{"id":"b9df6463-ff29-4d7f-b0e2-31ebcd05be8b","arxiv_id":"2501.15311","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A Kalman filter applied on top of U-Net OCT segmentation reduces reported epithelium and Descemet's membrane boundary errors by about 60% and 48% on ex vivo rabbit eyes.","lead":"This paper combines a deep-learning image segmentation network with a Kalman filter to smooth the detected cornea layer boundaries in optical coherence tomography images from eye surgery. The hybrid method reports lower segmentation errors than deep learning alone on ex vivo rabbit eye data, which could improve depth guidance during cornea transplant surgery.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Constant-position Kalman model (F=1, Q=1e-5, R=1) gives a Kalman gain near 3e-3; no temporal-response test supports the 'tracking' claim during needle-induced motion.","rationale":"The reader's weakest_assumption identifies exactly the vulnerable point: the Kalman model is a constant-position random walk with no temporal-response validation. My analysis sharpens this by noting the quantitative consequence: with Q=1e-5 and R=1, the steady-state gain is approximately 3e-3, so the filter's effective memory is hundreds of M-mode lines. This is not a fast tracker; it is a heavy smoother. The paper's strongest evidence is Table 1, which reports average absolute errors on static ex vivo M-mode images. Those numbers are consistent with variance reduction from smoothing and do not demonstrate tracking of real boundary motion. The video mention is anecdotal and lacks quantitative temporal alignment. Because the claim is specifically about tracking during DALK needle insertion, this untested dynamic response is the single most load-bearing gap. I do not see an internal inconsistency in the reported static error reduction, and the method is a conventional, plausible combination; the verdict should remain CONDITIONAL pending a temporal-response evaluation. The manuscript also contains duplicated and broken sentences in the Introduction, but this does not affect the mathematical argument; the main concern is the unsupported dynamic tracking claim.","tokens_in":4622,"tokens_out":3711,"duration_ms":38284,"concrete_test":"Compute the exact steady-state Kalman gain and step response of the scalar filter in Section 2.3 (F=H=1, Q=1e-5, R=1, including the 50-point 70/30 sliding-window averaging). Then, on held-out M-mode sequences containing rapid DM or epithelium displacement (e.g., frames during needle insertion), compare per-frame lag and absolute error of KDH versus deep-learning outputs against manual annotations. If KDH lags the true boundary by more than one pixel (2.61 um) during displacement, or if the reported error reduction is not reproduced on moving segments, the 'more accurate tracking' claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that adding the Kalman filter yields more accurate and stable DM/epithelium boundary tracking, with reported 59.55%/48.15% error reductions (Section 3.2, Table 1). The load-bearing assumption is that the filter tracks real boundary motion rather than merely smoothing static noise. Section 2.3 sets F=1, H=1, Q=1e-5, R=1, i.e., a scalar constant-position random walk with Q/R=1e-5. The implied steady-state Kalman gain is K = sqrt(Q/R) approximately 3e-3, giving a response time constant near 300 M-mode lines; the additional 50-point 70/30 sliding-window average further low-pass filters the estimate. During needle insertion or tissue deformation, the DM and epithelium can move by several pixels over a few lines, but this filter will lag such motion severely. The paper reports only average absolute errors on static M-mode images and provides no temporal-response, step-response, or lag analysis. Therefore the reported error reductions could be an artifact of smoothing jagged deep-learning outputs on quasi-static data, while the 'real-time tracking' claim is unsupported for the dynamic surgical conditions the method is intended for.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Kalman filter/deep-learning hybrid (KDH) approach for tracking the Descemet membrane (DM) and epithelium boundaries in M-mode OCT images acquired during ex vivo DALK procedures. A U-Net produces boundary segmentations, and a scalar Kalman filter (F=1, H=1, Q=1e-5, R=1) with a sliding-window observation smoothing step is applied to the resulting boundary positions. On a test set of 50 patches from 12 rabbit eyes, the reported average absolute errors are 0.36 pixels (epithelium) and 0.28 pixels (DM) for KDH, versus 0.89 and 0.54 pixels for the deep-learning-only baseline, corresponding to 59.55% and 48.15% error reductions. The paper also provides qualitative examples and a video of real-time robotic-assisted needle insertion, and claims improved robustness in dynamic surgical conditions.","tokens_in":4861,"tokens_out":4558,"duration_ms":39710,"significance":"If the reported error reduction is robust and the dynamic tracking behavior is sound, the method could offer a simple, low-cost way to smooth deep-learning segmentation outputs in OCT-guided DALK, potentially improving boundary stability and surgical safety. The approach is easy to implement and builds directly on prior work by the same group. However, the quantitative evidence is limited to a single test set without uncertainty quantification, and the central claim of real-time tracking under motion is not validated by any temporal-response or lag analysis. The paper is an incremental but potentially useful engineering contribution; its practical significance depends on whether the Kalman smoother truly tracks boundary motion rather than merely smoothing static noise.","major_comments":[{"comment":"The Kalman configuration F=1, H=1, Q=1e-5, R=1 defines a constant-position random walk whose implied steady-state gain is K ≈ sqrt(Q/R) ≈ 3.2e-3, giving a response time constant on the order of 1/K ≈ 300 M-mode lines. The additional 70/30 sliding-window observation (introduced in Section 2.3) further low-pass filters the input. During needle insertion or tissue deformation, the DM and epithelium can move by several pixels over a few lines, and this filter configuration would lag such motion severely. The paper claims real-time tracking and robustness 'even under dynamic surgical conditions' (Section 3.2), but all quantitative results are on static M-mode images and no step-response, lag, or temporal-response analysis is provided. Consequently, the reported error reductions could be an artifact of smoothing jagged deep-learning outputs on quasi-static data, and the dynamic tracking claim is unsupported. Please add either (a) a temporal evaluation with known boundary displacement (e.g., simulated steps or annotated dynamic sequences during needle insertion) reporting lag and tracking error as a function of displacement speed, or (b) revise the claims to explicitly limit the method to smoothing of static boundary estimates.","section":"Section 2.3 and Section 3.2"},{"comment":"The performance evaluation reports only point estimates of mean absolute error for a single test set of 50 patches from 12 rabbit eyes, with no error bars, confidence intervals, or significance tests. The claimed reductions of 59.55% and 48.15% are not accompanied by any measure of variability, so it is impossible to assess whether they are statistically reliable or how they vary across eyes or patches. The paper should report per-eye or per-patch error distributions (e.g., standard deviation, box plots), a paired significance test (e.g., Wilcoxon signed-rank or paired t-test), and a statement about whether the 50 test patches are independent (e.g., how many eyes contributed and whether multiple patches from the same eye are treated independently).","section":"Table 1 and Sections 3.1–3.2"},{"comment":"The paper repeatedly claims 'real-time' capability for the KDH approach (e.g., abstract, Section 4) and states that the evaluation emphasized 'inference time efficiency,' yet no latency, throughput, or frame-rate numbers are reported anywhere. Without a measurement of the combined U-Net plus Kalman filter processing time on the stated GPU (NVIDIA GeForce RTX 4070 Ti SUPER) and a comparison to the OCT acquisition rate or the surgical update requirement, the real-time claim is not supported. Please include a timing table or, if the claim is intended only qualitatively, temper the language accordingly.","section":"Section 3.2 and Abstract"}],"minor_comments":[{"comment":"There is a duplicated and garbled passage: 'However, these methods are often challenged by signal noise, motion artifacts, and data inconsistencies, which can lead to jagged or inaccurate segmentation of the DM and epithelium boundary as shown in Figure 1 deep learning approaches like U-Net based approach , were employed to segment corneal layers in OCT M-scan data, enabling autonomous or semi-autonomous needle guidance. However, these methods are often challenged by signal' — this appears to be an editing error and should be rewritten as a single coherent sentence.","section":"Section 1, paragraph 3"},{"comment":"The caption contains a placeholder URL 'http://dx.doi.org/doi.number.goes.here'; please replace it with a working link or remove the URL.","section":"Figure 4 caption"},{"comment":"The Kalman equation notation is underspecified: please define x, z, K, and the covariance matrices explicitly, and state that all quantities are scalars in this application. In particular, explain how the sliding-window weighted average replaces the measurement z_k in Eq. (2) and whether the filter covariance P is updated in the usual way (the update equations for P are not shown).","section":"Section 2.3, equations"},{"comment":"The column headers should indicate units explicitly (e.g., 'Average Epithelium Error (pixel/µm)') and state whether the errors are mean absolute errors averaged over all test patches, over M-mode lines, or per eye. This would improve reproducibility.","section":"Table 1"},{"comment":"Reference [4] is a preprint (arXiv:2410.14577); if it has been published in a peer-reviewed venue, please update the citation. Also consider adding a reference for the Kalman filter itself in addition to the two application references [6,7] currently cited.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is a short engineering contribution that builds directly on the authors' prior work (especially references [4] and [8]). The incremental addition is a Kalman smoother on top of an already-reported U-Net segmentation. While this is a reasonable technical step, the manuscript's claims about real-time dynamic tracking are not supported by the experiments, and the statistical reporting is minimal. The editor may also consider whether the paper's length and novelty meet the journal's standards; if the dynamic-tracking concern is addressed with a proper temporal-response evaluation and uncertainty quantification, I would be willing to see a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful core of this paper is simple: take U-Net segmentations of the DM and epithelium in OCT M-mode images and smooth them with a Kalman filter. As an application of off-the-shelf components to a specific surgical guidance problem, that is legitimate and mildly new—I don't know of another group doing exactly this for DALK. The reported error reductions (59.55% for epithelium, 48.15% for DM) are internally consistent with what a low-pass filter does to jagged segmentation output, and the qualitative figures match the claim.\n\nThe paper's real weakness is the gap between the word \"tracking\" and what was actually measured. The Kalman filter is configured as a constant-position random walk (F=1, H=1, Q=1e-5, R=1), which gives a steady-state gain around 3e-3 and a response time constant on the order of hundreds of M-mode lines. That will seriously lag any fast boundary movement during needle insertion, yet the evaluation is entirely static: average absolute error on 50 test patches. There is no step-response, no lag measurement, no test on a dynamic recording with ground truth. So the headline claim that the method improves \"real-time tracking\" is unsupported; it improves static segmentation smoothness. The stress-test note lands.\n\nOther weaknesses are more ordinary: no error bars or significance tests, small test set from 12 rabbit eyes, hand-set Kalman parameters without sensitivity analysis, no code or data. The writing is also rough, with a visibly broken repeated passage in the Introduction and typos throughout. These are fixable but need attention.\n\nWhat the paper does well is stay honest in the numbers—the table reports the errors, and the direction is plausible. The authors just need to either weaken the tracking claim or back it with temporal experiments.\n\nWho should read this? People working on OCT-guided ophthalmic surgery, especially DALK needle guidance, will find it a reasonable incremental data point. It is not a breakthrough; it is a small engineering step. I would send it to peer review because the idea is sensible and the results, while not rigorously characterized, are directionally positive. But I would ask for temporal-response analysis and proper uncertainty quantification before accepting.","headline":"A modest engineering application that plausibly reduces segmentation error on static OCT M-mode data, but the 'tracking' claim outruns the evidence without any temporal-response or lag analysis.","tokens_in":5423,"tokens_out":1313,"would_cite":false,"duration_ms":14106,"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":"Adding a Kalman filter to deep-learning segmentation cuts average epithelium boundary error by 59.55% and Descemet's membrane error by 48.15% in ex vivo OCT M-mode tracking for DALK.","keywords":["Kalman filter","cornea transplant","OCT","DALK","image segmentation","deep learning","M-mode OCT","boundary tracking"],"falsifier":"Run the proposed KDH pipeline on synthetic M-mode OCT sequences whose boundary moves as a known step or ramp, and compare the tracked depth against the known ground truth frame by frame. If the tracking error on fast-moving boundaries is much larger than the 0.28–0.36 pixel errors reported for the static ex vivo test set, the constant-position Kalman model is the reason.","tokens_in":4418,"feed_emoji":"👁️","tokens_out":10429,"duration_ms":86205,"temperature":0.7,"pith_summary":"This paper tries to show that a Kalman filter wrapped around a deep-learning segmentation network can make real-time optical coherence tomography (OCT) boundary tracking reliable enough for use during deep anterior lamellar keratoplasty (DALK), a partial-thickness cornea transplant. In M-mode OCT data, repeated one-dimensional depth scans over time, the hybrid reduces the average epithelium boundary error by 59.55% and the average Descemet's membrane (DM) error by 48.15% compared with deep-learning-only output on 50 ex vivo test images. The improvement matters because DALK needle insertion requires micron-accurate depth feedback: if the needle perforates DM, the operation must convert to a full-thickness transplant. The measured gains come from smoothing jagged deep-learning segmentations, not from changing the imaging hardware.","feed_headline":"Kalman filter cuts cornea-layer tracking error by up to 60%","feed_subtitle":"Smoothing AI segmentation makes real-time tracking of corneal layers steadier for partial-thickness transplants.","key_machinery":"The load-bearing mechanism is the Kalman filter integration: a one-dimensional state-space smoother that takes the U-Net's per-frame boundary detection as the observation $z_k$ and produces a smoothed boundary state $\\hat{x}_{k|k}$. The prediction and update steps are $\\hat{x}_{k|k-1}=F\\hat{x}_{k-1|k-1}$ and $\\hat{x}_{k|k}=\\hat{x}_{k|k-1}+K_k(z_k-H\\hat{x}_{k|k-1})$, with the paper's parameters $F=1$, $H=1$, $Q=10^{-5}$, $R=1$. That parameter choice lets the filter reject isolated jagged spikes while otherwise trusting the observations, and the adaptive sliding-window observation, 70% weight on the most recent 50 points and 30% on the preceding 50, turns the deep-learning boundary into a continuous, stable trace.","core_discovery":"On its own terms, the paper's central discovery is that feeding a deep-learning network's boundary predictions into a Kalman filter with state transition $F=1$, observation $H=1$, process noise $Q=10^{-5}$, and observation noise $R=1$ substantially improves automatic tracking of the epithelium and Descemet's membrane in OCT M-mode scans from rabbit eyes undergoing Big Bubble DALK. The filter treats each boundary depth as a slowly time-varying random walk, uses the network prediction as a noisy observation, and outputs a smoothed depth trace. In the reported ex vivo comparison, the average epithelium error drops from 0.89 pixel (2.3229 µm) to 0.36 pixel (0.9396 µm), and the average DM error drops from 0.54 pixel (1.4094 µm) to 0.28 pixel (0.7308 µm), reductions of 59.55% and 48.15%, respectively. The authors also show qualitative examples where the filtered trace stays smooth under signal loss, motion artifacts, and jagged segmentation, and they demonstrate the pipeline in real time on an eye-mounted robotic OCT-guided needle system.","pith_inferences":["A natural extension is to replace the constant-position random walk ($F=1$, small $Q$) with a constant-velocity or maneuver-aware model, since the paper does not analyze how much the filter would lag a fast-moving boundary during needle insertion.","Because the reported error reductions are averaged over 50 test images, partitioning those images by signal quality would reveal whether the gain is entirely noise suppression; on high-SNR cases the filtered and unfiltered results already align, according to the paper's qualitative examples.","The 70/30 sliding-window weighting is effectively a hand-tuned observation model; sweeping the $Q/R$ ratio and window weights on the existing test set would quantify how robust the 59.55% and 48.15% reductions are.","The same Kalman wrapper is directly transferable to other OCT-guided needle procedures that track a single tissue boundary in real time, such as subretinal injection."],"forward_implications":["The average epithelium tracking error is reduced from 0.89 to 0.36 pixel (2.3229 µm to 0.9396 µm).","The average DM tracking error is reduced from 0.54 to 0.28 pixel (1.4094 µm to 0.7308 µm).","The filtered traces remain smooth and stable in low-signal, motion-artifact, and signal-loss cases where the deep-learning-only traces become jagged.","The pipeline can be applied during real-time robotic-assisted DALK needle insertion, as demonstrated in the sample video.","The resulting depth feedback is steadier than deep-learning-only segmentations, which is the property the authors argue is needed for surgical safety during Big Bubble DALK."],"supporting_citations":[{"why":"Supplies the clinical comparison context and the manual ground-truth annotation practice for DM and epithelium boundaries.","marker":"[1]"},{"why":"Defines the Big Bubble needle insertion technique whose safety depends on avoiding DM perforation.","marker":"[3]"},{"why":"Provides the eye-mounted robotic OCT-guided needle system used in the real-time demonstration.","marker":"[4]"},{"why":"Provides the U-Net segmentation architecture whose jagged boundary output is the baseline the Kalman filter smooths.","marker":"[5]"},{"why":"Supplies the prior deep-learning M-mode OCT segmentation method and the rabbit-eye dataset used for training and testing.","marker":"[8]"},{"why":"Contributes to the ex vivo rabbit-eye dataset and the needle-detection context for the M-mode images.","marker":"[9]"}],"fun_headline_variants":["Kalman filter cuts OCT cornea tracking error 60% for DALK","Smoother cornea layer tracking: Kalman filter reduces error 60%","AI and Kalman filter steady OCT boundaries for cornea surgery","DALK depth sensing improved: Kalman filter trims tracking error"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that each boundary is almost stationary between frames, so the filter can treat any difference as noise; if the cornea moves quickly during needle insertion, the smoothed trace will lag behind the true boundary.","fun_headline_variants_meta":{"raw":{"variants":["Kalman filter cuts OCT cornea tracking error 60% for DALK","Smoother cornea layer tracking: Kalman filter reduces error 60%","AI and Kalman filter steady OCT boundaries for cornea surgery","DALK depth sensing improved: Kalman filter trims tracking error"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000881,"raw_usage":{"total_tokens":3841,"prompt_tokens":1011,"completion_tokens":2830,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":627,"completion_tokens_details":{"reasoning_tokens":2754}},"tokens_in":627,"tokens_out":2830,"duration_ms":17987,"temperature":1.0,"reasoning_tokens":2754,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T14:23:31.706727+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the proposed KDH pipeline on synthetic M-mode OCT sequences whose boundary moves as a known step or ramp, and compare the tracked depth against the known ground truth frame by frame. If the tracking error on fast-moving boundaries is much larger than the 0.28–0.36 pixel errors reported for the static ex vivo test set, the constant-position Kalman model is the reason.","supporting_citations":[{"cited_title":"Comparison of vertical cannula insertion techniques for big bubble deep anterior lamellar keratoplasty,","cited_arxiv_id":null,"evidence_quote":"Supplies the clinical comparison context and the manual ground-truth annotation practice for DM and epithelium boundaries."},{"cited_title":"Novel vertical big bubble needle approach vs. the standard approach for deep anterior lamellar keratoplasty (DALK),","cited_arxiv_id":null,"evidence_quote":"Defines the Big Bubble needle insertion technique whose safety depends on avoiding DM perforation."},{"cited_title":"Common -path optical coherence tomography guided vertical pneumodissection for DALK,","cited_arxiv_id":null,"evidence_quote":"Provides the U-Net segmentation architecture whose jagged boundary output is the baseline the Kalman filter smooths."},{"cited_title":"Topology-based deep-learning segmentation method for deep anterior lamellar keratoplasty (DALK) surgical guidance using M-mode OCT data","cited_arxiv_id":"2501.04735","evidence_quote":"Supplies the prior deep-learning M-mode OCT segmentation method and the rabbit-eye dataset used for training and testing."},{"cited_title":"The neural networks -based needle detection for medical retinal surgery,","cited_arxiv_id":null,"evidence_quote":"Contributes to the ex vivo rabbit-eye dataset and the needle-detection context for the M-mode images."}],"review_version":1}