{"id":"7afa7458-8883-4bfe-8e9d-e8bd0bd8cf64","arxiv_id":"2506.21886","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Leadfield-based Pareto optimization predicts that transcutaneous interference spinal cord stimulation can increase the spinal cord-to-skin electric field ratio by over 10-fold compared to conventional transcutaneous stimulation.","lead":"This computational study combines temporal interference with transcutaneous spinal cord stimulation and uses Pareto optimization to find electrode montages that concentrate the electric field in the spinal cord while reducing it in the skin. The model predicts that optimized montages reduce peak skin electric field by more than 20-fold compared to conventional stimulation, suggesting a more comfortable path to non-invasive spinal cord stimulation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 20-fold skin-intensity reduction is computed from the envelope E_AM, not from the actual carrier-frequency field; if cutaneous afferents respond to the carrier or to peak instantaneous fields, the optimized montage's real off-target exposure may be much higher than reported.","rationale":"The reader's weakest-assumption analysis identifies exactly the condition on which the paper's central claim hinges: the envelope-extraction model of temporal interference. My independent read of the manuscript reinforces this with an additional internal tension. In the Introduction, the authors state that superficial areas are predominantly influenced by high-frequency components, which would produce no activation or conduction block on cutaneous afferents (citing [10]); yet in Section 2.4 and in all results, they use E_AM in the skin as the off-target metric to be minimized. If the high-frequency carrier is what the skin actually experiences, then minimizing E_AM in skin is the wrong objective; if the envelope is what drives skin activation, then the Introduction's claim about high-frequency dominance is inconsistent. Either way, the quantitative comparison in Table 2 does not establish the claimed reduction in off-target skin intensity as a physical or clinical benefit. The mathematical form of Eq. (3) makes this concrete: when one field dominates, E_AM ≈ 2|E2| while the physical peak is |E1|+|E2|, so the optimized montages could have large carrier-frequency fields at the skin that the E_AM metric simply ignores. Budde et al. [14] provides direct experimental evidence that peripheral nerve TI stimulation is not driven by envelope extraction, so the assumption is not merely outside current consensus; it is contradicted by a study the paper itself cites. I do not see a fatal internal error in the leadfield computation, Pareto optimization, or the robustness analysis: the six-trial variability is reported, the 44-grid refinement is a reasonable heuristic, and the linear scaling argument for the '5-fold increase' claim is arithmetically consistent if E_AM is accepted. But because the central feasibility claim is entirely contingent on the envelope model, the computational results as presented are not sufficient to support the conclusion that tISCS selectively stimulates spinal cord circuits with greatly reduced skin exposure. The appropriate disposition remains conditional: the paper is a useful computational foundation, but it should be published only with the envelope-model assumption flagged as load-bearing and with a neural-activation-based sensitivity analysis or an explicit statement that the results are predictions of the envelope model, not of neural response. This matches the reader's CONDITIONAL verdict, so no change to the verdict is needed.","tokens_in":11781,"tokens_out":5539,"duration_ms":63208,"concrete_test":"Using the leadfield matrix, recompute for the optimized tISCS montage in Table 2 the peak instantaneous skin electric field max_t |E1(r)cos(2πf1t) + E2(r)cos(2πf2t)| with carrier frequencies in the kHz range, and compare it with the reported skin E_AM value (6.32 V/m) and with the tSCS skin value (148.3 V/m). If the peak carrier field exceeds E_AM by more than a factor of 10, or is comparable to the tSCS value, then the 20-fold skin-reduction claim is an artifact of the envelope metric and the central conclusion is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claims (≥20× skin reduction, ≥10× SC/skin ratio, 5× intensity headroom) are all evaluated using the amplitude-modulated envelope E_AM defined in Eq. (3), and the load-bearing assumption is that spinal cord and skin/muscle activation both follow this envelope. That assumption is weak in two concrete ways. First, E_AM can be much smaller than the actual instantaneous field: when one electrode pair dominates at a skin location (|E1| ≫ |E2| and angle < 90°), Eq. (3) gives E_AM ≈ 2|E2|, whereas the peak carrier-frequency field is |E1| + |E2|. Thus the optimizer minimizes the envelope, not the physical field experienced by skin, so the reported 20.5-fold reduction in skin intensity may be largely a metric artifact. Second, the manuscript itself cites [10] for the view that superficial areas are predominantly influenced by high-frequency components, and it cites Budde et al. [14], which demonstrated that temporal interference in peripheral nerves is not driven by envelope extraction. If cutaneous nociceptors or sensory fibers are activated by the carrier through nonlinear membrane processes, the optimized tISCS value in Table 2 (skin E_AM 6.32 V/m vs. 148.3 V/m for tSCS) does not imply a corresponding reduction in skin activation or pain. The paper contains no neural activation model, so the entire focality advantage rests on this contested envelope-extraction mechanism, which the paper's own cited literature undermines.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents a computational feasibility study of transcutaneous interferential spinal cord stimulation (tISCS). A finite element model of the lower thorax at T11 is combined with a leadfield matrix of 162 electrode positions, allowing rapid evaluation of 10,000 random two-pair montages. Using the amplitude-modulated envelope E_AM of the vector sum of two kilohertz electric fields defined in Eq. (3), the authors perform Pareto optimization over metrics consisting of the maximum E_AM in the spinal cord and the ratio of spinal cord E_AM to skin or muscle E_AM. They report that optimized tISCS reduces peak skin E_AM by 20.5- to 23.5-fold relative to 2-tSCS and tSCS, increases the spinal cord-to-skin E_AM ratio by roughly 10-fold or more, and permits about a fivefold increase in spinal cord stimulation intensity without exceeding the skin exposure of conventional tSCS. A six-trial random-subset robustness check is included, and a reduced electrode grid is derived from a relevant electrode map. The authors conclude that tISCS is a promising non-invasive approach with substantially reduced off-target effects.","tokens_in":12026,"tokens_out":6904,"duration_ms":66996,"significance":"If the envelope-based activation assumption were valid, the leadfield/Pareto workflow would be a useful and efficient design tool for tISCS, and the reported improvements in simulated field ratios would be practically significant. The study's strengths include the use of a standard FEM approach, the leadfield matrix acceleration, and the six-trial robustness check, which supports internal consistency. However, the central quantitative claims rest entirely on the amplitude-modulated envelope E_AM as the neural drive for both spinal cord and superficial tissues, an assumption that the manuscript's own cited references contest. Without a neural activation model or carrier-frequency field metrics, the reported >20-fold skin reduction and >10-fold ratio improvements are predictions about a field envelope, not about skin perception, pain, or spinal cord activation. The paper is therefore a useful computational feasibility demonstration, but its translational claims are not yet supported.","major_comments":[{"comment":"The entire optimization and evaluation are built on E_AM as the neural drive for both spinal cord and skin/muscle. The manuscript itself cites reference [10] for the view that superficial areas are predominantly influenced by high-frequency components, and reference [14] (Budde et al.) for the finding that temporal interference in peripheral nerves is not driven by envelope extraction. With no neural activation model included, the reported reductions in skin E_AM do not establish reduced skin perception, nociceptor activation, or pain. Please re-evaluate the optimized montage with a biophysical nerve model for cutaneous afferents, or at minimum report the peak instantaneous carrier-frequency field magnitudes, and clearly reframe the claims as predictions about the amplitude-modulated envelope rather than about neural activation.","section":"Section 2.3, Eq. (3); Section 4"},{"comment":"The optimization objectives are exactly the metrics used for evaluation: maximizing E_AM(SC) and the ratio E_AM(SC)/E_AM(skin) (or muscle), with Table 2 then reporting those same quantities. The comparison against the fixed conventional tSCS montage provides some independent grounding, but the headline improvements (e.g., 17.0-fold ratio increase over 2-tSCS and 11.3-fold over tSCS) are selected-for outcomes, not out-of-sample predictions. Please report the distribution of the ratios across all 10,000 random montages, the rank of the selected montage within that distribution, and ideally evaluate the optimized montage on metrics not used in the optimization, such as neural activation predictions.","section":"Section 2.4; Table 2"},{"comment":"The E_AM metric can be much smaller than the actual instantaneous carrier-frequency field. When one electrode pair dominates at a skin location (|E1| >> |E2| and angle < 90 degrees), Eq. (3) gives E_AM approximately equal to 2|E2|, whereas the peak instantaneous field magnitude is approximately |E1| + |E2|. The optimizer therefore minimizes the envelope depth, not the physical field experienced by skin, so the reported 23.5-fold reduction in skin E_AM in Table 2 may be partly a metric artifact. Please quantify the peak instantaneous fields |E1 + E2(t)| over the carrier cycle for the optimized montage and report whether the skin-intensity conclusions survive under that metric.","section":"Section 2.3, Eq. (3); Table 2"}],"minor_comments":[{"comment":"The Laplace equation is numbered (2) although no equation (1) appears in the text; the numbering should be corrected.","section":"Section 2.3"},{"comment":"The definition of the muscle focality ratio has a typo: the text writes max|E_AM(Skin)| in the denominator for the muscle ratio, which should be max|E_AM(Muscle)|.","section":"Section 2.4"},{"comment":"The text refers to the angle alpha between |E1| and |E2|; since these are magnitudes, the angle should be described as the angle between the vectors E1 and E2.","section":"Section 2.3, Eq. (3)"},{"comment":"The Relevant Electrode Map depends on a score threshold of 0.5 and on ad-hoc scoring weights (one point for Pareto-front montages, two points for the best montage), but no sensitivity analysis is provided for these choices; varying them could change the reduced 44-electrode grid.","section":"Section 2.5, Section 3.2"},{"comment":"The claim that tISCS 'still achieves the minimum electric field threshold for neuromodulation (0.35 V/m)' citing reference [42] is an extrapolation from transcranial alternating current stimulation thresholds to the spinal cord and to interferential envelope fields; this threshold should be justified with spinal-cord-specific data or removed.","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within scope for a computational neuromodulation journal. The FEM and leadfield implementation appear sound, and the robustness check is a positive feature. The key risk is that the headline factors hinge on the contested envelope-extraction mechanism; if the authors add carrier-level field metrics or a neural activation model, the paper would be substantially stronger. I do not see this as a reject, because the computational workflow is a valid contribution, but the central claims need to be re-scoped or re-supported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nThe headline on arXiv:2506.21886 is a genuinely new idea—applying temporal interference to transcutaneous spinal cord stimulation—and the paper supports it with a clean computational pipeline. Leadfield matrices, Pareto fronts, and a \"Relevant Electrode Map\" to cut the search grid are all executed competently, with a six-trial robustness check that gives the montage search internal consistency. The FEM model is standard and the computational speed-up is real. Credit where due: this is the first computational demonstration of tISCS, and the pipeline itself is reusable.\n\nThe problem is where the claims outrun the model. Every quantitative headline—the 20-fold skin reduction, the 10-fold ratio gain, the 5-fold intensity headroom—is computed from the amplitude-modulated envelope E_AM in Eq. (3). The optimizer minimizes that envelope, not the peak carrier-frequency field. In skin regions where one electrode pair dominates, Eq. (3) yields E_AM ≈ 2|E2| while the physical field is |E1|+|E2|. So the dramatic skin-intensity reduction is partly a metric artifact. The authors cite Mirzakhalili et al. [10] and Budde et al. [14]—the latter of which explicitly concludes peripheral-nerve TI is not envelope extraction—but they never reconcile those results with their own model. There is no neural activation model, peak E_AM is used as a proxy for pain, and a tACS threshold is borrowed for the spinal cord without support.\n\nI don't think this is a takedown; the paper is honest about conductivity and quasi-static limitations, and the fixed tSCS comparison gives the results some grounding. But the central premise—that spinal and cutaneous fibers both respond to the low-frequency envelope—is exactly what the paper's own references challenge.\n\nWho gets value? Computational neuromodulation researchers working on TI methods, and clinicians wanting a foundation for future experimental design. As a methods paper, it deserves engagement; as evidence for a clinical effect, it must be read with the envelope caveat front and center.\n\nRecommendation: yes, send to peer review. The revision should force a confrontation with the envelope debate, add a sensitivity analysis based on the instantaneous peak field rather than E_AM, and make the leadfield code and model available. If the envelope premise collapses, the tISCS advantage is unproven.\n\nBest,\n\n[Your name]","headline":"A clean computational pipeline for a genuinely new idea, but the headline focality numbers are envelope-metric artifacts unless neural activation follows the contested envelope model.","tokens_in":12618,"tokens_out":3276,"would_cite":false,"duration_ms":33319,"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":"Optimized two-tone electrode montages can cut off-target skin fields in spinal cord stimulation by more than twentyfold, the paper argues.","keywords":["transcutaneous spinal cord stimulation","temporal interference","electrode montage optimization","leadfield matrix","Pareto optimization","finite element model","focality","electrical stimulation"],"falsifier":"Replace the $E_{AM}$ objective with a cable-model simulation of the same T11 model in which dorsal-column fibers and skin afferents are driven by the complete two-carrier waveform, and compare recruitment thresholds for the optimized montage; if skin afferents reach threshold at or below the current where dorsal columns fire, or if the twentyfold skin reduction disappears, the central claim fails. In an animal or human surrogate, recording dorsal-column evoked potentials and skin nerve responses across graded currents with the optimized montage would settle the same question.","tokens_in":11528,"feed_emoji":"⚡","tokens_out":9474,"duration_ms":100064,"temperature":0.7,"pith_summary":"This paper tries to establish that transcutaneous interference spinal cord stimulation (tISCS) can deliver focused, comfortable stimulation to the spinal cord by canceling the electric field in the skin and letting a low-frequency envelope dominate at depth. Using a finite-element model of the lower thorax and a leadfield-based Pareto search over electrode montages, the authors report that an optimized two-pair electrode montage cuts the peak skin field by more than twentyfold compared with conventional transcutaneous spinal cord stimulation (tSCS), and raises the spinal-cord-to-skin field ratio by about an order of magnitude. If the model is right, tISCS could let clinicians apply roughly five times more current to the spinal cord before skin exposure reaches tSCS levels, reducing pain while preserving or improving target engagement. That matters because tSCS is currently limited by skin and muscle activation that makes higher-intensity stimulation uncomfortable.","feed_headline":"Interference montage cuts skin fields 20-fold in spinal stimulation","feed_subtitle":"A 10-fold better spinal-to-skin ratio means stronger stimulation can be delivered with less skin pain.","key_machinery":"The load-bearing machinery is the leadfield matrix $\\boldsymbol{A}_n$, which maps the current injected at electrode $n$ to the electric field at every mesh node, together with the amplitude-modulated envelope $E_{AM}$ defined by Eq. (3). The envelope is computed from the two carrier fields' magnitudes and the angle between them; the paper assumes that this envelope, not the carriers, drives spinal fibers. Once the leadfield matrix is precomputed (161 simulations, about 30 hours), any two-pair montage's envelope field costs about 80 seconds to evaluate, enabling a Pareto search over 10,000 random montages on a front that jointly maximizes $\\max|E_{AM}|$ in the spinal cord and its ratio to skin and muscle. The 'Relevant Electrode Map' then prunes the 162-position grid to 44 dorsal positions, which the paper shows improves the spinal-cord-intensity versus focality trade-off.","core_discovery":"The paper's central claim, stated on its own terms, is that temporal interference—superimposing two kilohertz currents from separate electrode pairs—creates a region inside the spinal cord where the amplitude-modulated envelope $E_{AM}$ is the effective neural stimulus, while superficial tissue experiences mostly the high-frequency carriers and is therefore less likely to be activated. To test this, the authors built a 12-tissue finite-element thorax model at the T11 level, precomputed a leadfield matrix for 162 electrode positions, and randomly sampled 10,000 two-pair montages; a Pareto front then picked montages maximizing spinal-cord envelope field while minimizing skin and muscle fields. The optimized montage reduced peak skin field from 148.3 V/m (tSCS) to 6.3 V/m, and raised the spinal-cord-to-skin ratio from 0.03 to 0.34 (about elevenfold) at the same total current; against a matched two-pair tSCS the skin reduction was about 20.5-fold and the ratio gain seventeenfold. A second optimization on a 44-position dorsal grid improved the spinal-cord intensity-to-off-target trade-off further, producing a spinal-cord field of 2.51 V/m with a spinal-cord-to-skin ratio of 0.39. The paper presents this as the first computational demonstration that tISCS is feasible and can outperform tSCS on focality.","pith_inferences":["Editorial extension: the fivefold current headroom is contingent on the envelope-extraction assumption; the paper cites [14] showing peripheral nerve temporal interference is not driven by envelope extraction, so a cable-model comparison between $E_{AM}$-based and full-waveform-based montage rankings would tell how much of the claimed advantage survives.","Editorial extension: a direct psychophysical experiment—ramping current with optimized tISCS versus tSCS in healthy volunteers and asking when skin sensation becomes painful—would test the twentyfold reduction and fivefold headroom without needing invasive spinal recordings.","Editorial extension: the same leadfield-plus-Pareto-plus-relevant-map recipe should transfer to other spinal levels or targets, but the relevant electrode map would need to be recomputed for each anatomy rather than reused from the T11 dorsal grid."],"forward_implications":["At the same 2.5 mA total current, the optimized tISCS montage keeps the spinal cord near the 0.35 V/m threshold while cutting peak skin field to less than one-twentieth of tSCS exposure.","Because the spinal-cord-to-skin ratio is about ten times higher, users could in principle increase current by roughly fivefold before off-target skin stimulation equals what tSCS produces, giving room for stronger spinal modulation.","Skin, not muscle, is the binding constraint in the optimized Pareto front; alignment with lower cutaneous activation thresholds means future tISCS protocols should track skin discomfort as the primary safety endpoint.","The leadfield/Pareto workflow makes montage optimization practical for high-resolution spinal models, reducing per-montage solve time from hours to about 80 seconds and enabling 10,000 configurations to be screened.","Reducing the electrode search space to the dorsal 'Relevant Electrode Map' improves rather than degrades the Pareto trade-off, supporting a focused, reproducible electrode-placement strategy."],"supporting_citations":[{"why":"Supplies the amplitude-modulated envelope formula and the computational precedent for temporal interference fields used in Eq. (3).","marker":"[18]"},{"why":"Introduced temporal interference as a deep-brain stimulation mechanism, the conceptual basis tISCS transfers to the spinal cord.","marker":"[38]"},{"why":"Provides the activation-function rationale for using the z-component of the spinal cord field as the fiber drive.","marker":"[37]"},{"why":"Reports the roughly 2.5 hours per montage cost that motivates the leadfield shortcut.","marker":"[19]"},{"why":"Established optimized multi-electrode stimulation via a leadfield matrix, the basis of Eq. (4).","marker":"[20]"},{"why":"Gives the 0.35 V/m minimum effective dose used to claim tISCS remains above threshold.","marker":"[42]"},{"why":"Argues temporal interference in peripheral nerves is not driven by envelope extraction, the main challenge to the study's activation assumption.","marker":"[14]"},{"why":"Supplies the lower thoracic (T11) anatomical geometry on which the finite-element model is based.","marker":"[24]"}],"fun_headline_variants":["Interference spinal stimulation slashes skin fields 20-fold","Temporal interference improves spinal cord stimulation focality 10x","Pareto-optimized interference montage cuts skin exposure 20-fold","Interference currents sharpen spinal targeting while sparing skin"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the premise that spinal fibers are driven by the slow amplitude envelope of two interfering kilohertz fields, while superficial nerves are not; if high-frequency carriers activate skin or muscle directly, the reported focality and comfort gains would shrink or disappear.","fun_headline_variants_meta":{"raw":{"variants":["Interference spinal stimulation slashes skin fields 20-fold","Temporal interference improves spinal cord stimulation focality 10x","Pareto-optimized interference montage cuts skin exposure 20-fold","Interference currents sharpen spinal targeting while sparing skin"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000561,"raw_usage":{"total_tokens":2766,"prompt_tokens":1145,"completion_tokens":1621,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":761,"completion_tokens_details":{"reasoning_tokens":1551}},"tokens_in":761,"tokens_out":1621,"duration_ms":13388,"temperature":1.0,"reasoning_tokens":1551,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T22:17:08.724405+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Replace the $E_{AM}$ objective with a cable-model simulation of the same T11 model in which dorsal-column fibers and skin afferents are driven by the complete two-carrier waveform, and compare recruitment thresholds for the optimized montage; if skin afferents reach threshold at or below the current where dorsal columns fire, or if the twentyfold skin reduction disappears, the central claim fails. In an animal or human surrogate, recording dorsal-column evoked potentials and skin nerve responses across graded currents with the optimized montage would settle the same question.","supporting_citations":[{"cited_title":"Botzanowski et al., ‘Noninvasive Stimulation of Peripheral Nerves using Temporally‐Interfering Electrical Fields’, Adv Healthc Mater, vol","cited_arxiv_id":null,"evidence_quote":"Supplies the amplitude-modulated envelope formula and the computational precedent for temporal interference fields used in Eq. (3)."},{"cited_title":"Huang and L","cited_arxiv_id":null,"evidence_quote":"Provides the activation-function rationale for using the z-component of the spinal cord field as the fiber drive."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Established optimized multi-electrode stimulation via a leadfield matrix, the basis of Eq. (4)."},{"cited_title":"Rattay, ‘Analysis of models for extracellular fiber stimulation’, IEEE Trans Biomed Eng, vol","cited_arxiv_id":null,"evidence_quote":"Gives the 0.35 V/m minimum effective dose used to claim tISCS remains above threshold."}],"review_version":1}