{"id":"e167a08d-cd17-457f-b919-e079278cf866","arxiv_id":"2506.04473","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Tractography of major white matter bundles is feasible on a 0.064 T portable MRI system, with visual but not quantitative agreement to 3 T reference data.","lead":"Researchers show that brain tractography, a method for mapping white matter fiber bundles, works on a low-cost portable 0.064 T MRI scanner. If confirmed, this could expand brain connectivity research to clinics and regions without high-field machines.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline 'strong correspondence' with high-field references rests on visual inspection of one representative subject; no quantitative tractography or scalar-map comparison is reported.","rationale":"I read the paper as a feasibility demonstration with comparative claims. The engineering effort is substantial and the artifact documentation is candid, which counts in its favor. However, the abstract's 'strong correspondence' and 'good agreement' are comparative claims, and comparative claims require quantitative comparison. The paper reports none: the 3 T reference data were acquired on the same participants, so the relevant metrics are readily computable, but the results section relies on visual judgment of a single subject. The reader's conditional verdict is appropriate, and my concern does not change it, so verdict_should_be is UNCHANGED. I partially agree with the reader's weakest_assumption: the B0 correction's scalar-scaling approximation is a genuine technical risk, and the paper itself shows residual spatial artifacts. But it is not the most load-bearing issue for the central claim as stated. Even a perfect B0 correction would not turn visual inspection of one subject into quantitative evidence of 'strong correspondence.' Conversely, if quantitative validation were supplied, the B0 approximation could be tested directly through agreement with the high-field reference. The concrete test uses data already in hand and would settle whether the correspondence claim is supported; it is a verification step, not a request for new experiments.","tokens_in":13552,"tokens_out":4961,"duration_ms":63205,"concrete_test":"Using the already-acquired 3 T Connectom data from the same five participants, compute for each of the nine listed bundles the Dice coefficient (or symmetric bundle-overlap) between ULF and HF tract density maps thresholded consistently, and compute voxelwise Pearson correlation and RMSE for FA and MD within the white-matter mask. Report results for all five participants with summary statistics. If median bundle Dice is below an a-priori threshold (e.g., 0.5) or FA/MD correlations are weak, the 'strong correspondence' claim should be softened; if all metrics are high, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim has two parts: feasibility of retrieving major bundles, and 'strong correspondence'/'good agreement' with high-field references. The feasibility part is supported by example tractograms, but the correspondence part is not quantitatively supported anywhere in the paper. Results Sections 3.1 and 3.2 compare ULF and HF only by visual inspection ('visually corresponded', 'broadly comparable', 'generally good correspondence') on a 'representative result of a compliant subject' (Section 3.1), despite five participants being scanned (Section 2.1). No Dice or bundle-overlap coefficients are reported, no voxelwise correlation or RMSE for FA or MD, no tractography metrics, and no group-level statistics. The discussion itself concedes that 'encouraging results shown must be validated against high field data in a range of populations before this approach can be endorsed.' Because the abstract asserts 'strong correspondence' and 'reliable tractography,' the load-bearing condition for the comparative half of the claim is a quantitative demonstration of agreement, and it is absent. The B0 scalar-correction approximation (Appendix Eq. 8) is a genuine accuracy risk, and the paper documents residual spatially varying artifacts (Section 3.3, Figure 9), but the more immediate problem is that the evidence base does not support the strength of the comparative claims.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a feasibility study of diffusion tensor MRI and spherical-deconvolution tractography on a commercial 0.064 T portable scanner. Five healthy adults were scanned with a 3D diffusion-weighted fast spin-echo sequence (18 diffusion directions, b=945 s/mm²) and with a high-field 3 T reference protocol. The authors apply retrospective corrections for B0 inhomogeneity, perform DTI and multi-tissue CSD, and generate tractograms with manual ROIs, TractSeg, and tract-orientation mapping. They report that most major white matter bundles can be retrieved in clinically tolerable scan times, and that scalar maps and fibre orientation distributions show good agreement with high-field references, based on visual comparisons of a representative subject.","tokens_in":13914,"tokens_out":3709,"duration_ms":36185,"significance":"If substantiated, this would be a valuable demonstration that tractography is feasible at ultra-low field, with implications for accessible neuroimaging in resource-limited settings. The strengths of the paper include the use of a commercial scanner, a detailed acquisition and processing pipeline, multiple tractography approaches, and a high-field validation dataset from the same participants. The feasibility of retrieving major bundles is supported by the example tractograms. However, the comparative claims of 'strong correspondence' and 'good agreement' are currently supported only by qualitative visual assessment, and several methodological approximations (B0 correction, manual noise tuning) limit the strength of the conclusions. The paper is a promising proof-of-concept that requires quantitative validation or tempered claims.","major_comments":[{"comment":"The central comparative claim that ULF scalar maps and fODFs show 'strong correspondence' and 'good agreement' with high-field references rests entirely on visual inspection of a single representative subject. No quantitative metrics (e.g., voxelwise correlation, RMSE, Dice/bundle overlap, streamline counts, or FA/MD differences) are reported for the five participants, and no group-level statistics are provided. The Discussion itself concedes that 'the encouraging results shown must be validated against high field data in a range of populations before this approach can be endorsed,' which weakens the abstract's assertion of 'reliable tractography.' Please either provide quantitative comparative analyses across the full cohort or substantially temper the wording of the claims.","section":"Abstract and Sections 3.1–3.2"},{"comment":"The B0 inhomogeneity correction assumes that the background gradient acts only as a scalar multiplicative scaling of the b-value, neglecting rotation of the effective encoding direction. The paper acknowledges that near the edge of the FOV (8 cm from isocentre) the static gradient is about 7% of the prescribed diffusion encoding gradient, where the 20:1 ratio used to justify negligible rotation does not hold. The residual spatially varying artifacts described in Section 3.3 (Figure 9) may in part reflect this neglected rotation. Please add a sensitivity analysis or apply a full spatially varying b-matrix correction to assess whether the scalar approximation biases FA, MD, or tractography in the affected regions; at minimum, state this limitation more prominently.","section":"Section 2.3.2 and Appendix Eq. (8)"},{"comment":"The RESTORE tensor fitting uses noise levels that were 'initially estimated with the DIPY estimatesigma method, then manually adjusted to maximise visual consistency.' This introduces a subjective, undocumented step into the estimation of all DTI-derived metrics and, indirectly, tractography. The claim of 'robust assessments of white matter microstructure' in the abstract requires that results are not dependent on this manual adjustment. Please report the range of adjustments applied, perform a sensitivity analysis using only the automated estimate, or use a fully automated noise estimation procedure.","section":"Section 2.3.5"}],"minor_comments":[{"comment":"The text says phase encoding was 'oversampled by 166%' and later that 'Oversampling ratios from 400% to 100% (Nyquist limit) were tested'; it would be clearer to define the oversampling factor explicitly and indicate which value was used for the reported results.","section":"Section 2.2"},{"comment":"The sentence 'this can lead to errors of 16.1% in the ADC' begins with a lowercase 'this' after a period; please correct the typo.","section":"Section 2.3.2"},{"comment":"Please specify how many voxels were manually selected for the corpus callosum response function and whether a single or multiple ROIs were used, for reproducibility.","section":"Section 2.3.4"},{"comment":"The comparison of fODFs in Figure 3 would be strengthened by reporting a quantitative measure of fODF similarity (e.g., angular correlation or cosine similarity) between ULF and HF, even if only for the representative subject.","section":"Section 3.1"}],"recommendation":"major_revision","confidential_remarks":"The paper describes work on a commercial scanner with co-authors from the manufacturer (Hyperfine) and from Siemens Healthineers, yet no conflict-of-interest statement is included; the editor may wish to request one. The claims of 'first successful demonstration' in the Discussion should also be verified against prior ISMRM abstracts that are cited. Overall, the feasibility result is interesting and likely publishable after major revisions that either provide quantitative validation or soften the comparative claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a read if you care about low-field diffusion. The genuinely new thing here is whole-brain CSD-based tractography at 64 mT on a commercial portable scanner, with same-subject 3 T comparison — that hasn't been shown before. The authors did a serious engineering job: 3D multi-shot DW-FSE, vendor deep-learning reconstruction, a B0 gradient correction derived from a phantom fieldmap, and careful artifact documentation. The tractograms of arcuate, cingulum, SLF, etc. look plausible for the core of each bundle.\n\nThe soft spot is exactly what the stress-test note flags: the abstract's 'strong correspondence' and 'good agreement' are supported by visual inspection of one representative subject out of five. There are no Dice or bundle-overlap numbers, no voxelwise FA/MD correlations or errors, no group stats. The Discussion itself concedes the results 'must be validated against high field data' before endorsement. That is a mismatch between claim strength and evidence, and it's the main reason this can't be accepted as-is. The B0 correction is a second concern: modeling the static gradient as a scalar b-value multiplier ignores encoding-direction rotation, and the authors' own numbers imply the 20:1 ratio doesn't hold across the FOV. They mention a spatially varying artifact in Fig 9, so it's not just theoretical.\n\nI'd still send it to peer review. The feasibility claim — that major bundles can be retrieved at 0.064 T in tolerable scan time — is genuinely new, plausibly supported, and useful to the low-field community. A good referee should push for quantitative agreement metrics on at least a few tracts, group-level summary, and a more careful statement of the B0 approximation's limits. The paper deserves serious engagement, and I think it will be cited.","headline":"First plausible whole-brain tractography at 64 mT, but the abstract's 'strong correspondence' claims outrun the quantitative evidence; conditionally worth engaging.","tokens_in":14404,"tokens_out":1919,"would_cite":true,"duration_ms":22373,"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":"This paper claims that a 0.064 T portable MRI scanner, corrected for its inhomogeneous field, can reconstruct most major white matter tracts in healthy adults within clinically tolerable scan times.","keywords":["ultra-low-field MRI","portable MRI","diffusion tensor imaging","tractography","spherical deconvolution","fibre orientation distribution","white matter","point-of-care imaging"],"falsifier":"Using the measured static gradient (up to 1.4 mT/m at 8 cm from isocentre), compute the full b-matrix including the rotation of the encoding gradient for the 24.9 mT/m diffusion gradients, and re-fit the diffusion tensor without the scalar-only approximation; if peripheral FA values and the hemispheric tract-density asymmetry remain unchanged, the scalar correction is sufficient, and if they shift, the paper's central tractography results are biased in those regions.","tokens_in":13364,"feed_emoji":"🧠","tokens_out":8343,"duration_ms":72574,"temperature":0.7,"pith_summary":"Ultra-low-field MRI (0.064 T) has typically been considered too noisy and too distorted for diffusion tractography. This paper claims that a portable point-of-care scanner can nevertheless reconstruct most major white matter bundles in healthy adults in about one hour, using a 3D multi-shot diffusion-weighted fast spin-echo readout, a retrospective correction for the scanner's large static magnetic-field gradient, and spherical-deconvolution fibre orientation estimation. The recovered fractional anisotropy, mean diffusivity, and fibre orientation distributions are reported to show good agreement with 3 T reference scans in the same participants. If the claim holds, white-matter microstructure and connectivity could be measured on low-cost, portable systems, bringing diffusion tractography to settings and populations that lack access to high-field MRI.","feed_headline":"Major brain pathways traced on a 0.064 T portable MRI","feed_subtitle":"A 64 mT point-of-care scanner retrieves most major white matter bundles with diffusion tractography in feasible scan times.","key_machinery":"The pipeline rests on three components. The first is a 3D multi-shot diffusion-weighted fast spin-echo (DW-FSE) sequence that collects many echoes per excitation to offset the low signal-to-noise ratio, using isotropic 3 mm voxels and a non-Cartesian centre-out phase-encoding order. The second is a retrospective correction (Eq. 2) that models the scanner's inhomogeneous static B0 field as a spatially varying scalar factor a(r) multiplying the b-value, then rescales each voxel's signal using the b=0 image; this assumes the static gradient does not rotate the effective encoding direction. The third is multi-tissue constrained spherical deconvolution to estimate fibre orientation distributions, followed by anatomically constrained tractography and prior-based bundle segmentation to filter streamlines into anatomically meaningful tracts.","core_discovery":"The central claim is that anatomically faithful tractography is achievable at 0.064 T on a commercially available portable MRI system, not just diagnostic diffusion imaging. Using 18 diffusion directions at b = 945 s/$mm^{2}$, a 3D multi-shot DW-FSE sequence, and the scalar b-value rescaling correction of Eq. (2), the authors reconstruct the arcuate, inferior longitudinal, inferior fronto-occipital, uncinate, and superior longitudinal fasciculi, the cingulum, the upper fornix, the corticospinal tract, and the corpus callosum. They also report that fibre orientation distributions from ultra-low-field data agree with high-field references, although the fODFs are blurred, show spurious grey-matter lobes, and contain a localised artifact that produces hemispheric asymmetry in tract density. The paper frames the result as the first successful demonstration of white-matter pathway reconstruction at 64 mT.","pith_inferences":["The residual hemispheric asymmetry and localised hypointensity after the scalar correction suggest that a full b-matrix treatment, including rotation of the encoding direction, would remove the remaining bias; this can be tested on the same acquisition.","A participant-specific in vivo field map, instead of a phantom-derived one, would test whether the correction's stability assumption is the limiting factor at the periphery of the field of view.","The same acquisition and correction chain could be transferred to longitudinal or bedside studies in populations that cannot be scanned at high field, but five healthy adults is not enough to establish reproducibility across ages or pathology.","Pairing ULF tractography with other low-field quantitative contrasts could yield a multi-contrast microstructural battery on a single portable device."],"forward_implications":["Portable 0.064 T MRI can produce tractograms of major association, projection, and commissural bundles within a one-hour protocol.","Diffusion-weighted fast spin-echo, with isotropic resolution and long echo trains, is a viable tractography sequence at ultra-low field.","Automated ROI and prior-based tracking methods transfer to ULF data, reducing operator dependence in tract extraction.","ULF-derived FA and MD maps align with single-shell high-field data, though peripheral white matter shows reduced anisotropy from the elevated noise floor.","Tractometry at ULF is identified as the next target, contingent on reducing the remaining field-gradient and eddy-current artefacts."],"supporting_citations":[{"why":"Defines the 3D multi-shot self-navigated diffusion-weighted fast spin-echo sequence used to acquire the data.","marker":"[11]"},{"why":"Earlier demonstration of tensor and tract estimation on the same 0.064 T hardware, which this work extends.","marker":"[4]"},{"why":"Supplies the multi-tissue constrained spherical deconvolution method used to reconstruct fibre orientation distributions.","marker":"[31]"},{"why":"Provides anatomically constrained tractography, which restricts streamlines using anatomical priors.","marker":"[17]"},{"why":"Describes automated white-matter bundle segmentation used to define the tracking ROIs.","marker":"[37]"},{"why":"Introduces tract orientation mapping, the alternative bundle-specific tracking method compared in the paper.","marker":"[38]"},{"why":"Provides the robust tensor-fitting algorithm used for the DTI scalar maps.","marker":"[25]"},{"why":"Defines the bias-field correction applied to all images before diffusion analysis.","marker":"[12]"}],"fun_headline_variants":["Portable 64 mT MRI maps key brain pathways","Tractography on a 64 mT point-of-care scanner","Ultra-low-field MRI traces major white matter tracts","Low-cost portable MRI achieves fiber tracking","64 mT scanner reconstructs major brain connections"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The correction for the scanner's inhomogeneous magnetic field assumes the field gradient only rescales the strength of diffusion encoding and does not rotate its direction; near the edge of the field of view the static gradient is large enough that this could bias every corrected map and tractogram.","fun_headline_variants_meta":{"raw":{"variants":["Portable 64 mT MRI maps key brain pathways","Tractography on a 64 mT point-of-care scanner","Ultra-low-field MRI traces major white matter tracts","Low-cost portable MRI achieves fiber tracking","64 mT scanner reconstructs major brain connections"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00026,"raw_usage":{"total_tokens":1627,"prompt_tokens":1023,"completion_tokens":604,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":639,"completion_tokens_details":{"reasoning_tokens":528}},"tokens_in":639,"tokens_out":604,"duration_ms":6176,"temperature":1.0,"reasoning_tokens":528,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T10:40:41.494753+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Using the measured static gradient (up to 1.4 mT/m at 8 cm from isocentre), compute the full b-matrix including the rotation of the encoding gradient for the 24.9 mT/m diffusion gradients, and re-fit the diffusion tensor without the scalar-only approximation; if peripheral FA values and the hemispheric tract-density asymmetry remain unchanged, the scalar correction is sufficient, and if they shift, the paper's central tractography results are biased in those regions.","supporting_citations":[{"cited_title":"Diffusion-Weighted Imaging at 0.064 TinProceedings of the International Society for Magnetic Resonance in Medicine 2022(2022)","cited_arxiv_id":null,"evidence_quote":"Defines the 3D multi-shot self-navigated diffusion-weighted fast spin-echo sequence used to acquire the data."},{"cited_title":"Tensors and Tracts at 64 mTinISMRM Workshop on Diffusion MRI: From Research to Clinic(Amsterdam, The Netherlands, 2022)","cited_arxiv_id":null,"evidence_quote":"Earlier demonstration of tensor and tract estimation on the same 0.064 T hardware, which this work extends."},{"cited_title":"D., Dhollander, T., Connelly, A","cited_arxiv_id":null,"evidence_quote":"Supplies the multi-tissue constrained spherical deconvolution method used to reconstruct fibre orientation distributions."},{"cited_title":"E., Tournier, J","cited_arxiv_id":null,"evidence_quote":"Provides anatomically constrained tractography, which restricts streamlines using anatomical priors."},{"cited_title":"& Maier-Hein, K","cited_arxiv_id":null,"evidence_quote":"Describes automated white-matter bundle segmentation used to define the tracking ROIs."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces tract orientation mapping, the alternative bundle-specific tracking method compared in the paper."},{"cited_title":"C., Jones, D","cited_arxiv_id":null,"evidence_quote":"Provides the robust tensor-fitting algorithm used for the DTI scalar maps."},{"cited_title":"J.et al.N4ITK: Improved N3 Bias Correction.IEEE transactions on medical imaging 29,1310.issn: 02780062.https : / / pmc","cited_arxiv_id":null,"evidence_quote":"Defines the bias-field correction applied to all images before diffusion analysis."}],"review_version":1}