{"id":"f4cb9e4f-2dc9-482e-a3d1-6ff88136c29b","arxiv_id":"2509.02760","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"In three cadavers, a VR digital twin allowed remote planning and robot execution of 132 needle insertions with a mean off-axis placement error of 5.30 mm.","lead":"Researchers built a virtual reality system that lets pathologists plan and run robotic needle biopsies on cadavers from a separate room. This matters because it points toward a way to collect autopsy tissue without putting medical staff in the same room as potentially infectious bodies.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Off-axis error conflates system accuracy with target displacement from repeated insertions, undermining the central accuracy claim.","rationale":"The reader's weakest assumption identifies exactly the same load-bearing concern: the accuracy metric assumes the target position is unchanged between planning and insertion. The paper's own text acknowledges that targets can be displaced by previous needle insertions, yet the reported error includes all insertions, including repeated ones to the same target. This is not merely a theoretical concern; it directly affects whether the 5.30 mm figure describes system placement accuracy or includes tissue displacement. The 65% histopath confirmation rate is a real-world outcome that is consistent with this confound. I agree with the reader's CONDITIONAL verdict because the concern is significant but potentially addressable by re-analyzing the data. No verdict change is needed; the reader already flagged the issue and recommended a revision. My agreement is 'agree' because the same weakest assumption was identified. The concrete test I propose would settle whether the concern actually lands: if first-insertion error is much lower, the reported metric is biased by repeated insertions, and the accuracy claim would need to be re-stated.","tokens_in":8892,"tokens_out":6541,"duration_ms":73389,"concrete_test":"Recompute the off-axis error using only the first insertion per target (the 46 insertions used for tissue extraction) and compare it to the reported 5.30 mm for all 132 insertions. If the first-insertion error is substantially lower, the reported figure is inflated by target displacement from repeated insertions, and the central accuracy claim must be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central accuracy claim (5.30±3.25 mm off-axis error, Section III.B) is computed by comparing the needle axis annotated in the post-insertion CT with the planned target position from the planning CT. This comparison is only valid if the target's anatomical location is identical in both CT frames. The paper itself states in Section II.F that 'targets can be displaced due to previous needle insertions,' yet all 132 insertions—including the second and third insertions to the same target—are included in the error analysis. The phantom is strapped to the skin, and Eq. (1) assumes a rigid chain from robot base to CT; it cannot capture internal organ motion relative to the skin. Consequently, the reported off-axis error could represent target displacement rather than needle placement error, especially for later insertions. The 65% histopath confirmation rate (Section III.B) is consistent with this confound. The validity of the headline accuracy figure is therefore in question.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a mobile robotic post-mortem biopsy system in which a physician plans needle trajectories in a virtual-reality digital twin and the robot executes the insertions in a physically separate intervention room. The system is evaluated in a usability study with 21 participants across three VR planning scenarios and in a clinical-feasibility study on three cadavers. The authors report 132 needle insertions with a mean off-axis error of 5.30±3.25 mm, a mean target error of 7.35±4.10 mm, and histopathological confirmation of 65% of tissue samples. The central claims are that fully remote VR planning is feasible and usable and that the system achieves clinically acceptable placement accuracy.","tokens_in":9058,"tokens_out":4217,"duration_ms":50430,"significance":"If the accuracy figure is valid, the system is a meaningful step toward reducing infection risk during post-mortem tissue sampling and enabling remote systematic biopsy collection. The paper provides a concrete and largely transparent evaluation protocol: an explicit registration chain (Eq. 1), three separate error measures, a multi-user usability study with physicians, and histopathological follow-up. The disclosure of excluded needles and the acknowledgment that targets can be displaced by prior insertions are also to the authors' credit. However, the central accuracy claim is threatened by the conflation of needle placement error with target motion, and the abstract overstates the histopathological result. The paper's contribution is promising, but the accuracy evidence needs strengthening before the headline number can be accepted at face value.","major_comments":[{"comment":"The headline off-axis error of 5.30±3.25 mm is computed for all 132 insertions, including the second and third insertion to the same target, by comparing the needle axis annotated in the post-insertion CT against the target position from the planning CT. The paper itself states in Section II.F that 'targets can be displaced due to previous needle insertions,' and indeed only tissue samples for the first insertion are extracted to avoid this bias. This creates a direct confound: for later insertions, the deviation between the annotated needle axis and the planned target may reflect target displacement rather than robot needle-placement error. Please report the error measures separately for first insertions only, or otherwise correct for target motion (e.g., by registering post-insertion CT to planning CT and updating target coordinates). Without this, the central accuracy claim is not sup","section":"Section II.F / Section III.B, Eq. (1)"},{"comment":"The abstract states 'Tissue samples were successfully biopsied and histopathologically verified,' which implies that the extracted samples were all confirmed. The results in Section III.B report histopathological confirmation in only 65% of samples, with the majority of misses for small structures such as the coronary artery. This is a substantial understatement of a limitation and should be corrected. Please report the 65% confirmation rate, or an equivalent qualification, in the abstract so that the summary matches the evidence.","section":"Abstract / Section III.B"},{"comment":"Five needles were removed from the study because they did not stay fixed after insertion, and one more was removed because it was touched by the robot EEF. Removing failure cases from the accuracy analysis can bias the reported error downward, especially because the five loose needles are attributed to the low-BMI cadaver (cadaver 2), making the exclusion potentially systematic rather than random. The disclosure is commendable, but the paper should quantify the sensitivity of the reported error to these exclusions, e.g., by reporting the error with the excluded needles included (with reasonable handling of their positions) or by reporting per-cadaver breakdowns that make the effect of the exclusion explicit.","section":"Section III.B (needle exclusion)"}],"minor_comments":[{"comment":"The comparison of the target point error with Neidhardt et al. [9] (7.19±4.22 mm vs. 7.35±4.10 mm) is presented without any statistical test or confidence interval. Please clarify whether this is an informal comparison, and avoid drawing conclusions about equivalence without a quantitative basis.","section":"Discussion, comparison to [9]"},{"comment":"The caption appears truncated: 'In red the learning curve is indicated' and then the sentence about users being tasked is incomplete. Please rewrite for clarity.","section":"Fig. 6 caption"},{"comment":"The text reads 'Gauge 13 biopsy needle'; standard notation is '13 Gauge' or '13G'. Also, please specify the needle diameter and length for reproducibility.","section":"Section II.F"},{"comment":"The calibration time 358±32 s is reported, but it is not clear whether this includes the CT scan and table positioning or only the hand-eye calibration. Please clarify what the reported interval covers.","section":"Section II.B"},{"comment":"The term 'digital twin' is used to describe the VR representation, but no formal definition is given. A brief statement of what aspects of the physical system are mirrored (robot kinematics, CT geometry, skin model) and which are not (e.g., tissue deformation, needle-tissue interaction) would help set expectations.","section":"Section II.C"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the scope of RA-L and the system is interesting, but the accuracy evaluation needs to be revisited before publication. The main risk is the target-displacement confound; if the authors can show that first-insertion-only errors are similar to the reported aggregate, or otherwise separate placement error from tissue motion, the paper would be much stronger. The abstract also needs to be aligned with the 65% histology rate. I do not see a fundamental circularity or fatal flaw, but the load-bearing claim is currently overstated."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short take: this is a real systems paper, not a simulation stunt. The new bit is putting a VR digital twin in the control loop for fully remote post mortem biopsy, with three planning modes and evaluation on actual cadavers. That combination is genuinely new, and the authors document the workflow—hand-eye calibration, registration chain, latency, end-to-end tissue extraction—carefully enough that the system could be replicated. The usability data from 21 users is credible, and the learning-curve numbers go in the right direction. Credit also for reporting the 65% histopathology confirmation rate in the body and for explicitly disclosing the six excluded insertions.\n\nThe main soft spot is whether the headline accuracy number means what it says. Eq. (1) registration and the post-insertion CT error computation assume no tissue shift between planning and insertion. The paper itself says previous insertions can displace targets, and all 132 insertions were included in the error analysis, including later ones on the same target. So part of the reported 5.30 mm off-axis error may be target movement, not needle placement error. That is a legitimate confound, though not fatal: the surface-point error is lower and the histology success rate is consistent with small targets and shallow insertions. The stress-test note is directionally right.\n\nThe other issues are real but minor: six excluded needles (five loose due to low BMI, one touched by the EEF) could bias accuracy in either direction and should be reported both ways; the abstract says samples were successfully biopsied and histopathologically verified without giving the 65% rate, which overstates the result; and the 'alternative to conventional approaches' claim lacks a manual baseline, so it should be framed as feasibility.\n\nI don't think the paper is circular or hiding anything. The comparison to their earlier robotic system [9] is a benchmark, not an input, and the registration chain is standard. The weakest assumption is standard for this kind of cadaver study, but the authors could address it by reporting per-target insertion order or excluding multiple insertions from the accuracy computation.\n\nWho is this for? People working on remote robot control, forensic pathology, or VR surgical planning. It deserves a serious referee; the thesis is clear and the evaluation is honest enough that a revision can fix the flags. I'd engage with it—if I did forensic robotics, I'd cite it.","headline":"A solid integrated feasibility study: remote VR planning plus robotic post mortem biopsy works on cadavers, but the headline accuracy number is softer than it looks because excluded insertions and possible target displacement are both in play.","tokens_in":9581,"tokens_out":1623,"would_cite":true,"duration_ms":18754,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A virtual-reality digital twin lets pathologists plan and supervise robotic post mortem biopsies from a separate room; 132 insertions landed within 5.30 ± 3.25 mm off-axis.","keywords":["virtual reality","digital twin","robotic needle placement","post mortem biopsy","minimally invasive tissue sampling","CT-guided intervention","teleoperated robot","human-robot interaction"],"falsifier":"Compare the planning CT and a post-insertion CT using radio-opaque markers embedded in the phantom and at the target sites: if internal targets shift relative to the strapped phantom, the registered needle-tip error should decrease once that displacement is compensated. If it does not, the 5.30 mm off-axis error is genuine placement accuracy.","tokens_in":8775,"feed_emoji":"🦾","tokens_out":7211,"duration_ms":76000,"temperature":0.7,"pith_summary":"The paper reports a complete remote workflow for post mortem tissue sampling: a pathologist wearing a VR headset plans needle trajectories on CT-derived anatomy in a life-size digital twin of the intervention room, and a lightweight robot in the physically separate room executes the insertions. Across three cadavers of different body types, 132 needle insertions achieved a mean off-axis placement error of 5.30 ± 3.25 mm and a mean surface puncture error of 2.62 ± 1.76 mm, comparable to the authors' earlier robotic system. Histopathological examination confirmed the intended tissue in 65% of 46 extracted samples, with misses concentrated in small targets such as coronary arteries. Eleven forensic pathologists and ten non-physicians completed the planning task after a five-minute introduction; the two interaction modes that let the user grab and orient the needle in 3D were faster and rated more intuitive than a conventional triplanar screen-based mode. If these results hold, the system offers a practical, low-exposure alternative to manual ultrasound-guided PM biopsy, especially for large-scale systematic sampling.","feed_headline":"VR-guided robot biopsies cadavers remotely at 5.3 mm accuracy","feed_subtitle":"Pathologists plan needle paths in a VR twin; the robot executes them from another room. Histology confirmed 65%.","key_machinery":"The load-bearing mechanism is the digital twin's single shared coordinate frame. A chain of rigid transformations—robot base, tracking camera, phantom on the cadaver's thorax, and CT volume—maps a trajectory drawn in VR into robot motion in the intervention room. In the VR scene, a CT-derived skin mesh is overlaid with a color map of the maximum Hounsfield unit along candidate needle paths, so the user immediately sees which insertions would hit bone and whether the robot's reach and joint limits make the trajectory feasible. The three planning scenarios differ only in how the needle path is specified; the underlying registration and feasibility checks are identical.","core_discovery":"The central claim is that a VR digital twin can support fully remote planning and execution of robotic post mortem biopsies with acceptable accuracy and usability. The paper demonstrates this by combining a CT-guided registration chain (robot base to tracking camera to a strapped phantom to the CT volume) with a VR application in which the user sees a life-size skin model, the robot, and CT slices in a shared coordinate frame. Users choose insertion points and needle angles in one of three interaction modes; the system immediately marks which trajectories are robot-feasible by overlaying a maximum-Hounsfield-unit color map on the skin. In the cadaver study, the mean off-axis error was 5.30 ±","pith_inferences":["Because the reported error is computed against the planning CT, part of the 5.30 mm off-axis value may reflect target displacement (e.g., tissue shifting after earlier insertions) rather than pure robot placement error; a marker-based repeat-CT experiment could separate the two.","The 65% histology confirmation rate suggests the limiting factor for small targets is not placement error alone but the biopsy needle and target size; a different needle design or intraprocedural confirmation could plausibly raise the hit rate.","The same digital-twin interaction could be adapted to living patients if respiratory and cardiac motion are added to the registration and planning, but that would require motion compensation the current post mortem setting does not need.","The strong preference for the 3D grab-and-orient modes over the conventional triplanar screen mode hints that pathologists may accept VR planning more readily if the interaction matches how they think about anatomy, not how legacy software displays slices."],"forward_implications":["Remote planning and execution: a physician can run the whole biopsy workflow from a control room, eliminating direct contact with the cadaver and reducing infection risk.","No robotics expertise required: novice and expert users planned feasible trajectories after a five-minute introduction, and planning times dropped 33–53% across repeated insertions.","Systematic sampling becomes practical: because the robot places each needle consistently, the workflow suits large multi-target collections such as national autopsy registries.","Accuracy is comparable to prior robotic PM biopsy systems (target error 7.35 vs 7.19 mm), while surface puncture error is lower (2.62 mm).","Scalability to infectious disease outbreaks: the system is mobile and can be deployed where autopsies are otherwise avoided."],"supporting_citations":[{"why":"Documents contamination of personal protective equipment during COVID-19 autopsies, motivating the need for remote, low-exposure tissue sampling.","marker":"[1]"},{"why":"Presents an earlier fixed ceiling-mounted robotic post mortem biopsy system that the mobile-platform design extends and compares against.","marker":"[8]"},{"why":"Earlier robotic PM tissue sampling system whose 7.19 ± 4.22 mm target error serves as the accuracy baseline for this study.","marker":"[9]"},{"why":"Supplies the QR24 hand-eye calibration algorithm used to determine the robot-base-to-tracking-frame transform.","marker":"[13]"},{"why":"Provides the motion planning and collision-avoidance framework used to check feasibility and execute planned needle trajectories.","marker":"[14]"},{"why":"Reports coronary artery dimensions used to explain why small targets such as the coronary artery are the most frequent histology misses.","marker":"[15]"}],"fun_headline_variants":["VR twin steers robot for remote autopsy biopsies","Pathologists pilot robot via VR for cadaver biopsies","Remote VR control of biopsy robot hits 5.3 mm accuracy","Digital twin VR lets doctors guide biopsy robot remotely","Robotic autopsies planned in VR, executed remotely"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The accuracy numbers assume the cadaver's skin and internal organs stay fixed relative to the tracked phantom between the planning CT and each insertion; if the body shifts under the tension straps, the reported error could partly be target displacement rather than robot placement error.","fun_headline_variants_meta":{"raw":{"variants":["VR twin steers robot for remote autopsy biopsies","Pathologists pilot robot via VR for cadaver biopsies","Remote VR control of biopsy robot hits 5.3 mm accuracy","Digital twin VR lets doctors guide biopsy robot remotely","Robotic autopsies planned in VR, executed remotely"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000374,"raw_usage":{"total_tokens":1826,"prompt_tokens":733,"completion_tokens":1093,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":477,"completion_tokens_details":{"reasoning_tokens":1016}},"tokens_in":477,"tokens_out":1093,"duration_ms":8992,"temperature":1.0,"reasoning_tokens":1016,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T11:24:27.653361+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the planning CT and a post-insertion CT using radio-opaque markers embedded in the phantom and at the target sites: if internal targets shift relative to the strapped phantom, the registered needle-tip error should decrease once that displacement is compensated. If it does not, the 5.30 mm off-axis error is genuine placement accuracy.","supporting_citations":[{"cited_title":"Contamination of personal protective equipment during covid-19 autopsies,","cited_arxiv_id":null,"evidence_quote":"Documents contamination of personal protective equipment during COVID-19 autopsies, motivating the need for remote, low-exposure tissue sampling."},{"cited_title":"Virtobot–a multi-functional robotic system for 3d surface scanning and automatic post mortem biopsy,","cited_arxiv_id":null,"evidence_quote":"Presents an earlier fixed ceiling-mounted robotic post mortem biopsy system that the mobile-platform design extends and compares against."},{"cited_title":"Robotic tissue sampling for safe post-mortem biopsy in infectious corpses,","cited_arxiv_id":null,"evidence_quote":"Earlier robotic PM tissue sampling system whose 7.19 ± 4.22 mm target error serves as the accuracy baseline for this study."},{"cited_title":"Non-orthogonal tool/flange and robot/world calibration,","cited_arxiv_id":null,"evidence_quote":"Supplies the QR24 hand-eye calibration algorithm used to determine the robot-base-to-tracking-frame transform."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the motion planning and collision-avoidance framework used to check feasibility and execute planned needle trajectories."},{"cited_title":"Assessment of the dimensions of coronary arteries for the manifestation of coronary artery disease,","cited_arxiv_id":null,"evidence_quote":"Reports coronary artery dimensions used to explain why small targets such as the coronary artery are the most frequent histology misses."}],"review_version":1}