{"id":"502f6629-13a9-48b7-9b5a-5cf76baa99a6","arxiv_id":"2505.23717","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":1.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A review synthesizing evidence on the required level of detail in atrial computer models for ablation planning, recommending personalized 3D or bilayer models with fiber direction and anisotropic conduction.","lead":"This review asks how much anatomical and biophysical detail is needed in computer models of the atria to plan atrial fibrillation ablation. It concludes that personalized 3D or bilayer models with fiber orientation and anisotropic conduction are currently recommended, while the suitability of simpler single-surface models remains an open question.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central recommendation presumes that in silico AF vulnerability and ablation success predict clinical outcome, a link the review itself says is unvalidated, so the 'reliable ablation planning' claim is expert opinion rather than evidence-backed conclusion.","rationale":"The paper is a broad, self-aware review rather than a new modeling study, and it deserves credit for being explicit about many open problems: it states that single-surface models remain an open question, that endo-epicardial dissociation may matter, that substrate localization from different clinical modalities is inconsistent, and that clinical validation is mandatory before clinical use. These admissions make the internal logic of the review fairly coherent. The central claim is the prescriptive Summary answer to the title question. For that claim to be true, two conditions must hold. First, the listed model components must be necessary to reproduce clinically relevant arrhythmia dynamics; the review marshals a reasonable body of simulation evidence for this, including comparisons of propagation models, wall-thickness effects, biatrial versus monoatrial models, and personalized ERP. Second, the in silico endpoints used to compare ablation strategies must predict clinical outcome. The second condition is the weak link. The evidence base cited for ablation planning relies on vulnerability reduction and non-inducibility as surrogates; no prospective, randomized, or even retrospective clinical outcome validation is presented. The review's own Section 10 wording, 'must pass clinical validation,' is effectively an admission that this link is missing. Adding to the concern, Section 7 shows that the choice of substrate data (LGE-MRI, low voltage, conduction velocity) changes the simulated vulnerability and ablation outcomes, so the recommended model detail may be necessary but is not sufficient, and the claim of 'reliable ablation planning' goes beyond what the present evidence establishes. This is not an internal inconsistency and not a reason to reject the review; it is a reason to keep the reader's conditional verdict: the authors should either temper the Summary to say 'to avoid known simulation artifacts, current evidence suggests...' or explicitly classify the recommendations as expert opinion pending clinical validation. The proposed retrospective test on the Azzolin 2023 cohort would settle whether the surrogate actually tracks recurrence, and a monolayer comparison would test whether the 3D/bilayer requirement itself changes the ranking of ablation strategies.","tokens_in":22160,"tokens_out":4925,"duration_ms":56235,"concrete_test":"Re-analyze the 29 personalized models in Azzolin et al. (2023, Europace 25:211-222) against the corresponding patients' clinical follow-up: for each patient, record the in silico predicted optimal ablation targets and vulnerability reduction, then test whether predicted non-inducible regions coincide with the sites of observed AF recurrence and whether vulnerability reduction discriminates recurrence-free patients. If the surrogate metric does not track recurrence, the recommended model detail is not justified for reliable clinical ablation planning. A cheaper complementary check: rerun the same 13 ablation strategies in a monolayer version of the same models and compare strategy rankings; if the ranking is unchanged, the claimed necessity of 3D/bilayer detail for ablation planning is weakened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing step is not any single equation but the inference from in silico vulnerability to clinical ablation outcome. Section 10 says that 'for reliable ablation planning' one needs a personalized 3D or bilayer model, fiber direction, anisotropic CV, remodeling, CV restitution, and regional ERP. The cited support uses surrogates: PEERP-based vulnerability (Azzolin et al. 2021) and vulnerability reduction after virtual ablation (Azzolin et al. 2023), not patient recurrence data. The review itself states in Section 10 that models 'must pass clinical validation' before clinical use, and Section 7 shows that substrate maps from LGE-MRI, low voltage, and conduction velocity disagree and change simulated outcomes. Thus the recommended 'reliable' package is an expert hypothesis; the link from model detail to patient outcome is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This review addresses the question of how much model detail is needed in computerized atrial electrophysiology simulations, with the stated intended use of planning and optimizing atrial ablation. It surveys cellular models, anatomical and fiber-orientation modeling, single-surface/bilayer/3D representations, propagation models from bidomain to eikonal, fibrosis modeling, verification and validation, in silico vulnerability assessment, personalization, cohort modeling, and ablation-strategy simulation. The authors conclude that for reliable ablation planning one needs a personalized 3D or bilayer atrial model with rule-based fiber direction, anisotropic conduction velocity, realistic remodeling, CV restitution, and regional effective refractory period, and they emphasize that models must pass clinical validation before use. The paper is a narrative synthesis rather than a new derivation, and it explicitly acknowledges several open questions.","tokens_in":22312,"tokens_out":5562,"duration_ms":50417,"significance":"If the recommendation package were established, it would provide a practical checklist for modelers and a target for clinical translation. The review has real strengths: it ties model choices to a specific context of use, it highlights the need for systematic verification and validation, it reports the PEERP protocol and the large-cohort ECG simulation efforts, and it is transparent about the current lack of clinical validation. The main caveat is that the central recommendation is an expert synthesis based largely on in silico surrogate endpoints (e.g., PEERP-based vulnerability and virtual ablation success) rather than on prospective clinical outcome data; the authors themselves state that the models must pass clinical validation before clinical use. The review would be more convincing if it labeled the summary package as a working hypothesis and more sharply separated established evidence from expert judgment.","major_comments":[{"comment":"Section 10 states that \"for reliable ablation planning, we need a personalized 3D model or a bilayer model of the atria,\" but Section 3.4 explicitly leaves open \"the question whether they [single-surface models] are suitable for in silico AFib vulnerability assessment.\" The manuscript does not provide evidence that single-surface models fail for the ablation-planning context, so the summary recommendation is stronger than the body of the review supports. Please either present the dimensional-model requirement as a research need or cite the specific studies that establish inferiority of single-surface models for this intended use.","section":"Section 10 vs. Section 3.4"},{"comment":"The central claim in Section 10 rests on in silico vulnerability surrogates: Section 6.1 describes the PEERP-based vulnerability metric (Azzolin et al. 2021), and Section 9 describes virtual ablation outcomes as vulnerability reduction in 29 patients (Azzolin et al. 2023), not patient-level arrhythmia recurrence. Because the review itself states in Section 10 that computational models \"must pass clinical validation\" before clinical use, the phrase \"for reliable ablation planning\" overstates the evidence. Please rephrase the conclusion to make explicit that the recommended package is a hypothesis that still requires prospective clinical validation with patient outcomes.","section":"Sections 6.1, 9 and 10"},{"comment":"Section 4.2 asserts that \"the most dangerous degree of fibrosis is between 30% and 60%.\" This is presented as a general conclusion, but the preceding text derives it from Keller et al.'s simulations of replacement fibrosis and from percolation-based modeling of fibrotic tissue; the evidence base does not appear to cover different fibrosis patterns, transmural extents, and patient-specific geometry. Given that this claim is used to argue against binary LGE-MRI fibrosis classification, please qualify it as model-based and pattern-dependent or provide a more systematic evidentiary basis.","section":"Section 4.2"}],"minor_comments":[{"comment":"Section 7 and Section 8 contain a figure-numbering inconsistency: Section 7 refers to \"Figure 5\" for the comparison of LGE-MRI, voltage, and conduction velocity, but the caption is numbered Fig. 4, and Section 8 then uses \"Figure 5\" for the P-wave fibrosis estimation figure. Please renumber the figures consistently.","section":"Sections 7 and 8"},{"comment":"The sentence listing the pacing protocols contains a duplicated \"(RP)\" after \"rapid pacing from 227 points,\" and the fibrosis labels \"H2, H3, H4 UII and UIV\" are not explained in the text; please clarify.","section":"Section 6.1"},{"comment":"Reference [38] is cited as \"Groot et al.\" in the text; the correct name is \"de Groot et al.\".","section":"Reference [38]"},{"comment":"In the paragraph on the universal atrial coordinate system, the text contains the typo \"Roney at al.\" instead of \"Roney et al.\".","section":"Section 3"},{"comment":"Reference [36] lists an author as \"Prof. Schmidt, C.\"; academic titles should be removed from the author list.","section":"Reference [36]"},{"comment":"The DREAM model and the cyclic fast iterative method are cited partly to preprint/conference items ([10,28]); if peer-reviewed versions are available, please cite those instead.","section":"Sections 4.3 and 8"}],"recommendation":"major_revision","confidential_remarks":"The paper is a review by a group that is itself a major contributor to many of the cited studies, so the concentration of self-citations is not surprising, but the summary recommendations should be tempered to reflect the level of evidence. The manuscript fits the journal's scope as a computational-physics review; I would support publication after revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a review from Dössel and Loewe's group asking how much model detail is needed for atrial electrophysiology simulations, with ablation planning as the intended use. The punchline: it's a solid, useful synthesis, but the summary recommendation overreaches the evidence it cites.\n\nThe paper does what a good review should do. It organizes a large, scattered literature—cell models, anatomy, fiber orientation, wall thickness, biatrial vs. monoatrial, 3D vs. bilayer vs. single-surface, propagation models, fibrosis modeling, personalization, cohort modeling, ablation strategies. The discussion of eikonal models and their speed-up is particularly helpful, as is the emphasis on verification and validation. I also appreciate the candid discussion of how LGE-MRI, low voltage, and conduction velocity maps disagree, and the suggestion that EP data may be more reliable than LGE-MRI for substrate localization. The paper is well written and the authors are honest about many open questions.\n\nNow the soft spots. The Section 10 claim that \"for reliable ablation planning\" we need a personalized 3D or bilayer model, fiber direction, anisotropic CV, and realistic remodeling is the central answer to the title question. But the evidence cited for that uses in silico vulnerability and virtual ablation outcomes, not patient recurrence. The review itself says models \"must pass clinical validation\" before clinical use—so the \"reliable\" is doing a lot of work. That recommendation should be framed as an expert hypothesis or a research priority, not a proven requirement. The \"most dangerous fibrosis 30-60%\" claim is also based on a small number of simulation studies and could be over-generalized. Self-citation is noticeable, but the authors are leading contributors and many claims are independently supported by other groups, so it's not a red flag.\n\nThe paper is not new in the sense of presenting results, but as a review it serves a clear purpose. I'd send it to peer review—a serious referee could push for the summary to be calibrated to the evidence and the clinical validation caveat to be elevated. The reader's CONDITIONAL verdict is about right.","headline":"A solid, honest review of atrial model complexity that overstates its own summary recommendation relative to the evidence it cites.","tokens_in":22763,"tokens_out":2489,"would_cite":true,"duration_ms":24355,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This review concludes that reliable atrial ablation planning requires personalized biatrial models with fiber direction, anisotropic conduction, and realistic refractory-period data.","keywords":["atrial fibrillation","computational electrophysiology","ablation planning","model complexity","eikonal model","biatrial modeling","fibrosis modeling","conduction velocity restitution"],"falsifier":"A prospective trial in which patients are randomized to ablation guided by a recommended-detail personalized model versus standard clinical practice; if post-ablation recurrence rates are statistically indistinguishable, the added model complexity is not clinically decisive.","tokens_in":21974,"feed_emoji":"🫀","tokens_out":10629,"duration_ms":94558,"temperature":0.7,"pith_summary":"This review asks how much anatomical and electrophysiological detail a computer model of the human atria needs in order to be trustworthy for a specific intended use, here the planning and optimization of ablation for atrial flutter and atrial fibrillation. It argues that the answer is context-dependent, and that for ablation planning the assembled simulation and clinical mapping evidence points to a definite minimum: a personalized volumetric or bilayer model of both atria, rule-based fiber direction with anisotropic conduction velocity, and realistic estimates of remodeling, conduction-velocity restitution, and regional effective refractory period. It also argues that single-surface isotropic models and simple fibrotic-versus-healthy tissue classifications are insufficient, and that extended eikonal models can simulate reentrant arrhythmias at a fraction of the computational cost of full reaction-diffusion models. If this prescription is correct, it converts an abstract debate about model complexity into a concrete checklist and indicates which patient-specific measurements are worth collecting.","feed_headline":"Model atria with fibers, anisotropy, and ERP for ablation","feed_subtitle":"Single-surface isotropic models are too crude; fast eikonal models may be the practical compromise for ablation planning.","key_machinery":"The argument is carried by several linked objects. The 'context of use' is the criterion that decides how much detail is needed, and here it is ablation planning. The bilayer model, a two-surface representation of the atrial wall that can hold different endocardial and epicardial fiber directions, is the recommended geometric compromise, with full volumetric three-dimensional models as the more expensive alternative. Rule-based fiber fields supply anisotropy without requiring patient-specific diffusion-tensor imaging. Anisotropic conduction velocity with its restitution and regional effective refractory period are the physiological properties identified as indispensable for reentry. The extended eikonal approach, including the reaction-eikonal and DREAM variants, provides the fast solver that can simulate reentrant arrhythmias several hundred times faster than reaction-diffusion models. The PEERP protocol is the suggested standardized measure of arrhythmia vulnerability that makes competing ablation strategies quantitatively comparable.","core_discovery":"The paper's central claim is that the level of detail in atrial electrophysiology models must be matched to the intended use, and that for ablation planning the minimum viable model is a personalized three-dimensional or bilayer representation of both atria with rule-based fiber architecture, anisotropic conduction velocity, and patient-specific estimates of remodeling, conduction-velocity restitution, and effective refractory period. The review supports the claim by surveying comparisons of biatrial with left-atrial-only models, volumetric with bilayer and single-surface geometries, bidomain/monodomain with eikonal formulations, and different ways of representing fibrosis. It reports concrete quantitative consequences, such as omitting ablation lines that connect to the nearest orifice lowering simulated success by about 20 percent, and imprecise fibrosis localization lowering it by about 35 percent. It also motivates a standardized in silico vulnerability protocol, PEERP (pacing at the end of the effective refractory period), which can interrogate hundreds of sites that could never be tested in a patient.","pith_inferences":["A direct clinical test the review does not run would randomize patients to ablation targets from a recommended-detail personalized model versus conventional pulmonary-vein isolation and compare atrial fibrillation recurrence; the recommended checklist stands or falls on that comparison.","If in silico vulnerability transfers to the clinic, the PEERP protocol could become a pre-ablation screening that finds latent triggers months before they manifest; that use is an extrapolation beyond the validation the review reports.","The same context-of-use logic could be run in reverse for cohort studies: when the question is population-level risk rather than an individual ablation plan, cheaper monolayer or purely eikonal models may be adequate, an option the review leaves only partially explored.","The review's 30-60 percent window of dangerous fibrosis suggests a testable prediction: clinical ablation outcome should depend on the fraction of moderately fibrotic tissue rather than total scar burden, which a retrospective imaging-outcome analysis could check."],"forward_implications":["Biatrial models should become the default for in silico vulnerability assessment, because the studies reviewed here show that including the right atrium can unmask additional inducing points in the left atrium.","For identifying fibrotic substrate, electrophysiological measurements such as voltage, conduction velocity, and restitution should be weighted more heavily than late gadolinium enhancement MRI, since the two modalities often locate the substrate differently.","Fast extended eikonal models make it practical to test hundreds of stimulation sites per virtual patient, so arrhythmia vulnerability can be checked before and after an ablation plan at a cost that full reaction-diffusion simulations cannot afford.","Ablation strategies should connect high-dominant-frequency ablation sites to the nearest orifice; the reviewed simulations find that omitting these connections lowers success by about 20 percent, and using the wrong fibrosis modality lowers it by about 35 percent.","Cohort simulations built from statistical shape models can generate thousands of virtual patients, which is how the review proposes to train machine-learning tools such as estimating fibrosis class from the 12-lead ECG."],"supporting_citations":[{"why":"Previous review that the article updates; supplies the baseline scope of atrial modeling that the detail question builds on.","marker":"[26]"},{"why":"Introduces the bilayer atrial model whose two surfaces hold different fiber directions, the geometric compromise the review recommends.","marker":"[51]"},{"why":"Biatrial versus left-atrial study showing the right atrium changes in silico vulnerability; underpins the recommendation to model both atria.","marker":"[66]"},{"why":"Clinical catheter measurements of conduction-velocity anisotropy and restitution in patients; supplies the data type the review says is indispensable.","marker":"[80]"},{"why":"Defines the PEERP vulnerability protocol and shows it finds inducing points other protocols miss; the standardized metric used to compare ablation strategies.","marker":"[9]"},{"why":"Systematic comparison of bidomain, monodomain, and eikonal propagation models; justifies the claim that fast eikonal models suffice for many uses.","marker":"[69]"},{"why":"Shows late gadolinium enhancement MRI, electrogram voltage, and conduction velocity localize atrial substrate differently; supports the review's preference for electrophysiological data over MRI.","marker":"[74]"},{"why":"Simulates 13 ablation strategies in 29 personalized atria; yields the quantitative success-rate differences for connections and fibrosis localization that the review reports.","marker":"[5]"}],"fun_headline_variants":["Oversimplified atrial models cut ablation success by 20-35%","Atrial ablation planning needs fibers, anisotropy, and ERP in models","Match atrial model detail to use: 3D or bilayer for ablation","Fast eikonal models with fibers and ERP may suffice for ablation","Personalized atrial modeling for ablation: what detail is required?"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that in silico measurements of how easily an atrium sustains atrial fibrillation, and of whether an ablation plan stops it, predict the real clinical outcome for patients.","fun_headline_variants_meta":{"raw":{"variants":["Oversimplified atrial models cut ablation success by 20-35%","Atrial ablation planning needs fibers, anisotropy, and ERP in models","Match atrial model detail to use: 3D or bilayer for ablation","Fast eikonal models with fibers and ERP may suffice for ablation","Personalized atrial modeling for ablation: what detail is required?"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000285,"raw_usage":{"total_tokens":1700,"prompt_tokens":989,"completion_tokens":711,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":605,"completion_tokens_details":{"reasoning_tokens":619}},"tokens_in":605,"tokens_out":711,"duration_ms":7701,"temperature":1.0,"reasoning_tokens":619,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:38:14.917662+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A prospective trial in which patients are randomized to ablation guided by a recommended-detail personalized model versus standard clinical practice; if post-ablation recurrence rates are statistically indistinguishable, the added model complexity is not clinically decisive.","supporting_citations":[{"cited_title":"MBEC 50(8), 773–799 (2012)","cited_arxiv_id":null,"evidence_quote":"Previous review that the article updates; supplies the baseline scope of atrial modeling that the detail question builds on."},{"cited_title":"Europace16 Suppl 4, iv21–iv29 (2014)","cited_arxiv_id":null,"evidence_quote":"Introduces the bilayer atrial model whose two surfaces hold different fiber directions, the geometric compromise the review recommends."},{"cited_title":"Heart Rhythm (2024)","cited_arxiv_id":null,"evidence_quote":"Biatrial versus left-atrial study showing the right atrium changes in silico vulnerability; underpins the recommendation to model both atria."},{"cited_title":"conventional ablation strategies to termi- nate atrial fibrillation and prevent recurrence","cited_arxiv_id":null,"evidence_quote":"Simulates 13 ablation strategies in 29 personalized atria; yields the quantitative success-rate differences for connections and fibrosis localization that the review reports."}],"review_version":1}