{"id":"6a2e2b50-8591-480e-92b4-30f44a115acc","arxiv_id":"2504.12772","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A systematic review that classifies photoacoustic imaging artifacts into five sources and illustrates each with simulations and in vivo examples.","lead":"This paper reviews the sources of image artifacts in photoacoustic imaging, an emerging clinical imaging technology. It proposes a five-source classification and illustrates each artifact with simulations and clinical examples, aiming to help clinicians recognize and mitigate them.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"All-artifacts framing overreaches: Section 11 excludes reconstruction-induced artifacts, so the five-source taxonomy cannot support the claimed completeness.","rationale":"The reader correctly identifies the isolation strategy in Section 4 as a didactic risk, but the more load-bearing issue is completeness: the paper's own Section 11 explicitly excludes reconstruction-induced artifacts, and Section 10 acknowledges that all reconstruction methods embed approximate physics models. The central claim as stated by the reader, that all PAI artifacts can be classified into five sources, is therefore broader than what the manuscript actually demonstrates. This is not a fatal flaw: the paper has strong independent support, including an open simulation code link, physics-based derivations, and reprinted experimental examples. The taxonomy is well suited to artifacts originating in patient, light-tissue, PA-effect, sound-tissue, and signal-detection processes. However, because the title and abstract promise a general account of artifacts in PAI, the explicit scoping in Section 11 should be moved to the abstract or a matching section on reconstruction-induced artifacts should be added. This is a modest revision rather than a rejection, so the verdict should move from ACCEPT to CONDITIONAL.","tokens_in":31024,"tokens_out":7739,"duration_ms":86807,"concrete_test":"Build an artifact inventory from 3-5 recent PAI reviews (e.g., Tian et al., Photonics Insights 3:R06, 2024; Hauptmann and Cox, JBO 25:112903, 2020) and map every distinct artifact named there onto the five sources in Table 2 and the two causes in Section 2. Count artifacts that are reconstruction-specific (backprojection streak artifacts, negative-value artifacts, learned-method hallucinations, time-reversal trapping) or that do not fit any source. If the count is nonzero, the exhaustive-taxonomy claim fails and the paper should be revised to narrow the claim or add the missing category.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim, as summarized, is that all PAI artifacts reduce to two causes (insufficient data, incorrect reconstruction assumptions) and sort into five sources. The paper's own Section 11 limits the scope: 'the scope of this paper is limited to artifacts that originate outside of the reconstruction step,' and acknowledges that reconstruction algorithms can 'introduce new artifacts' (e.g., backprojection streak artifacts, negative-value artifacts, deep-learning hallucinations). These reconstruction-induced artifacts are not classified or demonstrated anywhere in Sections 4-9, so the five-source taxonomy is a taxonomy of a subset, not of all PAI artifacts. The two-cause statement in Section 2 can conceptually encompass reconstruction artifacts, but the paper never tests or shows that mapping. A clinical reader encountering an artifact from a specific reconstruction algorithm has no entry in the proposed classification. Thus the load-bearing premise that the taxonomy is exhaustive is unsupported; the title and abstract should either include this class or explicitly state the narrower scope up front.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This review paper provides a systematic account of image artifacts in photoacoustic imaging (PAI). It proposes that the artifacts considered arise from two fundamental causes, insufficient data and incorrect assumptions embedded in the reconstruction model, and organizes them into five sources: patient, light-tissue interactions, the photoacoustic effect, sound-tissue interactions, and signal detection. For each source, the paper explains the underlying physics, illustrates the artifact with simulations that break exactly one idealizing assumption, and provides experimental images from the literature. It closes with a discussion of advanced mitigation strategies, including improving the physics model, supplementing the data, deep learning, and deconvolution. The simulation code is provided through SIMPA, and the figures are generally clear.","tokens_in":31127,"tokens_out":7780,"duration_ms":80808,"significance":"If the scope issue is resolved, this will be a timely and valuable educational resource for clinical users and a useful reference for system developers. The simulation protocol is transparent, uses no fitted parameters, and is reproducible via the linked code. The simulated artifact signatures are corroborated by experimental examples from the literature, including in vivo images. The proposed taxonomy provides a helpful mental model even though, as discussed below, its claimed completeness needs to be scoped.","major_comments":[{"comment":"The paper frames the artifact taxonomy as covering PAI artifacts generally, with Section 2 stating that there are two corresponding classes of underlying causes (insufficient data and incorrect assumptions in the image reconstruction algorithm). Section 11, however, explicitly limits the scope to artifacts that originate outside the reconstruction step and acknowledges that reconstruction algorithms can introduce new artifacts, such as backprojection streaks, negative-value artifacts, and deep-learning hallucinations. These reconstruction-induced artifacts are not classified in the five-source taxonomy, so the classification is not exhaustive as presented. The title, abstract, and Section 2 should either incorporate reconstruction-induced artifacts or state prominently that the taxonomy covers only artifacts arising before the reconstruction step; otherwise the reader is left without an entry point for a whole class of artifacts.","section":"Section 2 and Section 11"},{"comment":"The isolation methodology in Section 4 (ideal settings except for the one difference giving rise to the artifact) is clear pedagogically, but the paper does not address how artifact signatures interact when several assumptions are violated simultaneously, which is the clinical situation. Because the stated purpose is to train clinical users to identify artifacts, the Discussion should warn explicitly that combined artifacts may not have the same visual signature as the isolated demonstrations and, ideally, include at least one simulation with multiple simultaneous violations to illustrate the interaction. The experimental examples partially mitigate this concern, but they are not a substitute for a stated caveat.","section":"Section 4"}],"minor_comments":[{"comment":"The phrase 'assess whether their impact' is incomplete; the sentence needs an object, e.g., 'assess whether their impact is clinically significant'.","section":"Abstract"},{"comment":"The text 'For a typical resolution of 100 µs' should read '100 µm'; the current wording mixes time and length units in the motion estimate.","section":"Section 9.1"},{"comment":"Assumption H7 is listed as 'Detectors are perfectly directional,' but Section 8.2.1 states the ideal assumption is omnidirectionality; the table wording should be corrected to avoid contradiction.","section":"Table 1"},{"comment":"The simulation volume is described with '0.25 mm/pixel'; for a 3D volume this should be '0.25 mm/voxel'.","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is technically sound and the simulations are reproducible; the key issue is framing. I recommend asking for the scope revision and a prominent statement about artifact interactions. With those changes it would be a strong candidate for acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this is a review, not a research result: no new physics, no new measurements, no new reconstruction method. Its contribution is a five-source taxonomy of PAI artifacts (patient, light–tissue interactions, PA effect, sound–tissue interactions, signal detection) plus explicit tables matching each artifact to the reconstruction assumption it violates. That taxonomy is well built and clearly illustrated, and for a field moving into the clinic it fills a real gap. The simulations are reproducible (SIMPA, code linked), the experimental reprints anchor the didactic images in real data, and the physics in Section 3 is correctly stated.\n\nSecond, the packaging overreaches. Section 2 says there are two classes of underlying causes of artifacts—insufficient data and incorrect assumptions—and the title promises artifacts in PAI generally. But Section 11 tells you the scope is artifacts that originate outside the reconstruction step; reconstruction-specific artifacts like backprojection streaks, negative-value artifacts, and deep-learning hallucinations are named but not classified or demonstrated. So the taxonomy is a taxonomy of a subset, not of all PAI artifacts. A clinical reader who meets an artifact from a specific algorithm has no entry in the proposed classification. This is a real mismatch between the explicit scope and the framing, and it should be fixed by moving the scope statement into the abstract and toning down the completeness claim. It's not fatal—the paper does own the limitation—but as written it will mislead a skimming reader.\n\nThe weakest methodological point is the one-at-a-time simulation design: each artifact is isolated in otherwise ideal settings. That is sensible for teaching, and the paper is open about it, but it means the visual signatures could shift in clinical images where several assumptions fail at once and the reconstruction algorithm differs. That's a limitation, not a flaw.\n\nMinor: the abstract has a grammatical error ('assess whether their impact'), and Figure 9's caption repeats 'are.' Nothing substantive.\n\nWho is this for? Clinicians starting to use PAI, developers checking which assumptions their system violates, and anyone writing an intro to PAI artifacts. It does not advance the research frontier, but it's not trying to.\n\nI'd send it to peer review. The taxonomy is worth publishing and the simulations support it. Request a revision that aligns the title and abstract with the actual scope and adds the reconstruction-artifact class to the classification discussion, even if only as a listed outside-scope branch.","headline":"Useful educational taxonomy of PAI artifacts, honestly scoped in the discussion but overbroad in title and abstract; worth publishing after a scope-claim fix.","tokens_in":31702,"tokens_out":3437,"would_cite":true,"duration_ms":33079,"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":"The paper argues that every photoacoustic imaging artifact arises from insufficient data or incorrect assumptions in reconstruction, organized into five artifact sources with simulation and clinical examples.","keywords":["photoacoustic imaging","imaging artifacts","artifact classification","image reconstruction","medical imaging","fluence decay","limited-view tomography","artifact mitigation"],"falsifier":"A single clinical or simulated artifact that persists when the data are complete, noise-free, and broadband and the reconstruction uses an exact physical model would falsify the two-cause claim. A more practical check is to simulate two artifacts together, such as fluence decay plus sound-speed error, and see whether their isolated signatures still predict the combined image; if they mask or transform each other, the single-cause illustrations cannot be extended directly to clinical images.","tokens_in":1631,"feed_emoji":"🩻","tokens_out":2557,"duration_ms":91404,"temperature":0.7,"pith_summary":"This review argues that every artifact in photoacoustic imaging has one of two root causes: the imaging system records insufficient data, or the reconstruction algorithm embeds assumptions about the physics that do not hold. Around that distinction the authors build a five-way classification by source—patient, light-tissue interactions, the photoacoustic effect, sound-tissue interactions, and signal detection—and show each artifact type with paired simulations and clinical examples. The practical goal is to give clinical users a field guide for recognizing artifacts before they cause misdiagnosis, and to give system developers a reference for which measurement or modeling gaps to close. If the classification holds, artifact recognition becomes a matter of mapping a visual signature back to a violated assumption, which in turn points to the mitigation.","feed_headline":"Two causes explain every photoacoustic artifact","feed_subtitle":"A five-source taxonomy with simulated and clinical examples helps users recognize and mitigate artifacts.","key_machinery":"The organizing device is the two-cause taxonomy and the assumption list behind it. The physical backbone is the initial-pressure equation $p_0(\\mathbf{x}) = \\Gamma(\\mathbf{x})\\mu_a(\\mathbf{x})\\Phi(\\mathbf{x})$ for light absorption and the linear acoustic wave equation for propagation, and each artifact corresponds to a violated assumption in that chain. The taxonomy maps every artifact source to a cause, such as fluence decay, spectral coloring, sound-speed mismatch, or limited view, and to a visual effect, such as dislocation, splitting, blurring, clutter, or signal loss. Because the simulations isolate one violated assumption at a time, the machinery also provides a controlled way to learn what each artifact looks like.","core_discovery":"On the paper's own terms, the central claim is that the confusing variety of photoacoustic artifacts is not arbitrary. Every artifact is produced either because the measured data are incomplete—limited detector coverage, sparse elements, finite bandwidth, noise—or because the reconstruction model assumes something false, such as uniform sound speed, no acoustic attenuation, omnidirectional detectors, or spatially constant light fluence. The authors organize the resulting phenomena into five sources (patient, light-tissue interactions, the photoacoustic effect, sound-tissue interactions, and signal detection) and connect each to concrete assumptions in the image-formation chain $p_0(\\mathbf{x}) = \\Gamma(\\mathbf{x})\\mu_a(\\mathbf{x})\\Phi(\\mathbf{x})$, followed by linear acoustic propagation to the detectors. Simulated phantoms that deliberately break one assumption at a time, together with in vivo examples, demonstrate the characteristic signature of each artifact.","pith_inferences":["Extension: If the two-cause framing is right, artifact identification becomes a diagnostic inverse problem: the violated assumption could be inferred from the image itself, pointing toward automated artifact annotation tools for clinical workflows.","Extension: The same two-cause dichotomy likely transfers to other hybrid modalities that reconstruct from physical models, so the taxonomy may apply to ultrasound-guided or optoacoustic tomography with re-labelled sources.","Extension: The paper leaves reconstruction-induced artifacts out of scope, so a natural extension of the classification would add a sixth source for algorithm-specific artifacts, including deep-learning hallucinations.","Extension: A testable extension of the isolation protocol is to break two assumptions at once; if the resulting signatures interact nonlinearly, clinical users would need signatures learned from combined artifact conditions rather than from single-cause examples."],"forward_implications":["Clinicians using multispectral PAI can learn to recognize spectral coloring and out-of-plane absorption as causes of false or missing features in oxygenation maps, reducing the risk of misdiagnosis.","Knowing that sound-speed mismatch causes dislocation, splitting, and blurring, scanner designers can incorporate measured sound-speed maps to recover vessels that are invisible under constant-sound-speed reconstruction.","For hardware, the taxonomy makes trade-offs explicit: more detector coverage or denser elements reduces limited- and sparse-view clutter, while detector directivity and bandwidth set a floor on resolution that deconvolution can only partially recover.","Artifacts can sometimes carry diagnostic information, so identifying an artifact correctly may assist diagnosis rather than merely confound it, as with the comet-tail artifact in ultrasound.","Because mitigation splits into improving the model versus supplementing the data, any reconstruction algorithm, including deep learning, will still produce artifacts whenever its implicit physics model is wrong or the data are incomplete."],"supporting_citations":[{"why":"Supplies the photoacoustic physics—absorption, Grüneisen parameter, and thermal and stress confinement—that defines the assumptions being broken.","marker":"[5]"},{"why":"The simulation toolkit with which every idealized artifact example is generated.","marker":"[25]"},{"why":"The delay-and-sum beamformer used in the simulations, the baseline reconstruction whose assumptions are violated.","marker":"[26]"},{"why":"Experimental phantom data showing fluence decay with depth in blood-filled tubes.","marker":"[29]"},{"why":"In vivo breast images demonstrating spectral coloring and the effect of sound-speed correction.","marker":"[33]"},{"why":"Experimental demonstration that an out-of-plane absorber is projected into the imaging plane.","marker":"[34]"},{"why":"Experimental demonstration that temperature-dependent Grüneisen parameter variations change the photoacoustic signal.","marker":"[52]"}],"fun_headline_variants":["Every photoacoustic artifact has one of two roots","All photoacoustic artifacts stem from two system flaws","Photoacoustic artifacts: incomplete data and wrong assumptions","Five-source taxonomy roots every artifact in two flaws"],"cache_read_input_tokens":33920,"weakest_assumption_plain":"The whole didactic apparatus depends on whether an artifact's isolated signature in an ideal simulation still appears recognizably the same in a real clinical image, where many assumptions fail at once and the reconstruction algorithm is different.","fun_headline_variants_meta":{"raw":{"variants":["Every photoacoustic artifact has one of two roots","All photoacoustic artifacts stem from two system flaws","Photoacoustic artifacts: incomplete data and wrong assumptions","Five-source taxonomy roots every artifact in two flaws"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000878,"raw_usage":{"total_tokens":3759,"prompt_tokens":873,"completion_tokens":2886,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":489,"completion_tokens_details":{"reasoning_tokens":2824}},"tokens_in":489,"tokens_out":2886,"duration_ms":21724,"temperature":1.0,"reasoning_tokens":2824,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:22:18.589330+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A single clinical or simulated artifact that persists when the data are complete, noise-free, and broadband and the reconstruction uses an exact physical model would falsify the two-cause claim. A more practical check is to simulate two artifacts together, such as fluence decay plus sound-speed error, and see whether their isolated signatures still predict the combined image; if they mask or transform each other, the single-cause illustrations cannot be extended directly to clinical images.","supporting_citations":[{"cited_title":"Kirchner, F","cited_arxiv_id":null,"evidence_quote":"The delay-and-sum beamformer used in the simulations, the baseline reconstruction whose assumptions are violated."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Experimental phantom data showing fluence decay with depth in blood-filled tubes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Experimental demonstration that an out-of-plane absorber is projected into the imaging plane."}],"review_version":1}