{"id":"a8c7e6da-213b-4c11-a9d3-eb45e53ad132","arxiv_id":"2607.03761","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.5,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Variance-driven Softmax annealing selects CT angles on the fly, improving SSIM over random sampling especially for anisotropic objects under noise.","lead":"SAVER adaptively picks CT projection angles in real time by tracking the variance of measured absorption values, using Softmax scheduling with annealing. This reallocates dose toward informative directions and yields higher reconstruction fidelity than fixed random sampling on anisotropic phantoms.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified beyond the reader's already-flagged proxy and regime limits.","rationale":"The central claim is an empirical performance statement inside a fully specified simulation regime (parallel-beam, monochromatic, 32\times32, N=500, SSIM on the inscribed circle). The mathematics (Tikhonov + Woodbury, Softmax annealing) is standard and inspectable; the code is released. The only place the claim could fail is if the variance proxy systematically mis-ranks angles under partial sampling and random orientation. The paper already tests that risk with oracles, axis-prior ablations, and noise levels, and the gains track structural anisotropy exactly as predicted. No internal contradiction or circularity appears. Therefore the reader’s CONDITIONAL (HIGH confidence) already correctly balances the positive numerical evidence against the acknowledged extrapolation limits; no adjustment is warranted.","tokens_in":13322,"tokens_out":497,"duration_ms":4790,"concrete_test":"Re-run the full 10-seed protocol of Fig. 5–6 on the same 8 phantoms but with only one initialization ray per angle (instead of two) and with η=0.01; if SAVER-A’s AUC/500 advantage over AIRS/Random disappears or reverses on the anisotropic set (Rectangle, Cross, Stripes), the early-variance proxy is too fragile for the claimed real-time adaptation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest assumption (sample variance of a few scalar rays as a real-time proxy for angular information content under random rotation) is the correct soft spot, but it is not a hidden inconsistency: the paper itself supplies the necessary controls. MAX-V vs MIN-V (Fig. 5–6) shows that true variance ordering is informative precisely when sinograms are anisotropic; SAVER-A closes most of that gap once two rays per angle seed the estimates (Eq. 3 + Phase-1 workflow); and the Softmax annealing (Eq. 4) plus axis-prior prevent collapse when the proxy is weak (isotropic phantoms, high noise). The remaining limitations—32\times32 monochromatic phantoms, no real scanner data—are scope restrictions already reflected in the CONDITIONAL verdict, not load-bearing errors that would reverse the claim inside the stated regime.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The manuscript proposes SAVER, an adaptive CT acquisition scheme that treats projection selection as a sequential decision process. At each round a single ray is measured; angles are chosen by a Softmax policy whose scores are the robustly scaled sample variances of the scalar projection values observed so far (Eqs. 3–4), with temperature annealed from exploration to exploitation. Reconstruction uses Tikhonov regularization updated recursively via the Woodbury identity. On eight 32\times32 monochromatic phantoms (ten random rotations, two noise levels) SAVER and its axis-initialized variant SAVER-A are compared with Random, AIRS, and four oracles that use ground-truth variances. The central claim is that variance-driven Softmax annealing reallocates dose to informative angles and yields higher SSIM (and higher AUC/500) than uniform random sampling, especially for anisotropic objects, while remaining stable under noise.","tokens_in":13585,"tokens_out":968,"duration_ms":8061,"significance":"If the result holds inside the stated regime, the work supplies a concrete, mathematically transparent alternative to fixed-geometry low-dose CT: a real-time, sample-dependent policy that needs no learned prior and is supported by clear oracle controls (MAX-V vs MIN-V). The public repository, the explicit Softmax-annealing schedule, and the Woodbury recursion are reproducible strengths. The contribution is primarily methodological and proof-of-concept; clinical impact remains prospective until larger, polychromatic, or real-scanner experiments appear. Within the adaptive-sampling / experimental-design literature the paper is a useful, carefully controlled demonstration rather than a definitive clinical solution.","major_comments":[{"comment":"The load-bearing proxy (sample variance of a few scalar rays as a real-time surrogate for angular information content) is only weakly validated outside the anisotropic phantoms. For Triangle, Gradient and Shepp-Logan the SSIM/AUC gaps between SAVER-A and AIRS/Random shrink to near zero (Figs. 5–7), and the paper never quantifies how many rays per angle are required before ˆσ_i^{2} becomes a reliable ranking. A short ablation that reports rank correlation between online and true variances versus n_i(t), or that freezes the Softmax scores after Phase 1, would make the claim falsifiable rather than visual.","section":null},{"comment":"All experiments remain on 32\times32 monochromatic linear phantoms with R=32 and N=500. The computational discussion correctly notes that the O(d^{2}) Woodbury update is already prohibitive at clinical resolutions, yet no scaling experiment (even 64\times64) or approximate-covariance alternative is shown. Without at least one higher-resolution or fan/cone-beam demonstration, the claim that SAVER “marks a shift toward sample-dependent CT acquisition” rests on a regime whose practical relevance is still unproven.","section":null}],"minor_comments":[{"comment":"Table 1 and the Methods text list η∈{1,0.1,0.01} while the main-text discussion of Fig. 5 mentions η=0.01 and the SI caption alludes to η=10; the set of annealing rates should be stated once and consistently.","section":null},{"comment":"Figure 8 caption asserts that image restoration is unnecessary for decision-making, yet every reported SSIM curve is generated by the Woodbury update at every round; a one-sentence clarification that the curves are diagnostic only would avoid confusion.","section":null},{"comment":"The notation for the robust score ˜σ_i(t) uses the same symbol for both the scaled variance and the Softmax argument; a distinct symbol (e.g., s_i(t)) would improve readability of Eq. 4.","section":null},{"comment":"Several references (e.g., Joseph 2007, Wang et al. 2003) appear with incomplete or non-standard bibliographic data; a quick clean-up would help.","section":null}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid methods contribution for an ML/imaging venue, but the 32\times32 monochromatic setting and the absence of any real or higher-resolution data make it a borderline fit for a top clinical-imaging journal. The major-revision request is intended to force a tighter quantification of the variance proxy and at least a minimal scaling study; if those are supplied, minor_revision or accept would be appropriate. No integrity or novelty concerns."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing worth knowing is that SAVER is a cleanly engineered, reproducible adaptive-acquisition loop: seed two rays per angle, score angles by sample variance of the scalars seen so far, then Softmax-anneal the selection probabilities. On eight 32\times32 phantoms with random rotations it beats plain random sampling (and axis-initialized random) on SSIM, especially when the sinogram is anisotropic; the MAX-V/MIN-V oracles confirm that true variance ordering is informative exactly when you would hope. Code is released.\n\nWhat is new is the concrete combination—online variance as the instantaneous score, Softmax with temperature annealing, axis-prior warm-start, and the systematic oracle baselines—not a first-principles derivation. The math is standard and correctly written: linear forward model, Tikhonov, Woodbury rank-one update, Softmax policy. Circularity is low; the oracles use ground-truth variances never available to SAVER, and the SSIM curves are genuine out-of-sample metrics. Free parameters (T_initial, T_min, η, ξ, Δθ) are declared and ablated in the supplement.\n\nSoft spots are real but already scoped by the authors. The proxy (sample variance of a few scalar rays for whole-angle information content under random rotation) is the weakest link; the paper itself shows it works best after the two-ray seed and for anisotropic objects, and Softmax + annealing keep it from collapsing on isotropic or noisy cases. Everything is monochromatic parallel-beam 32\times32; no beam hardening, scatter, or real scanner data. That is a scope limit, not a hidden contradiction. Novelty is incremental (adaptive sampling + Softmax annealing are established tools).\n\nThis is for people who work on sequential experimental design or low-dose CT acquisition, not for pure reconstruction theorists. It deserves a serious referee: the claim is supported inside the stated regime, the design is inspectable, and the limitations are honest. I would send it out.","headline":"Clean adaptive CT acquisition that uses online projection variance + Softmax annealing; solid gains on anisotropic 32\times32 phantoms, but still a simulation-only incremental result.","tokens_in":14148,"tokens_out":496,"would_cite":false,"duration_ms":4525,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"SAVER reallocates CT radiation dose in real time by chasing high-variance projection angles, yielding higher reconstruction fidelity than uniform random sampling on anisotropic objects.","keywords":["low-dose CT","adaptive sampling","projection variance","Softmax annealing","sequential decision","SSIM reconstruction","anisotropic phantoms"],"falsifier":"On a phantom whose true angular information is known a priori, replace the online sample-variance scores with the true variances (the oracle SAVER-O already present in the paper) and check whether the practical SAVER SSIM curve collapses to random-sampling performance once the early-round variance estimates become noisy or biased.","tokens_in":14257,"feed_emoji":"🧮","tokens_out":635,"duration_ms":5569,"temperature":0.7,"pith_summary":"Conventional low-dose CT irradiates every angle the same way, even when an organ's structure is strongly directional and some angles carry far more information than others. This paper introduces SAVER, a sequential acquisition policy that treats each new X-ray ray as a decision: it estimates the sample variance of the projection values already seen at each angle, then chooses the next angle with a Softmax probability that is annealed from exploratory to exploitative. On eight 32\times32 phantoms the method reaches higher Structural Similarity (SSIM) with fewer rays than pure random sampling, especially when the object is anisotropic, and the same stochastic rule remains stable under substantial measurement noise. The practical claim is that diagnostic quality per unit dose can be raised by making the scan geometry itself sample-dependent rather than fixed in advance.","feed_headline":"CT dose reallocated by chasing high-variance angles","feed_subtitle":"Softmax annealing beats uniform random sampling on anisotropic phantoms and stays stable under noise","key_machinery":"The Softmax selection probability Pi(t) = exp(σ̃i(t)/Tt) / ∑ exp(σ̃j(t)/Tt), where σ̃i is the robust-scaled sample variance of rays already acquired at angle i and Tt is a temperature that cools by simulated annealing; this single stochastic rule both ranks angles by estimated structural information and keeps enough exploration to avoid premature collapse.","core_discovery":"Across eight diverse phantoms, a Softmax policy driven by real-time sample variance of projection values, annealed from exploration to exploitation, produces consistently higher reconstruction SSIM than conventional random angular sampling, with the largest gains on objects whose sinograms show strong angular anisotropy, and with retained stability under high measurement noise.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Softmax variance sampling beats random angles for low-dose CT","Adaptive high-variance projections lift CT SSIM on anisotropic phantoms","Real-time sample variance steers angles for higher-fidelity CT","Annealed Softmax reallocates CT dose to informative projections","SAVER variance policy outperforms random sampling under noise"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The sample variance computed from the few rays already measured at an angle is treated as a reliable enough proxy for how much structural information that entire angle still carries, even when the object has been randomly rotated relative to the grid.","fun_headline_variants_meta":{"raw":{"variants":["Softmax variance sampling beats random angles for low-dose CT","Adaptive high-variance projections lift CT SSIM on anisotropic phantoms","Real-time sample variance steers angles for higher-fidelity CT","Annealed Softmax reallocates CT dose to informative projections","SAVER variance policy outperforms random sampling under noise"]},"model":"grok-4.5","effort":"low","cost_usd":0.002904,"raw_usage":{"total_tokens":1024,"prompt_tokens":714,"num_sources_used":0,"completion_tokens":68,"cost_in_usd_ticks":29040000,"prompt_tokens_details":{"text_tokens":714,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":242,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":714,"tokens_out":68,"duration_ms":2579,"temperature":1.0,"reasoning_tokens":242,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-12T00:07:25.396468+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"On a phantom whose true angular information is known a priori, replace the online sample-variance scores with the true variances (the oracle SAVER-O already present in the paper) and check whether the practical SAVER SSIM curve collapses to random-sampling performance once the early-round variance estimates become noisy or biased.","supporting_citations":[],"review_version":1}