{"id":"150e7df4-9785-4864-a62d-ddea0845169d","arxiv_id":"2502.07922","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Adding a locally rendered ultrasound preview to haptic model-mediated teleoperation removes the effect of up to 1000 ms round-trip delay on operator effort and completion time in a phantom study.","lead":"A teleoperated ultrasound system gives the remote operator an instant visual preview by re-slicing a previously recorded ultrasound sweep, together with instant touch feedback, so the operator does not depend only on delayed live images. In a 15-person test, operators using this preview finished scans faster and with less mental effort at simulated delays up to one second round trip, which could make delayed tele-ultrasound practical.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Complete compensation is inferred from a non-significant difference without an equivalence test, so the headline claim overstates what n=15 with large variance can support.","rationale":"The reader's verdict is already CONDITIONAL and its rationale explicitly mentions that the complete-compensation claim is \"supported by a null result with 15 subjects and no equivalence test.\" The reader's weakest_assumption, however, is the static-model validity rather than the statistical inference. I agree that the static-model limitation is important and is acknowledged in Section IV, but the more load-bearing concern is that the headline claim is an equivalence claim backed only by a non-significance test. This concern applies even in the idealized static-phantom setting where the pre-acquired sweep is perfectly valid, so it directly attacks the central claim rather than its scope. The statistical fix is straightforward: report confidence intervals or a TOST analysis. Because the paper's contribution is otherwise coherent and the missing analysis is addressable, the conditionality of the verdict is appropriate; no verdict shift is needed. Agreement with the reader is partial because the reader listed the equivalence-test issue in the rationale but did not make it the weakest assumption.","tokens_in":10859,"tokens_out":7016,"duration_ms":66615,"concrete_test":"Obtain or reconstruct the per-subject paired data for total completion time and NASA-TLX effort from Tables I and II. Compute the two-sided 95% confidence interval for the difference (500 ms VH-MMT minus 0 ms MMT) and run a two one-sided tests (TOST) equivalence procedure with a prespecified margin, for example ±20% of the control mean for completion time and ±1 TLX point for effort. If both confidence intervals lie fully within the margins, the complete-compensation claim is supported; if either interval crosses the margin, the abstract should be revised to claim comparability or non-inferiority rather than complete compensation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that VH-MMT \"compensates completely for time delays up to 1000 ms round trip in terms of operator effort and completion time\" rests on the observation in Section III that none of the effort or completion-time values differed significantly between the 0 ms MMT control and the 500 ms one-way VH-MMT condition. This is a null result, not an equivalence result. With 15 subjects and large standard deviations in Tables I and II, a non-significant p-value has low power to detect meaningful differences. The paper reports no confidence intervals, no prespecified equivalence margin, no two one-sided tests (TOST), and no power analysis. The point estimates are consistent with complete compensation, but they are also consistent with no detectable difference or with non-inferiority. The phrase \"compensates completely\" therefore goes beyond the statistical evidence. The claim also generalizes \"up to 1000 ms round trip\" from a single 500 ms one-way delay condition. The static-model limitation flagged by the reader is real but secondary: it confines the result to the idealized phantom setting, whereas the missing equivalence test weakens the claim even within that idealized setting.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes visual-haptic model-mediated teleoperation (VH-MMT) for remote ultrasound, in which a pre-acquired ultrasound sweep is re-sliced in real time on the expert side to provide immediate visual feedback, alongside a haptic virtual fixture reconstructed from an RGB-D point cloud. A prototype is built around a Franka Panda robot and a Touch X haptic device, and an experiment with 15 volunteer operators compares conventional MMT with VH-MMT under 500 ms and 1000 ms one-way delays, with a no-delay conventional MMT control. The reported results show that VH-MMT significantly reduces completion time, mental demand, effort, vessel-following error, and force inconsistency relative to conventional MMT at the same delays. The paper further claims that VH-MMT completely compensates for the 500 ms one-way (1000 ms round-trip) delay in terms of effort and completion time, because those outcomes did not differ significantly from the no-delay condition.","tokens_in":11057,"tokens_out":3887,"duration_ms":35055,"significance":"The core idea is novel and practically motivated: replacing delayed video with a locally re-sliced model of the ultrasound volume directly addresses the visual-haptic asynchrony that prior model-mediated teleoperation systems leave unsolved. The prototype is fully implemented, the system description is detailed enough to be reproduced, and the experimental comparison includes multiple clinically relevant metrics (completion time, workload, vessel centering, force consistency). If the complete-compensation claim survives proper statistical scrutiny, the work would be an important step toward tele-ultrasound over realistic communication links. However, as presented, the strongest claim rests on a non-significant difference rather than a demonstrated equivalence, so the significance of the paper is currently limited by missing statistical evidence. I also credit the authors for transparently stating the static-model limitation and for providing detailed algorithms and calibration equations.","major_comments":[{"comment":"The central claim that VH-MMT 'compensates completely' for the 500 ms one-way delay is not supported by the statistics actually reported. The p-values in Tables I and II compare conventional MMT with VH-MMT at the same delay; there is no test reported that directly compares VH-MMT at 500 ms with the no-delay MMT control. The text states that 'none of the effort or completion time values differed significantly between 0 delay and 500 ms delay with VH-MMT,' but that comparison is not shown. A non-significant difference is also not evidence of equivalence without a prespecified equivalence margin, confidence intervals, or a two one-sided tests (TOST) procedure. Please report the missing pairwise comparison and an appropriate equivalence test before claiming complete compensation.","section":"Section III, Tables I and II"},{"comment":"The statistical methodology is under-specified. The paper does not state which significance test was used (paired t-test, Wilcoxon signed-rank, or other), whether the data satisfied normality assumptions, or how the many p-values in Tables I-IV were corrected for multiple comparisons. Several reported p-values are close to conventional thresholds (e.g., p=0.047 for effort and p=0.07 for sweeping), so the conclusions could depend on these choices. Please provide the test names, effect sizes, confidence intervals, and a multiple-comparison correction or a clear rationale for not applying one.","section":"Section II-C and Section III, Tables I-IV"},{"comment":"The claim of complete compensation 'for time delays up to 1000 ms round trip' is supported by only a single delay condition: 500 ms one-way (approximately 1000 ms round trip). No intermediate or longer delays were tested within the range claimed, and for the 1000 ms one-way (approximately 2000 ms round trip) condition, Tables III and IV show significant remaining differences in vessel-centering error and force consistency. The phrase 'up to 1000 ms round trip' should be restricted to the condition actually tested, or additional delay levels should be included to justify the generality.","section":"Abstract and Section III"}],"minor_comments":[{"comment":"The task description says 'VH-MMT and short (500 ms) communication delay' and 'once for each delay with standard MMT'; this should be rephrased for clarity, e.g., 'VH-MMT with a 500 ms communication delay, VH-MMT with a 1000 ms delay, conventional MMT with each delay, and conventional MMT with no delay as a control.'","section":"Section II-C"},{"comment":"The sentence 'With this visual-haptic MMT concept, several other avenues of further research.' is missing a verb; it should read '...several other avenues of further research exist' or 'are possible.'","section":"Section IV"},{"comment":"The 15 participants are described only as 'volunteers,' not as trained sonographers. The paper should state their level of ultrasound expertise and discuss how this affects generalizability to expert operators.","section":"Section II-C"},{"comment":"The eccentricity metric is described as 'proportional to the applied force'; a citation or a brief derivation would help readers understand why eccentricity is a valid proxy for probe pressure.","section":"Table IV"},{"comment":"The text uses 'VH-MMT' and 'MMT' in close proximity without always clarifying which condition is being discussed (e.g., the paragraph starting 'Interestingly' in Section III); please make the comparisons explicit so the reader is not forced to infer the intended pairings.","section":"Section II-B"}],"recommendation":"major_revision","confidential_remarks":"The paper's main contribution, the visual re-slicing preview, is novel and the implementation is credible. The reviewers should insist on proper statistical treatment of the 'complete compensation' claim, including the direct 0 ms vs 500 ms VH-MMT comparison and an equivalence test. I see no issue with novelty or scope for this venue. The paper is not ready for acceptance in its current statistical form, but the required changes are incremental rather than requiring new experiments in a different setting, so major revision rather than rejection is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper does something genuinely new—it puts a re-sliced, pre-acquired ultrasound volume on the expert console as a delay-free visual preview alongside the usual haptic virtual fixture. Prior MMT for tele-US gave haptics or robot motion preview, not ultrasound content. That is the right idea, and the paper shows it works in a controlled phantom study.\n\nThe engineering is solid and honestly reported. Fifteen operators, randomized order between conditions, a practice scan before testing, and objective metrics (completion time, vessel-centering RMSE, eccentricity as a proxy for pressure) plus TLX scores. Results consistently favor VH-MMT over conventional MMT: shorter completion times, lower mental demand and effort, better centering, more consistent pressure. The static-phantom limitation is stated plainly in the discussion, and the simulated delay queue is a reasonable way to test the concept before a real network.\n\nThe soft spot is the headline claim. 'Compensates completely for time delays up to 1000 ms round trip' rests on a non-significant difference between VH-MMT at 500 ms one-way and the no-delay control. With 15 subjects and large standard deviations, that is a null result, not an equivalence result. There are no confidence intervals, no equivalence margin, no TOST, no power analysis. The point estimates are consistent with complete compensation, but so are they with a meaningful residual effect. The claim also generalizes 'up to 1000 ms round trip' from a single 500 ms one-way condition; the 1000 ms one-way condition shows significant improvement over MMT, not full compensation. The paper's own wording in Section III is more careful than the abstract, but the abstract and discussion push further.\n\nThe static-model limitation is real but secondary, and the paper already flags it. The missing equivalence analysis weakens the claim even in the idealized phantom setting, which is why the abstract should be toned down. This is fixable: report effect sizes with confidence intervals, run an equivalence test for the 500 ms condition against a pre-specified margin, and restrict the 'complete compensation' language to the tested one-way delay.\n\nThe citation pattern looks appropriate; no red flags. This is a proof-of-concept for people working on remote ultrasound and teleoperation with delay. It deserves serious peer review, but the revision needs to tighten the statistics and align the abstract with what the data can support.","headline":"A useful proof-of-concept for visual-haptic model-mediated tele-ultrasound, but the 'complete compensation' claim is stronger than the statistics support.","tokens_in":11627,"tokens_out":1827,"would_cite":true,"duration_ms":16998,"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":"Remote ultrasound operators can work as if a round-trip delay of up to one second were absent if the local system shows them a re-sliced preview of a pre-acquired scan alongside the delayed live image.","keywords":["tele-ultrasound","telerobotic ultrasound","model-mediated teleoperation","time delay","haptic feedback","ultrasound re-slicing","virtual fixture","visual-haptic asynchrony"],"falsifier":"Repeat the five-step phantom scanning task after physically shifting the phantom by about 1 cm or tilting it by about 10 degrees once the sweep has been acquired; if VH-MMT at 500 ms one-way delay no longer matches the zero-delay baseline in completion time and effort, or no longer beats conventional MMT in vessel-following accuracy, the complete-compensation claim fails for moving anatomy. Equivalently, on a human subject the same comparison during normal breathing or with visible bowel gas would test whether the static-model limitation, which the paper itself names as the primary limitation, actually erases the preview's benefit.","tokens_in":10641,"feed_emoji":"🩺","tokens_out":8202,"duration_ms":71106,"temperature":0.7,"pith_summary":"Telerobotic ultrasound is useful for remote care, but network delays break the loop between moving the probe and seeing the image. This paper argues that the fix is to give the operator a local model that is visual as well as haptic: a pre-acquired ultrasound sweep, re-sliced in real time at the haptic device's current pose, so the operator sees an instant preview of what the delayed live image will show. In a study with 15 volunteer operators performing a five-step vessel-scanning task, visual-haptic model-mediated teleoperation made a 500 ms one-way (about 1 s round-trip) delay statistically indistinguishable from no delay in completion time and operator effort, while conventional model-mediated teleoperation did not. At a 1000 ms one-way delay it still significantly improved motion accuracy and force consistency. If the claim holds, an expert could scan a patient through a poor connection with much of the ease and control of being physically present.","feed_headline":"A preview image cancels tele-ultrasound delay up to 1 second","feed_subtitle":"A local visual model restored zero-delay performance at 500 ms one-way lag.","key_machinery":"Model-mediated teleoperation (MMT) represents the remote environment locally so the operator can touch it immediately; the paper's addition is a co-registered visual twin. The central mechanism is a point-cloud virtual fixture, built by compounding RGB-D frames into an octree and fitting a proxy-plane surface for spring-damper forces, plus a pre-acquired ultrasound volume that is re-sliced by GPU in real time at the plane corresponding to the haptic device pose. The correspondence is set by the chain of transforms in Eq. (2), which maps the expert handle through the hand-eye calibration, the robot pose, and the probe calibration so the predicted image and the haptic surface occupy the same space. The re-sliced image is what removes the visual delay, while the delayed live stream and periodic integration of new frames keep the model tied to reality.","core_discovery":"The central discovery, stated on the paper's own terms, is that visual-haptic model-mediated teleoperation compensates completely for time delays up to 1000 ms round trip in operator effort and completion time, and significantly outperforms conventional MMT at longer delays in motion accuracy and force control. The new element is the local visual model: because the re-sliced preview tracks the expert's hand immediately while the virtual fixture tracks it haptically, the operator no longer experiences visual-haptic asynchrony, which the paper identifies as the main residual problem in delayed MMT. The actual delayed ultrasound image remains on display for fine positioning, and incoming frames are integrated back into the sweep to slow model aging. The evidence is a controlled comparison in which 15 volunteers each performed five scans; VH-MMT at 500 ms one-way delay had completion times and task-load scores that were not significantly different from the zero-delay control, and both delayed VH-MMT conditions beat conventional MMT on vessel-centering error and on consistency of applied force, measured through vessel eccentricity.","pith_inferences":["If incoming live frames were used to non-rigidly deform the stored sweep rather than just being appended, the same architecture could tolerate small patient motion and breathing; this is the natural next test because the integration pipeline already exists.","The complete-compensation figure was obtained on a rigid phantom, so on real abdomens the static-model assumption will set the practical ceiling; a clinical evaluation should measure how much translation, rotation, and compression the preview can absorb before it misleads the operator.","A decoupled control variant, in which the follower robot servos to make the live image match the preview the expert is viewing, could in principle extend delay tolerance to non-static anatomy, although the matching problem is hard in real time."],"forward_implications":["An expert performing tele-ultrasound over a link with roughly 1 s round-trip delay can expect zero-delay levels of effort, mental demand, and completion time during the positioning and sweeping phases of a scan, according to the 15-operator study.","Conventional MMT's remaining weakness under delay is visual, not haptic: restoring synchronized visual feedback is what closes the gap, so the delayed image stream is the bottleneck this method removes.","At longer delays (1000 ms one-way), VH-MMT no longer fully compensates but still yields significant improvements in motion accuracy and force control over conventional MMT.","The method is framed as a rough-positioning aid rather than a diagnostic replacement: the expert still uses the delayed real image for the final, precise assessment."],"supporting_citations":[{"why":"Documents that delayed ultrasound video made finding the desired plane impossible in 24% of cases, motivating the visual-preview approach.","marker":"[1]"},{"why":"Applies model-mediated teleoperation to robot-assisted tele-ultrasound, the baseline this paper extends with a visual model.","marker":"[15]"},{"why":"Establishes model-mediated telemanipulation as the conceptual basis for interacting with a local model despite delay.","marker":"[16]"},{"why":"Surveys model-mediated teleoperation and states the visual-haptic asynchrony and near-real-time update problems the paper targets.","marker":"[20]"},{"why":"Shows that delayed visual feedback on haptic manipulation is disruptive even at small delays, quantifying why delay-free haptics alone are insufficient.","marker":"[22]"},{"why":"Provides a prior VR point-cloud preview for delayed haptic teleoperation, which the paper distinguishes from previewing the ultrasound image itself.","marker":"[24]"},{"why":"Supplies the proxy-point method used to render haptic forces from the point-cloud model.","marker":"[32]"}],"fun_headline_variants":["Visual preview erases tele-ultrasound delay up to 1 sec","Haptic-plus-visual model beats delayed tele-ultrasound","Local visual model restores zero-delay tele-ultrasound","Real-time preview cancels 1-second tele-ultrasound lag","Visual-haptic MMT fully compensates 1000 ms delay"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that the pre-recorded ultrasound sweep still matches the patient during the exam: the anatomy must stay still, not shift with breathing or flatten under probe pressure, and the geometric calibration linking the robot, camera, and probe must stay accurate.","fun_headline_variants_meta":{"raw":{"variants":["Visual preview erases tele-ultrasound delay up to 1 sec","Haptic-plus-visual model beats delayed tele-ultrasound","Local visual model restores zero-delay tele-ultrasound","Real-time preview cancels 1-second tele-ultrasound lag","Visual-haptic MMT fully compensates 1000 ms delay"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000193,"raw_usage":{"total_tokens":1355,"prompt_tokens":954,"completion_tokens":401,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":570,"completion_tokens_details":{"reasoning_tokens":307}},"tokens_in":570,"tokens_out":401,"duration_ms":4358,"temperature":1.0,"reasoning_tokens":307,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T11:25:39.340656+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the five-step phantom scanning task after physically shifting the phantom by about 1 cm or tilting it by about 10 degrees once the sweep has been acquired; if VH-MMT at 500 ms one-way delay no longer matches the zero-delay baseline in completion time and effort, or no longer beats conventional MMT in vessel-following accuracy, the complete-compensation claim fails for moving anatomy. Equivalently, on a human subject the same comparison during normal breathing or with visible bowel gas would test whether the static-model limitation, which the paper itself names as the primary limitation, actually erases the preview's benefit.","supporting_citations":[{"cited_title":"Telerobotic ultrasound to provide obstetrical ultra- sound services remotely during the covid-19 pandemic,","cited_arxiv_id":null,"evidence_quote":"Documents that delayed ultrasound video made finding the desired plane impossible in 24% of cases, motivating the visual-preview approach."},{"cited_title":"Robot- assisted teleoperation ultrasound system based on fusion of augmented reality and predictive force,","cited_arxiv_id":null,"evidence_quote":"Applies model-mediated teleoperation to robot-assisted tele-ultrasound, the baseline this paper extends with a visual model."},{"cited_title":"Model-mediated telemanipulation,","cited_arxiv_id":null,"evidence_quote":"Establishes model-mediated telemanipulation as the conceptual basis for interacting with a local model despite delay."},{"cited_title":"Model-mediated teleoperation: Toward stable and transparent teleoperation systems,","cited_arxiv_id":null,"evidence_quote":"Surveys model-mediated teleoperation and states the visual-haptic asynchrony and near-real-time update problems the paper targets."},{"cited_title":"Delayed visual and haptic feedback in a reciprocal tapping task,","cited_arxiv_id":null,"evidence_quote":"Shows that delayed visual feedback on haptic manipulation is disruptive even at small delays, quantifying why delay-free haptics alone are insufficient."},{"cited_title":"Virtual reality-based time-delayed haptic teleoperation using point cloud data,","cited_arxiv_id":null,"evidence_quote":"Provides a prior VR point-cloud preview for delayed haptic teleoperation, which the paper distinguishes from previewing the ultrasound image itself."},{"cited_title":"Proxy method for fast haptic rendering from time varying point clouds,","cited_arxiv_id":null,"evidence_quote":"Supplies the proxy-point method used to render haptic forces from the point-cloud model."}],"review_version":1}