the CART-RAF response assessment framework. This figure presents the algorithmic decision tree of the proposed CART-RAF, integrating PET/CT and ctDNA results at Day 28 into four response categories: Complete Metabolic-Molecular Response (PET−/ctDNA−), Metabolic Response with Molecular Residual Disease (PET−/ctDNA+), Indeterminate Probable Pseudoprogression (PET+/ctDNA−), and Molecular-Confirmed Progressive Disease (PET+/ctDNA+).

Table 4.

Systematic Comparison of Existing Response Assessment Frameworks and the Proposed CART-RAF

Assessment Dimension	Lugano Classification (2014)	LYRIC (2016)	iRECIST (2017)	RECIL 2017	CART-RAF (Proposed)	Evidence Status / Validation Level
A. Foundational Design
1. Original disease/therapy target	Lymphoma (HL/NHL); conventional cytotoxic and targeted therapies.	Lymphoma; immunomodulatory therapies (eg., checkpoint inhibitors).	Solid tumours; immunotherapeutics.	Lymphoma; standard cytotoxic and targeted therapies.	Large B-cell lymphoma (LBCL); CAR-T cellular immunotherapy.	Defines the baseline scope for each response framework.
2. Founding publication	[3]	[48]	[49]	[50]	Proposed in current manuscript.	Standard guidelines [GREEN]; Proposed [RED].
3. Primary imaging modality	18F-FDG PET/CT.	18F-FDG PET/CT.	CT or MRI (morphological).	18F-FDG PET/CT & CT.	18F-FDG PET/CT + functional Brain PET/MRI.	PET/CT is gold standard [GREEN].
4. Measurement method	Bidimensional (SPD ≤6 lesions); visual Deauville (D5PS).	Bidimensional (SPD) + visual D5PS.	Unidimensional (RECIST 1.1) + new lesions.	Unidimensional (SLD ≤3 target lesions) + D5PS.	Continuous quantitative (∆SUVmax, ∆TMTV) + D5PS.	Quantitative metrics improve PPV [AMBER].
B. Response Categories & Progression Rules
5. Response categories	CMR, PMR, NMR, PMD.	CR, PR, SD, PD + Indeterminate Response (IR).	iCR, iPR, iSD, iUPD, iCPD.	CR, PR, Minor Response (MR), SD, PD.	Continuous metabolic/molecular response; integrated IR category.	Standard categories yield high false-positives [GREEN].
6. Handling of pseudoprogression	✗ None; flares classified as progressive disease.	✓ IR1, IR2, IR3 categories capture inflammatory flares.	✓ iUPD designation accounts for flare.	◐ Provisional allowance; relies mostly on unidimensional SLD.	✓ “Indeterminate, likely pseudoprogression” designated if PET+ but ctDNA-.	Pseudoprogression affects 5–10% of CAR-T patients [AMBER].
7. Progression definition	D5PS 4–5 with >50% SPD increase or new FDG-avid lesions.	Confirmed PD after initial IR, or clear clinical decline.	iCPD requires subsequent scan confirming prior iUPD.	>20% increase in SLD or new lesion appearance.	Dual-positivity (PET+ and ctDNA+) or confirmed growth at Day 60.	Highly discordant PD rates across standard criteria [AMBER].
8. Confirmation of progression required?	✗ No.	✓ Repeat scan at 12 weeks for IR cases.	✓ Repeat scan in 4–8 weeks.	✗ No formal requirement.	✓ Repeat PET/ctDNA at Day 60 for IR cases.	Prevents premature cessation of CAR-T therapy [AMBER].
C. Molecular Integration
9. ctDNA/MRD integration	✗ None.	✗ None.	✗ None.	✗ None.	✓ Mandatory at baseline, D28, D90, and D180.	ctDNA clearance strongly predicts OS/PFS [GREEN].
10. Molecular override of imaging?	✗ No.	✗ No.	✗ No.	✗ No.	✓ Undetectable ctDNA + PET+ = likely pseudoprogression.	Resolves false-positive inflammatory PET signals [AMBER].
D. Time-Point & Logistics
11. Recommended assessment time points	Mid-treatment & End-of-treatment (EOT).	Baseline, EOT, repeat at 12 weeks for IR.	Varies by individual trial protocol.	Mid-treatment & EOT.	Day 28 (M1), Day 90 (M3), Day 180 (M6).	Early kinetic markers predict durability [GREEN].
12. Dual-baseline (TD + TT) concept	✗ Single baseline.	✗ Single baseline.	✗ Single baseline.	✗ Single baseline.	✓ Integrates bridging interval tumour growth rate (TGR).	Bridging TGR impacts prognosis and toxicity [AMBER].
13. Site-specific response weighting	✗ None.	✗ None.	✗ None.	✗ None.	✓ Extranodal (lung/GI) relapses heavily penalise prognostic score.	Extranodal relapse OS is exceptionally poor [AMBER].
E. CAR-T-Specific Adaptations
14. CRS/ICANS imaging provisions	✗ None.	✗ None.	✗ None.	✗ None.	✓ Incorporates brain PET/MRI for ICANS; accounts for CRS flare.	Structural MRI often fails to capture ICANS [AMBER].
15. Applicability to CAR-T therapy	◐ Suboptimal PPV (~60%); strict metabolic triggers.	◐ IR is helpful, but lacks molecular/quantitative depth.	✗ Poor fit for hypermetabolic haematological flares.	◐ SLD fails to capture metabolic-only shifts.	✓ Designed specifically for CAR-T cellular kinetics and mimics.	Standard criteria cause discordance [GREEN]; CART-RAF needs trial [RED].

Notes: Evidence Status / Validation Level Colour Code: [GREEN] = Validated in ≥1 prospective multicentre trial or represents widely accepted current gold standard. [AMBER] = Supported by retrospective data, single-centre studies, or extrapolated from other immunotherapy cohorts. [RED] = Proposed/theoretical concept requiring future prospective trial validation.

Abbreviations: CART-RAF, CAR-T Response Assessment Framework; HL, Hodgkin Lymphoma; NHL, Non-Hodgkin Lymphoma; SPD, Sum of the Product of the Diameters; SLD, Sum of the Longest Diameters; D5PS, Deauville 5-Point Scale; SUVmax, Maximum Standardised Uptake Value; TMTV, Total Metabolic Tumour Volume; CMR, Complete Metabolic Response; PMR, Partial Metabolic Response; NMR, No Metabolic Response; PMD, Progressive Metabolic Disease; CR, Complete Response; PR, Partial Response; SD, Stable Disease; PD, Progressive Disease; IR, Indeterminate Response; iCR/iPR/iSD/iUPD/iCPD, immune Complete Response / Partial Response / Stable Disease / Unconfirmed Progressive Disease / Confirmed Progressive Disease; ctDNA, circulating tumour DNA; MRD, Minimal Residual Disease; TD, Time of Decision; TT, Time of Transfusion; TGR, Tumour Growth Rate; CRS, Cytokine Release Syndrome; ICANS, Immune Effector Cell-Associated Neurotoxicity Syndrome.

Operational Challenges and Implementation Barriers

While the CART-RAF framework offers a conceptually robust solution to the limitations of legacy criteria, its clinical implementation faces several practical challenges that must be candidly acknowledged. First, the requirement for dual-modality assessment (PET/CT and ctDNA) at every landmark time point doubles the diagnostic workload and increases per-patient costs. Current ctDNA assay turnaround times of 7–14 days for NGS-based platforms may delay critical treatment decisions at day 28, when clinicians face pressure to initiate salvage therapy in patients with apparently progressive disease.³⁹ Second, ctDNA assays lack universal standardisation: clonoSEQ (IgHTS-based), PhasED-Seq (phased variant detection), and Signatera (tumour-informed mPCR) employ fundamentally different detection methodologies, and inter-assay concordance in the post-CAR-T setting has not been established. Third, tumour-informed assays require adequate baseline tumour tissue (typically FFPE), which may be unavailable in patients with limited biopsy material or prior exhaustive molecular profiling. Fourth, the framework demands multidisciplinary coordination between nuclear medicine physicians, haematologists, and molecular pathologists—a team structure that may not be available outside high-volume academic centres. Potential solutions include the development of faster ctDNA platforms (digital droplet PCR achievable in <48 hours), centralised reference laboratory networks,⁴² AI-assisted PET interpretation algorithms to reduce inter-reader variability, and tiered implementation models where ctDNA assessment is prioritised for discordant or equivocal PET results rather than applied universally.

Positioning CART-RAF Against Existing Immunotherapy Response Frameworks

Several immunotherapy-adapted response criteria exist, and it is essential to position the CART-RAF in relation to them. The LYRIC criteria (Cheson et al, 2016)⁴⁸ introduced the indeterminate response (IR) category to capture pseudoprogression in lymphoma patients receiving immune checkpoint inhibitors. While conceptually relevant, LYRIC has three critical limitations in the CAR-T context: (i) its 12-week confirmation window for IR cases is too long for CAR-T kinetics, where immune flare occurs as early as day 8 and most relapses manifest within 30 days;⁸ (ii) it relies entirely on imaging without molecular integration, meaning it cannot distinguish inflammatory PET positivity from genuine residual disease; and (iii) it has never been validated in any CAR-T-treated cohort. In classical Hodgkin lymphoma treated with nivolumab, IR occurred in 35% of patients at 3 months, with 29% of those subsequently achieving objective response; equivalent data for CAR-T populations do not exist.⁵¹ The iRECIST framework (Seymour et al, 2017)⁴⁹ addresses pseudoprogression in solid tumours through the iUPD/iCPD confirmation concept, requiring a follow-up scan at 4–8 weeks before confirming progression. This confirmation principle is the closest precedent to the CART-RAF dual-modality approach. However, iRECIST uses unidimensional CT measurements (RECIST 1.1), does not incorporate PET/D5PS, and was designed for the slower kinetics of solid-tumour checkpoint inhibitor responses—making it poorly suited for the hypermetabolic, rapidly evolving haematological flares characteristic of CAR-T therapy. The RECIL 2017 framework introduced a minor response category and simplified unidimensional measurement, but it shares the fundamental limitation of all imaging-only systems: it cannot resolve the molecular ambiguity of post-infusion FDG uptake.⁵⁰ The CART-RAF represents the first framework to integrate imaging with molecular biomarkers, incorporate product-specific considerations, and mandate site-specific response weighting—features absent from all existing criteria.

Practical Barriers to ctDNA Integration and Future Solutions

The integration of ctDNA into routine post-CAR-T assessment faces several practical barriers. The most immediate is cost: the Medicare Clinical Laboratory Fee Schedule rate for clonoSEQ was set at $2,007 per test effective January 2025, with episode pricing of $8,029, while research-grade assays such as PhasED-Seq are estimated at approximately $5,000 per test. For a framework mandating ctDNA at baseline, day 28, and day 90 (minimum three tests per patient), the incremental cost per patient exceeds $6,000, a non-trivial addition to already expensive CAR-T episodes costing $373,000–$475,000 per infusion. However, cost-effectiveness modelling suggests that ctDNA-guided strategies may ultimately reduce total healthcare expenditure by $33,000–$34,000 per patient through avoiding false-positive-directed salvage therapies, including unnecessary second-line CAR-T infusions.⁵² Regulatory barriers also persist: while clonoSEQ is the only commercially available, Medicare-covered ctDNA MRD test for DLBCL, broader insurance coverage remains incomplete, and the NCCN has only recently recommended ctDNA testing for PET-positive DLBCL patients at end-of-frontline therapy.⁵³ Assay-specific limitations include the ~4% clonotype identification failure rate, the inability of tumour-informed approaches to detect antigen-loss escape variants (eg., CD19-negative relapses), and the potential for clonal evolution to render baseline-established tracking primers obsolete at late surveillance time points. Future solutions under development include tumour-agnostic ctDNA approaches (fragmentomics, methylation-based assays), point-of-care digital droplet PCR platforms with sub-48-hour turnaround, and integration of ctDNA ordering into electronic health record workflows to streamline multidisciplinary coordination.

Limitations, Negative Results, and Scenarios Where the Framework May Underperform

Critical engagement requires acknowledging scenarios where the proposed approach may fail or underperform. The ctDNA-based molecular arbitration at the core of the CART-RAF assumes adequate tumour DNA shedding into the bloodstream. However, certain lymphoma subtypes and anatomical compartments are known to shed ctDNA poorly: primary mediastinal B-cell lymphoma, primary CNS lymphoma, and testicular lymphoma may yield false-negative ctDNA results even in the presence of active disease. In such cases, a PET-positive/ctDNA-negative result would be incorrectly classified as “indeterminate, likely pseudoprogression,” potentially delaying necessary salvage therapy.³⁹ Similarly, antigen-loss escape variants, a well-documented mechanism of CAR-T resistance in which tumor cells downregulate or lose CD19 expression, may not be captured by tumor-informed VDJ clonotype tracking, because the escaping clone may have undergone sufficient genomic divergence from the baseline sample. The 39% false-positive rate of D5PS cited from Bes et al also warrants contextualisation:¹⁷ this figure derives from end-of-treatment assessment in conventionally treated DLBCL, and the false-positive rate specifically in the post-CAR-T setting at day 28 may differ and has not been prospectively quantified.⁵⁴ Furthermore, the utility of serial ctDNA surveillance beyond month 6 remains unproven in prospective trials; extrapolation from retrospective cohorts and economic modelling carries inherent uncertainty. These limitations underscore that the CART-RAF should be regarded as a hypothesis-generating framework requiring rigorous prospective validation, not as a ready-to-implement clinical standard.²⁹ Moreover, emerging evidence demonstrates that the tumour immune microenvironment and novel cell death pathways, such as cuproptosis, independently influence prognosis in B-cell NHL, further underscoring that metabolic imaging alone captures only a fraction of the biological determinants of treatment outcome.⁵⁵

Research Agenda and Future Directions

Priority research questions must focus on optimizing assessment timing, validating ctDNA integration, and evaluating novel tracers (⁶⁸Ga-Grazytracer, ⁶⁸Ga-Pentixafor) for direct imaging of CAR-T effector function. Prospective multicentre trials, such as the proposed CAR-T-IMAGE study (mandating serial PET/CT and ctDNA at D0, D28, D60, D90, D180), are essential to confirm the concordance between the CART-RAF and long-term PFS. Furthermore, the development of radiomic biomarker signatures to predict pseudoprogression requires large-scale, TRIPOD-compliant, multi-institutional discovery studies. Successful execution of this agenda carries profound implications for health economics, rationalizing surveillance protocols, and reducing the incidence of unnecessary biopsies and toxic salvage therapies.

Conclusion

CAR-T cell therapy has definitively outpaced its legacy imaging assessment tools. Managing these complex patients requires a cohesive multidisciplinary approach encompassing haematologists, radiologists, and molecular pathologists. Pre-infusion metabolic tumour burden is indispensable for risk stratification, while post-infusion PET/CT utilizing the standard D5PS yields unacceptable false-positive rates due to pseudoprogression and CRS mimics. The longitudinal monitoring of ctDNA provides an essential molecular complement that successfully bypasses inflammatory radiologic noise, resolving severe diagnostic dilemmas. Clinical practice must transition from purely morphological frameworks to the integrated dual-axis response assessment formalized in the CART-RAF. Prospective multicentre validation of this framework is urgently required to establish a new global diagnostic standard, ensuring precise, personalized care for patients undergoing cellular immunotherapy.

Funding Statement

This work did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data Sharing Statement

The datasets used and/or analysed during this review are available from the author upon reasonable request.

Ethical Approval

Not applicable. This article is a review of previously published studies and did not involve human participants or animal research.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The author(s) report no conflicts of interest in this work.