{"id":"99801318-6d18-4168-a297-fb0d8b2b13d1","arxiv_id":"2607.07240","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":3,"one_line_summary":"EP-SAM improves ultrasound image segmentation by injecting edge-aware features and self-generated mask prompts into SAM's encoder pipeline.","lead":"EP-SAM adapts the Segment Anything Model for ultrasound image segmentation by adding edge-aware feature extraction and self-generated mask prompts. It improves boundary delineation on six ultrasound benchmarks, outperforming prior SAM-based methods.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"Ablation Table II shows sub-additive gains when EAM and PEM are combined, contradicting the paper's central 'complementary/synergistic' claim — the mechanistic framing is not supported by the paper's own evidence.","rationale":"The reader correctly identified that the 'synergy' claim is not causally verified and that the ablation doesn't rule out the simpler explanation of additional supervised signals. I sharpen this by noting that the ablation numbers actually contradict the synergy claim: the combined gains are sub-additive on both datasets, which is the quantitative signature of redundancy, not complementarity. The paper explicitly claims 'complementary nature' in §IV.C, but its own data shows the opposite. This is an internal inconsistency between the stated mechanism and the evidence, not merely an unverified assumption. The reader's verdict of CONDITIONAL is appropriate. The method is a legitimate empirical contribution with consistent improvements across datasets, but the mechanistic framing — which is the paper's primary novelty argument — is not supported by the evidence presented. If the additional supervision control experiment shows that a generic auxiliary head matches PEM's contribution, the paper's distinguishing claim weakens to 'we added more supervised signals to SAM adapters,' which is incremental rather than conceptually novel. The lack of significance testing on small improvements (0.13–1.1% Dice) compounds this: the headline 'consistently outperforms' could be within noise for some comparisons, particularly the CC-SAM comparison (86.95 vs 86.82). I agree with the reader's CONDITIONAL verdict and moderate confidence. The concern does not warrant REJECT because the method is sound and the empirical results are broadly consistent, but it does prevent ACCEPT until the mechanistic claim is either softened to match the evidence or supported by a proper causal isolation experiment.","tokens_in":10991,"tokens_out":1995,"duration_ms":125606,"concrete_test":"Re-run the ablation in Table II with an additional condition: EAM with an equivalent auxiliary supervised signal (e.g., a third segmentation head on intermediate features without the edge-aware mechanism) to match the supervision count of the full model. If EAM+auxiliary-supervision matches or exceeds EAM+PEM, the 'synergy' claim collapses — the gains are from additional supervision, not from the specific prompt-enhancement mechanism. Additionally, run 5-fold cross-validation with paired t-tests on the headline comparisons (EP-SAM vs UN-SAM, EP-SAM vs SAMUS) to verify the 0.6–1.1% Dice improvements are statistically significant.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central mechanistic claim is that EAM and PEM are 'complementary' and produce 'enhanced synergy' between SAM's image and prompt encoders (§IV.C: 'These results demonstrate the complementary nature of edge-aware feature extraction and prompt enhancement'). However, the ablation numbers in Table II show the opposite pattern — sub-additive gains. On TN3K: EAM alone gives +1.22 Dice, PEM alone gives +1.92, but combined gives +2.47 (vs. 3.14 if additive). On BUSI: EAM gives +1.14, PEM gives +3.23, combined gives +3.78 (vs. 4.37 if additive). In both cases, the combined improvement is substantially less than the sum of individual improvements, indicating overlapping/redundant information capture rather than complementarity. The paper misinterprets diminishing returns as synergy. This matters because the 'synergy between SAM components' is the paper's primary conceptual contribution distinguishing it from prior adapter-based methods (Fig. 1c). If the gains are simply from adding two supervised auxiliary signals that partially overlap, the novelty claim weakens considerably. Additionally, the absolute improvements over baselines are small (e.g., +0.61 Dice over UN-SAM without prompts, +0.13 over CC-SAM with prompts) with no significance testing, making it unclear whether these differences are robust.","agreement_with_reader":"agree"},"referee_report":{"model":"glm-5.2","summary":"The manuscript proposes EP-SAM, an adaptation of the Segment Anything Model (SAM) for ultrasound image segmentation. The method introduces two modules: an Edge-Aware Module (EAM) that extracts boundary cues from intermediate features of SAM's image encoder under explicit edge supervision, and a Prompt Enhanced Module (PEM) that fuses these edge-aware features with multi-block semantic features to generate dense mask-form prompts for the prompt encoder. The approach is evaluated on six ultrasound datasets (three in-domain, three cross-domain) against 14 baselines under prompt-free and single-point-prompt settings, with ablation studies isolating the contributions of EAM and PEM.","tokens_in":11187,"tokens_out":1121,"duration_ms":186564,"significance":"Ultrasound image segmentation suffers from boundary ambiguity and speckle noise, and adapting SAM to this domain is an active area of research. The paper's approach of using intermediate encoder features to generate self-prompts, rather than relying on external models or sparse user-provided prompts, is a reasonable architectural contribution. The experimental evaluation is relatively thorough, covering multiple anatomical structures and including both in-domain and cross-domain generalization tests. The inclusion of ablation studies (Tables II and III) to isolate component contributions is a positive practice.","major_comments":[{"comment":"§IV.C, Table II: The paper's central conceptual claim is that EAM and PEM are 'complementary' and produce 'enhanced synergy' between SAM's image and prompt encoders. However, the ablation results show sub-additive gains when the two modules are combined. On TN3K, EAM alone gives +1.22 Dice and PEM alone gives +1.92, but combined gives +2.47 (vs. 3.14 if additive). On BUSI, EAM gives +1.14 and PEM gives +3.23, but combined gives +3.78 (vs. 4.37 if additive). This pattern indicates overlapping or redundant information capture rather than complementarity. The paper should either temper the 'synergy/complementary' framing to match the evidence or provide additional analysis (e.g., feature similarity analysis, gradient-weighted class activation maps) demonstrating that the two modules capture genuinely non-redundant information. As stated, the mechanistic claim is not supported by the paper's","section":null},{"comment":"§IV.B, Table I: The absolute improvements over the strongest baselines are small (e.g., +0.61 Dice over UN-SAM without prompts on average, +0.13 Dice over CC-SAM with point prompts on average). No statistical significance tests (e.g., p-values, confidence intervals, or multiple random seed runs) are reported. Given the modest margins, it is unclear whether these differences are robust or within noise. The authors should add significance testing or report variance across multiple runs to substantiate the claim of consistent outperformance.","section":null},{"comment":"§III.A, §III.C: The EAM introduces additional supervision via the edge loss (L_edge), and PEM introduces the coarse mask loss (L_coarse). The ablation in Table II does not isolate whether the performance gains come from the architectural modules themselves or simply from the addition of auxiliary supervised signals. A fairer test of the 'synergy' claim would include a baseline that adds equivalent auxiliary supervision signals (e.g., a simple edge prediction head and a coarse mask prediction head) without the proposed cross-module feature routing, to verify that the gains are attributable to the architectural design rather than the multi-task supervision alone.","section":null}],"minor_comments":[{"comment":"§IV.A.1: The dataset split ratios for BUSI are rendered as unreadable Unicode escape sequences (e.g., '/uni00000038/uni00000010/...'). This should be corrected to the standard '7:1:2' format.","section":null},{"comment":"§III.C: The loss term weights for L_edge, L_coarse, and L_final are all set to 1, and α is fixed to 0.8. No sensitivity analysis is provided for these hyperparameters. A brief discussion of robustness to these choices would strengthen the paper.","section":null},{"comment":"Figure 2: The diagram is dense and some data flow paths (e.g., how the dynamic gating weights w1–w4 are computed and applied) are difficult to follow. Clarifying the flow or splitting the figure would aid readability.","section":null},{"comment":"§IV.B: The comparison with CC-SAM (an LLM-assisted method) is mentioned in the text but CC-SAM is only evaluated on the average column in Table I, not on individual datasets. For completeness, per-dataset results for CC-SAM should be included or the comparison scope should be clarified.","section":null}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern regarding sub-additive gains in Table II is valid and is the primary basis for the major revision recommendation. The paper frames EAM and PEM as synergistic, but the numbers show diminishing returns. This is a presentation/interpretation issue that can be fixed by tempering claims or adding analysis, but it is load-bearing for the paper's novelty positioning. The modest absolute improvements over baselines without significance testing is a secondary concern that should be addressed."},"author_rebuttal":null,"desk_editor":{"model":"glm-5.2","letter":"The paper proposes EP-SAM, which adds two modules to SAM for ultrasound segmentation: an Edge-Aware Module (EAM) that extracts boundary features from intermediate transformer blocks with ground-truth edge supervision, and a Prompt Enhanced Module (PEM) that fuses multi-level encoder features with edge cues to generate dense mask-form self-prompts. The combination is tested on six ultrasound datasets with in-domain and cross-domain evaluation against 14 baselines. The empirical work is thorough and the results are consistent — EP-SAM achieves the best or near-best Dice and HD across most settings, and the cross-domain generalization results are genuinely strong. The idea of using SAM's own intermediate features to generate both edge supervision and dense prompts, rather than relying on external operators like Sobel or external prompt sources, is a reasonable and clean design choice. Credit earned there. The ablations in Table II and III properly isolate the two modules and the contribution of individual transformer blocks. No code release is mentioned, which is a reproducibility gap but not unusual for the venue. The loss weights (α=0.8, equal term weights) are unjustified but this is a minor concern. The real soft spot is the mechanistic framing. The paper repeatedly claims EAM and PEM are 'complementary' and produce 'enhanced synergy' between SAM's image and prompt encoders. But the ablation numbers in Table II show sub-additive gains: on TN3K, EAM alone gives +1.22 Dice and PEM alone gives +1.92, but combined gives +2.47 (not 3.14). On BUSI, the pattern is the same — combined gains fall well short of the sum of individual gains. This is the opposite of synergy; it suggests the two modules capture overlapping information and the improvements may simply come from adding more supervised auxiliary signals. The paper interprets diminishing returns as complementarity, which is a misread of its own evidence. The absolute improvements over strong baselines are also small (e.g., +0.61 Dice over UN-SAM without prompts, +0.13 over CC-SAM with prompts) with no significance testing, so robustness is unclear. This is a competent empirical paper with a clean design and solid results, but the conceptual contribution is overstated relative to what the ablations actually show. It deserves a serious referee who can push back on the mechanistic claims and ask for significance tests. The method itself is sound enough to warrant review.","headline":"Solid SAM adaptation for ultrasound segmentation, but the 'synergy' claim is undermined by the paper's own ablation numbers","tokens_in":11950,"tokens_out":577,"would_cite":false,"duration_ms":200656,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"glm-5.2","headline":"Edge-aware prompts fix SAM's ultrasound boundary problem","keywords":["ultrasound image segmentation","Segment Anything Model","edge-aware segmentation","prompt enhancement","boundary delineation","medical image segmentation"],"falsifier":"If feeding the same edge-aware features directly to the mask decoder (bypassing the prompt encoder loop) yields equivalent or better results, the claim that prompt-level integration is the key mechanism would be undermined.","tokens_in":11204,"feed_emoji":"🔬","tokens_out":942,"duration_ms":321216,"temperature":0.7,"pith_summary":"The Segment Anything Model (SAM), despite its success on natural images, produces blurry or drifting boundaries when applied to ultrasound images—a modality plagued by speckle noise, low contrast, and gradual tissue transitions. This paper proposes EP-SAM, which adapts SAM by adding two modules that connect the model's image encoder to its prompt encoder. The first, the Edge-Aware Module (EAM), extracts boundary cues from intermediate features of SAM's image encoder and supervises them against ground-truth edge maps. The second, the Prompt Enhanced Module (PEM), fuses these edge-aware features with multi-level semantic features to generate coarse mask predictions, which are then fed back into SAM's prompt encoder as dense mask-form prompts. The central claim is that closing the loop between what the image encoder sees and what the prompt encoder receives—specifically by injecting explicitly supervised boundary information—resolves the boundary ambiguity that causes SAM to fail on ultrasound data. Experiments on six ultrasound benchmarks show EP-SAM outperforming prior SAM-based methods under both prompt-free and single-point-prompt settings.","feed_headline":"Edge-aware prompts fix SAM's ultrasound boundary problem","feed_subtitle":"By extracting boundary cues from SAM's own encoder features and feeding them back as dense prompts, the method resolves the edge-drift that ","key_machinery":"The Edge-Aware Module (EAM), which applies gated residual convolutional refinement to intermediate SAM transformer features and supervises them against Canny-extracted ground-truth edges; and the Prompt Enhanced Module (PEM), which fuses multi-level encoder features with EAM outputs through dynamic gating to produce coarse masks that serve as dense prompts for SAM's frozen prompt encoder.","core_discovery":"The paper identifies that SAM's failure on ultrasound images stems from treating the image encoder, prompt encoder, and mask decoder as isolated modules, causing boundary cues present in intermediate encoder features to be lost before they reach the prompt stage. By extracting edge features from intermediate transformer blocks, supervising them with ground-truth boundaries, and feeding the result back as dense mask prompts, the model achieves boundary-informed segmentation without requiring external prompt sources or heavy architectural changes.","pith_inferences":["If the performance gain comes primarily from the additional supervision signal rather than from the architectural coupling between encoder and prompt encoder, then simply adding an auxiliary edge-loss branch to the mask decoder might achieve similar gains without the prompt-generation machinery.","The claim that EAM and PEM improve 'synergy' between SAM components would be strengthened by showing that removing the feedback loop (feeding edge features directly to the decoder rather than through the prompt encoder) degrades performance—if it does not, the synergy claim is weaker than the architectural narrative suggests.","The use of Canny-extracted edges as ground truth introduces a dependency on classical edge detection; if the quality of these pseudo-labels varies across anatomical structures, the method's boundary improvement may be uneven across datasets in ways not captured by aggregate Dice scores."],"forward_implications":["If intermediate encoder features in SAM carry boundary information that the mask decoder cannot recover on its own, then any SAM adaptation for modalities with ambiguous boundaries (not just ultrasound) could benefit from explicit edge supervision at the encoder level.","Self-generated dense mask prompts derived from internal features may substitute for external prompting (points, boxes, or language models), reducing the need for user interaction or large language model integration in medical imaging workflows.","The principle of bridging encoder and prompt encoder with task-specific intermediate supervision could generalize to other foundation-model adaptation scenarios where the pre-trained model's components are under-coupled."],"fun_headline_variants":["Recycling SAM's internal edge features as prompts fixes ultrasound segmentation","Feeding SAM's own boundary cues back as prompts improves ultrasound segmentation","Edge features extracted from SAM's encoder and fed back as prompts beat baseline","Supervising SAM's intermediate edge features and feeding them back as dense prompts","EP-SAM uses edge-aware supervision and dense prompts to resolve ultrasound boundary drift"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The paper assumes that the boundary features extracted from intermediate transformer blocks provide information that is complementary to, rather than redundant with, what the mask decoder already processes, and that the observed gains stem from the coupling mechanism rather than simply from adding more supervised training signals.","fun_headline_variants_meta":{"raw":{"variants":["Recycling SAM's internal edge features as prompts fixes ultrasound segmentation","Feeding SAM's own boundary cues back as prompts improves ultrasound segmentation","Edge features extracted from SAM's encoder and fed back as prompts beat baseline","Supervising SAM's intermediate edge features and feeding them back as dense prompts","EP-SAM uses edge-aware supervision and dense prompts to resolve ultrasound boundary drift"]},"model":"glm-5.2","effort":"low","cost_usd":0.0,"raw_usage":{"total_tokens":538,"prompt_tokens":444,"completion_tokens":94,"prompt_tokens_details":null},"tokens_in":444,"tokens_out":94,"duration_ms":21398,"temperature":1.0,"reasoning_tokens":null,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-09T16:23:48.034435+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"If feeding the same edge-aware features directly to the mask decoder (bypassing the prompt encoder loop) yields equivalent or better results, the claim that prompt-level integration is the key mechanism would be undermined.","supporting_citations":[],"review_version":1}