{"id":"3c93dfea-ee8a-4969-b34b-7d9058aaa9de","arxiv_id":"2506.20683","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A multimodal contrastive learning method, PTACL, aligns ECG segments with cardiac MRI frames at the same heart-cycle phase, improving ECG-based estimates of cardiac volumes and function.","lead":"This paper pairs heart MRI videos with ECG traces from the same patients and trains an AI so that an ECG alone can estimate heart volumes and pumping strength. Adding a time-matching step between the two signals improves those ECG-only estimates, which could make cheap heart checks more informative.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The core claim that local temporal alignment transfers CMR information to ECG is not yet supported: gains over global-only are small, and no ablation shows that correct phase correspondence, rather than any extra contrastive signal, causes them.","rationale":"The paper's central claim is empirical: adding a local temporal contrastive loss to global patient-level alignment improves ECG-based retrieval and regression of CMR phenotypes. The most load-bearing condition is that the local loss's P_{k,t} alignment corresponds to true cardiac phase; otherwise the method's novelty and mechanism are unsupported. The paper asserts this in Section 3.2 but does not validate it. The reported gains are small: e.g., Table 2 LP LVEDV R2 0.514 vs 0.507, and Table 1 P@1 EDVLV 0.510 vs 0.492. The local loss introduces many additional in-patient positive/negative pairs; even a randomly permuted alignment could provide a useful training signal by forcing ECG tokens to be temporally discriminative, so the improvement over Global cannot be attributed to correct R-peak-to-CMR correspondence without a control. Table 1 also contains per-metric decreases (P@5 EDVLV and EF LV), so the text's claim that local loss 'consistently improves' retrieval is overclaimed. A permutation/shift ablation and bootstrap confidence intervals would settle whether the local loss's benefit is due to genuine phase alignment. The reader identified the same R-peak alignment concern; our proposed test sharpens it into a falsifiable check. The concern is addressable, so the conditional verdict stands unchanged.","tokens_in":9369,"tokens_out":9123,"duration_ms":100991,"concrete_test":"Train PTACL with identical hyperparameters but three corrupted alignments: (a) a fixed random permutation of the T CMR frames shared across all patients, (b) a per-patient random permutation, and (c) a constant circular shift of q=3 frames. Evaluate LP R2 for all phenotypes in Table 2 and P@1/P@5 in Table 1. If the correct alignment does not outperform all corrupted variants, the gain is not due to true R-peak/CMR phase correspondence, and the central claim should be downgraded. Also compute bootstrap 95% confidence intervals over test subjects for the Global vs PTACL retrieval metrics, since none are currently reported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing premise is that the local loss in Eq. 3 uses a physiologically correct phase correspondence between interpolated ECG heartbeats and CMR frames, via the P_{k,t} alignment in Eq. 2. Section 3.2 asserts this ('The detected R peaks are later used to align and reconstruct a single representative heartbeat cycle. This physiological alignment motivates our local contrastive loss formulation'), but no experiment verifies that the improvements depend on this correspondence being correct. The measured gains are small (Table 2: LVEDV LP R2 0.514 vs 0.507; Table 1: P@1 EDVLV 0.510 vs 0.492), and the local loss adds many extra positive/negative pairs within each patient. It is therefore possible that any fixed within-patient temporal mapping, or even a shuffled one, would produce a similar or identical improvement by forcing ECG tokens to be phase-discriminative, independent of true cardiac-phase alignment. The paper's own Table 1 shows some retrieval metrics do not improve (P@5 EDVLV 0.445 vs 0.444; P@5 EF LV 0.378 vs 0.377), contradicting the claim that local loss 'consistently improves' retrieval. Without a permutation/shift ablation or error bars, the central claim that temporal alignment transfers CMR information is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes PTACL, a self-supervised multimodal contrastive learning framework that combines a global patient-level contrastive loss with a local temporal-alignment contrastive loss to align ECG and CMR embeddings. ECG and CMR encoders are first pre-trained separately with masked autoencoders; the local loss uses ECG R-peak gating to align interpolated single-heartbeat ECG tokens with CMR frames at matched cardiac phase. The method is evaluated on 27,951 UK Biobank subjects for retrieval of patients with similar cardiac phenotypes and for regression of ten CMR-derived cardiac parameters. The authors report that PTACL improves over a global-only baseline without adding learnable parameters, using only a single middle short-axis CMR slice.","tokens_in":9805,"tokens_out":5265,"duration_ms":56442,"significance":"If the central claim holds, the paper demonstrates a practically useful transfer of CMR-derived cardiac functional information into ECG representations, with no additional learnable parameters and with a relatively simple self-supervised objective. Strengths of the work include the large paired dataset, a self-supervised training setup that does not leak downstream labels into the ECG representation, public code release, and comparisons with several established baselines. The main weakness is that the evidence for the causal role of temporal alignment is incomplete: the gains over the global-only baseline are small, several reported metrics are flat or slightly worse, and no ablation isolates whether the physiological phase correspondence, rather than merely the extra intra-patient contrastive signal, is responsible for the improvements.","major_comments":[{"comment":"The claim that PTACL 'achieves better performance' and that the local loss 'consistently improves patient retrieval' is contradicted by the reported numbers. In Table 2, linear-probing R2 for LVEF LV drops from 0.258 (Global3 LP) to 0.255 (PTACL3 LP), and in Table 1 several metrics do not improve (e.g., P@5 EDVLV 0.445 vs. 0.444, P@5 EF LV 0.378 vs. 0.377, P@15 SV LV 0.380 vs. 0.377). Since the Table 2 footnote reports standard deviations only as 'not shown and in ~0.001–0.002' and Table 1 has no error bars, the differences driving the headline claims are within or near the noise level. Please weaken the 'consistently improves' wording and provide confidence intervals or significance tests for the key comparisons.","section":"Abstract; Table 1 caption; Table 2"},{"comment":"The central mechanism is the phase correspondence encoded by the alignment matrix P_{k,t}, but no experiment verifies that the improvements depend on this correspondence being correct. The local loss in Eq. (3) adds many within-patient positive and negative pairs even if the alignment matrix is arbitrary; a shuffled, constant, or shifted P_{k,t} could in principle produce the same small gains by forcing ECG tokens to be temporally discriminative. Please add an ablation that permutes or shifts the alignment matrix, or replaces it with a fixed non-physiological mapping, and report the retrieval and regression metrics for each variant. This is necessary to support the conclusion that local temporal alignment, rather than any auxiliary contrastive signal, transfers CMR information to ECG.","section":"Section 3.2, Eqs. (2)-(3)"},{"comment":"The retrieval evaluation reports a single run for each metric without standard deviations, confidence intervals, or significance tests. Given that the reported improvements are small (e.g., P@1 EDVLV 0.510 vs. 0.492, P@1 SVLV 0.472 vs. 0.458), these differences could easily arise from training stochasticity. Please report mean and standard deviation over multiple seeds, or bootstrap confidence intervals, for both retrieval and regression results so that the reader can assess whether the observed gains are statistically reliable.","section":"Section 4.1, Table 1"}],"minor_comments":[{"comment":"The abbreviation 'w/ I phase of SMP' is unclear and appears to be a typo; please clarify whether it means 'with one phase of single modality pre-training' and use a consistent notation.","section":"Table 2 footnotes"},{"comment":"The normalization condition 'PT t=1Pk,t = 1' should be written as sum_{t=1}^T P_{k,t} = 1, and the symbol δ_{k,t} should be explicitly defined as a Kronecker delta for the σ = 0 case.","section":"Eq. (2)"},{"comment":"The text calls the local alignment 'parameter-free', but the full objective still depends on hyperparameters β, σ, τ, and T; please state this more precisely to avoid overclaiming.","section":"Section 3.2"},{"comment":"The acronyms LP, FN, and FN+AP are used without full definitions; please define 'linear probing', 'fine-tuning', and 'attention pooling' at first use.","section":"Section 4.2, Table 2"},{"comment":"The notation for ECG local embeddings uses the index k while CMR local embeddings use t, but the text sometimes refers to 'time segments' and 'frames' interchangeably; please define all indices consistently.","section":"Section 3.2"},{"comment":"The Δ/□ formatting in Table 1 is visually confusing because the global and local symbols are embedded in the numeric cells; consider presenting the two models in separate columns or rows.","section":"Table 1"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for a medical imaging journal, and the large-scale UK Biobank evaluation is a strength. However, the main novelty is local temporal alignment, and the missing alignment ablation is the decisive point: without a shuffled or otherwise incorrect alignment control, the small and partly inconsistent gains cannot be attributed to physiological phase correspondence. This is fixable within the manuscript's scope, so major revision rather than rejection seems appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take on arXiv:2506.20683 (PTACL). The paper does something concrete: it adds a parameter-free local temporal contrastive loss to a CLIP-style global alignment between ECG and CMR, using R-peak gating to define ECG heartbeat segments and matching them to CMR frames. On 28k UK Biobank pairs, the global model is already strong, and the local loss adds a small bump in retrieval and regression for most phenotypes. The code is public. That is a real, if incremental, contribution.\n\nWhere it gets soft: the paper claims the local loss improves all phenotypes, but its own tables show drops (LVEF R2 from 0.258 to 0.255 in Table 2; P@5 for EF and EDV in Table 1 go down by 0.001). Those are tiny, but the claim is still overstated. More importantly, there's no ablation that tests whether the physiological R-peak correspondence actually matters. The local loss introduces many extra within-patient contrastive pairs; a shuffled or purely arbitrary temporal mapping might yield a similar bump. Without a permutation control, the stated mechanism—that correct cardiac-phase alignment transfers CMR functional information to ECG—is not established. That's the load-bearing claim of the paper, and it's currently supported only by the design, not by evidence.\n\nAlso, retrieval metrics lack error bars. Table 2 has a footnote saying std is 0.001–0.002 for one set of runs, but it's not clear which rows, and retrieval doesn't have any. Given the small effect sizes, this matters.\n\nThe paper is honestly written otherwise—the dependence on paired data is acknowledged, the hyperparameters are stated, the baselines are reasonable. The local loss is parameter-free, which is a plus. The method is likely to be useful even if its mechanism is not fully pinned down; the clinical motivation (cheap ECG getting a bit closer to CMR) is solid.\n\nWho should read it: anyone building multimodal contrastive models for time-series + imaging, or working on ECG-based screening. It deserves a serious referee, not a desk reject. But a competent reviewer should ask for a shuffled-alignment ablation, error bars on retrieval, and a rewrite of the 'all phenotypes' claims. If those come back, this could be a solid workshop-tier or modest conference paper; as is, it's a borderline accept.\n\nMy recommendation: engage with it, and if you're the editor, send it out.","headline":"PTACL adds a parameter-free local temporal alignment to ECG-CMR contrastive learning, with small real gains, but overclaims 'all phenotypes' and lacks an ablation showing the physiological correspondence matters.","tokens_in":10234,"tokens_out":3308,"would_cite":true,"duration_ms":33548,"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":"PTACL adds a parameter-free local temporal contrastive loss to global ECG-CMR alignment, improving phenotype retrieval and regression from ECG.","keywords":["contrastive learning","ECG","cardiac MRI","multimodal representation learning","temporal alignment","masked autoencoder","patient retrieval","phenotype regression"],"falsifier":"Train PTACL on the same paired data but with the ECG-to-CMR segment correspondence randomly permuted, or shifted by a fixed phase, while keeping all other settings fixed. If retrieval and regression $R^2$ stay at the same level as with the true alignment, the physiological correspondence is not what drives the reported gains; if performance drops, the temporal alignment is doing real work.","tokens_in":9195,"feed_emoji":"🫀","tokens_out":8873,"duration_ms":82073,"temperature":0.7,"pith_summary":"This paper argues that ECG embeddings can be made to carry functional cardiac information that a resting ECG does not measure directly, if they are trained against paired cardiac MRI (CMR) at two levels at once: the whole patient and the individual time point. The proposed method, PTACL (Patient and Temporal Alignment Contrastive Learning), keeps the standard global contrastive pull between a patient's ECG and CMR representations and adds a local contrastive loss that aligns ECG heartbeat segments with CMR frames at the same cardiac phase. The local loss is parameter-free, so any gain comes from the training signal rather than extra model capacity. Across 27,951 subjects with paired ECG-CMR data, PTACL improves retrieval of patients with similar CMR-derived phenotypes and raises linear-probing $R^2$ for left-ventricular end-diastolic volume from 0.507 to 0.514. If these results hold, a cheap and widely available ECG could transfer a meaningful share of expensive CMR functional assessment, although training still requires paired data.","feed_headline":"Local time alignment helps ECG learn cardiac function from MRI","feed_subtitle":"A parameter-free local loss lifts LVEDV R² from 0.507 to 0.514 and improves patient retrieval","key_machinery":"The load-bearing object is the local temporal contrastive loss between ECG-derived segments and CMR frames, paired with the R-peak-based alignment that creates the correspondence. ECG tokens between consecutive R peaks are extracted per heartbeat, interpolated to a fixed length $T$, and averaged across heartbeats, while CMR local embeddings average spatial tokens within each temporal frame. The local loss uses the alignment matrix $P_{k,t}$, which is one-to-one when $\\sigma=0$ and Gaussian-smoothed when $\\sigma>0$, inside a symmetric contrastive objective over segments and frames. This loss introduces no new learnable weights, so any improvement it produces must come from reshaping the existing encoders' representations rather than from additional capacity. It also supplies the temporal structure that global average pooling discards, which is the missing signal the paper identifies in earlier ECG-CMR contrastive methods.","core_discovery":"The central claim is that fine-grained temporal alignment between ECG and CMR is an effective, learnable-parameter-free way to transfer cardiac functional information into ECG representations. PTACL pre-trains separate masked-autoencoder encoders on each modality, then jointly trains them with a total loss $L = L_{\\text{global}} + \\beta L_{\\text{local}}$. The global term is a standard cross-modal contrastive loss of the InfoNCE form applied to pooled patient-level embeddings. The local term contrasts each of $T$ ECG time segments, extracted between detected R peaks and interpolated to a fixed length, against the corresponding CMR frame using an alignment matrix $P_{k,t}$ that is either hard ($\\sigma=0$) or Gaussian-soft, following a supervised-contrastive formulation. The authors report that adding this local loss improves retrieval precision and ranking across all ten tested CMR-derived phenotypes and improves linear-probing regression $R^2$ for every phenotype, with LVEDV rising from 0.507 to 0.514 and to 0.544 after fine-tuning.","pith_inferences":["A testable implication is that PTACL's gains should shrink on arrhythmic subgroups, because R-peak gating is the source of the temporal correspondence; splitting the test set by arrhythmia burden would test this.","The same local-contrastive recipe should transfer to other paired physiological recordings, such as echocardiography with ECG, since the loss itself is modality-agnostic.","The hard ($\\sigma=0$) versus soft ($\\sigma>0$) alignment choice is a controllable trade-off; per-phenotype tuning of $\\sigma$ could yield further gains.","The result suggests that temporal correspondence, not just shared patient identity, is what lets one modality teach another, which would extend to any paired time series and imaging sequence."],"forward_implications":["PTACL improves retrieval precision and ranking across all ten CMR-derived phenotypes; for example, LVEDV P@1 rises from 0.492 to 0.510.","Adding the local loss improves linear-probing $R^2$ for every phenotype, with LVEDV going from 0.507 to 0.514 and to 0.544 after fine-tuning.","These gains come without additional learnable parameters, so the improvement is attributable to the training signal rather than model capacity.","A single middle short-axis CMR slice suffices to enrich ECG representations, reducing the imaging burden needed for multimodal pre-training.","The pre-trained ECG encoder can be used downstream for patient stratification and as a non-invasive proxy for cardiac function parameters."],"supporting_citations":[{"why":"Provides the masked-autoencoder pre-training scheme used to train both the ECG and CMR encoders before multimodal contrastive learning.","marker":"[7]"},{"why":"Supplies the supervised-contrastive formulation with the alignment matrix that defines the local temporal contrastive loss.","marker":"[10]"},{"why":"Supplies the InfoNCE global contrastive loss used for patient-level ECG-CMR alignment.","marker":"[17]"},{"why":"Supplies the paired ECG-CMR cohort of 27,951 subjects used for training and the 5,991-subject test split.","marker":"[21]"},{"why":"A global-only ECG-CMR contrastive baseline that PTACL extends and is compared against.","marker":"[5]"},{"why":"A multimodal ECG-CMR baseline whose reported ECG regression results PTACL must beat.","marker":"[18]"},{"why":"A multimodal ECG-CMR baseline with the strongest prior ECG regression results, used as a key comparison.","marker":"[25]"},{"why":"Defines the ten CMR-derived cardiac phenotypes used as retrieval and regression evaluation targets.","marker":"[2]"}],"fun_headline_variants":["No new weights: temporal alignment boosts ECG from MRI","Local contrastive loss improves ECG predictions without extra params","Fine-grained ECG-MRI alignment boosts cardiac phenotype retrieval","PTACL: local loss aligns ECG and MRI without added parameters"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes that the detected R peaks and the interpolation map ECG time segments onto CMR frames at the same true cardiac phase; if arrhythmia, heart-rate drift, or gating errors break that correspondence, the local contrastive loss aligns unrelated moments.","fun_headline_variants_meta":{"raw":{"variants":["No new weights: temporal alignment boosts ECG from MRI","Local contrastive loss improves ECG predictions without extra params","Fine-grained ECG-MRI alignment boosts cardiac phenotype retrieval","PTACL: local loss aligns ECG and MRI without added parameters"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000711,"raw_usage":{"total_tokens":3260,"prompt_tokens":1067,"completion_tokens":2193,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":683,"completion_tokens_details":{"reasoning_tokens":2127}},"tokens_in":683,"tokens_out":2193,"duration_ms":14706,"temperature":1.0,"reasoning_tokens":2127,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:24:00.288225+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Train PTACL on the same paired data but with the ECG-to-CMR segment correspondence randomly permuted, or shifted by a fixed phase, while keeping all other settings fixed. If retrieval and regression $R^2$ stay at the same level as with the true alignment, the physiological correspondence is not what drives the reported gains; if performance drops, the temporal alignment is doing real work.","supporting_citations":[{"cited_title":"In: Larochelle, H., et al","cited_arxiv_id":null,"evidence_quote":"Supplies the supervised-contrastive formulation with the alignment matrix that defines the local temporal contrastive loss."},{"cited_title":"In: ICML","cited_arxiv_id":null,"evidence_quote":"Supplies the InfoNCE global contrastive loss used for patient-level ECG-CMR alignment."},{"cited_title":"PLoS Med.12(3), e1001779 (2015)","cited_arxiv_id":null,"evidence_quote":"Supplies the paired ECG-CMR cohort of 27,951 subjects used for training and the 5,991-subject test split."},{"cited_title":"In: MICCAI 2024","cited_arxiv_id":null,"evidence_quote":"A global-only ECG-CMR contrastive baseline that PTACL extends and is compared against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"A multimodal ECG-CMR baseline whose reported ECG regression results PTACL must beat."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"A multimodal ECG-CMR baseline with the strongest prior ECG regression results, used as a key comparison."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the ten CMR-derived cardiac phenotypes used as retrieval and regression evaluation targets."}],"review_version":2}