REVIEW 4 major objections 4 minor 1 references
Lifespan Pancreas Morphology for Control vs Type 2 Diabetes using AI on Largescale Clinical Imaging
T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash
Pith's one-line read Diabetes shifts pancreas aging on 10 of 13 shape-and-size measures
desk verdict A large lifespan reference for pancreas morphology, but the CT/MRI measurement gap needs a thorough sensitivity analysis before the T2D trend claims land. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The analysis rests on automated pancreas segmentation followed by extraction of 13 morphological features, including volume, surface area, and shape descriptors, all computed after resampling scans to 3 mm isotropic resolution so CT and MRI can be compared. Lifespan trends are modeled with GAMLSS, generalized additive models for location, scale, and shape, which lets the model capture how the whole distribution of each feature changes with age, sex, and diabetes status rather than only the mean. The diabetes comparison is carried out on 1,350 age- and sex-matched patients, 675 per group.
What would settle it
Take patients who have had both abdominal CT and MRI within a short window, run the same segmentation pipeline, and compare per-patient morphology differences between modalities. If those differences are as large as the reported type 2 diabetes versus control differences, the lifespan and diabetes trends would need to be re-estimated separately by modality; a CT-only matched analysis of the 675 pairs would be a simpler first check.
Extended reading notes
Core claim
The paper's central claim is that type 2 diabetes alters the way the pancreas ages in both size and shape. Using 675 matched pairs of diabetic and non-diabetic patients scanned with CT or MRI, the authors found that age trends for 10 of 13 pancreas morphology features, spanning volume and shape descriptors, were significantly different between groups after adjusting for confounders and correcting for multiple comparisons. They further conclude that the pancreas is smaller in type 2 diabetes, and they present normative lifespan trends for 13 morphological features from non-diabetic controls as a reference.
Load-bearing premise
The load-bearing assumption is that CT and MRI scans, after resampling to 3 mm isotropic voxels, yield comparable pancreas morphology measurements; the paper's own finding that MRI and CT give different measurements means this pooling assumption may be the first link to break.
Editorial extensions
If this is right
- Lifespan pancreatic morphology can be treated as a measurable clinical quantity rather than a coarse radiologic impression.
- Type 2 diabetes is associated with divergent aging trajectories in most pancreatic size and shape features, so pancreatic involvement is not limited to volume loss.
- The non-diabetic reference curves give future studies a baseline for detecting deviations in other pancreatic diseases.
- Because MRI and CT differ, studies mixing modalities need explicit harmonization or separate modality-specific models.
- Image-derived morphology could support early-detection or monitoring research in type 2 diabetes.
Reading between the lines
- An implication left implicit: if CT and MRI are not interchangeable, the reported diabetes differences could be partly confounded by modality unless the models included modality as a covariate; this could be checked by re-running the comparisons on CT-only patients.
- A natural testable extension is to ask whether pancreas shape deviations precede clinical diabetes diagnosis in longitudinal scans, which would make morphology a potential early marker.
- Because these scans are obtained for unrelated clinical reasons, AI-measured pancreatic morphology could become an opportunistic screening signal for metabolic risk at no extra radiation or patient burden.
- Shape features such as sphericity or compactness may track pancreatic fat and fibrosis better than volume alone, linking this method to non-diabetic pancreatic disease.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes an AI-based pipeline to segment the pancreas in clinically acquired abdominal CT and MRI scans, resample to 3 mm isotropic resolution, and extract 13 morphological features across ages 0–90. Using GAMLSS regression on 1,350 patients matched for age, sex, and type 2 diabetes status, it reports that the aging trends for 10 of 13 features differ significantly between T2D patients and non-diabetic controls (p < 0.05 after multiple-comparison correction), and that the pancreas is smaller in T2D. The abstract also reports that MRI yielded different pancreas measurements than CT with the same AI method. The supplied full text is largely unreadable, so this assessment is based on the abstract and visible table/figure fragments.
Significance. If the central claim holds, this would be a valuable large-scale clinical-imaging reference for pancreas morphology and a potentially useful AI-derived biomarker for type 2 diabetes. The use of a clinically acquired cohort spanning the lifespan, a matched design, and flexible GAMLSS models to characterize nonlinear aging trends are strengths. The study also explicitly attempts to assess modality consistency, which is an important design consideration. However, the current evidence is insufficient to establish the central claim: the abstract reports no effect sizes or confidence intervals, provides no segmentation validation, and the acknowledged CT/MRI measurement difference is not shown to be controlled in the diabetes comparison. These are not mere presentation issues; they bear directly on whether the reported diabetes-specific aging trends reflect biological morphology rather than acquisition-driven bias.
major comments (4)
- [Abstract, Results] The paper states: 'MRI appeared to yield different pancreas measurements than CT using our AI-based method.' This is direct evidence of a systematic modality-dependent measurement bias. The central diabetes comparison pools CT and MRI scans after 3 mm isotropic resampling, but the abstract does not state whether the 1,350 matched patients were matched on modality, whether modality was included as a covariate or interaction in the GAMLSS models, or whether modality-stratified sensitivity analyses were performed. In clinically acquired data, CT versus MRI mix can plausibly differ by age and diabetes status. If so, the reported differences in aging trends for 10 of 13 features could be artifacts of modality rather than diabetes. This concern is load-bearing for the central claim and must be resolved, e.g., by reporting modality-stratified trends and a model that includes modality-by-age and
- [Abstract, Results] The only quantitative evidence for the central claim is 'p < 0.05 after multiple comparisons corrections' for 10 of 13 features. No effect sizes, confidence intervals, or model coefficients are reported. With 1,350 patients, small systematic measurement differences can be statistically significant yet biologically negligible. The paper should report adjusted mean differences or standardized effect sizes with confidence intervals for each morphological feature, not only p-values, so that the magnitude and direction of the diabetes-associated changes can be evaluated.
- [Abstract, Approach and Results] The morphological features are derived from automated pancreas segmentation, but the abstract reports no validation of the segmentation accuracy. Without Dice/Jaccard or similar metrics on a held-out set, ideally stratified by modality, age, and sex, it is impossible to distinguish true biological differences from segmentation artifacts. The reported CT/MRI discrepancy could also reflect modality-dependent segmentation errors rather than measurement differences inherent to the modalities. A validation subsection with quantitative segmentation performance is essential before the lifespan morphology trends can be interpreted.
- [Abstract, Approach and Conclusions] The normative aging reference and the diabetes comparison are estimated from the same 1,350-patient matched cohort. The 'deviation from normative aging' is therefore a within-cohort contrast, not an out-of-sample or externally validated norm. This does not invalidate the diabetes comparison, but it weakens the claim of providing a 'normative' reference. The authors should either validate the reference on an independent cohort of non-diabetic patients or explicitly reframe the claims as descriptive of the non-diabetic subpopulation of this cohort.
minor comments (4)
- [Abstract, Results] Specify the multiple-comparison correction procedure (e.g., Bonferroni, Benjamini-Hochberg FDR) rather than the generic phrase 'after multiple comparisons corrections.'
- [Abstract, Approach] The abstract mentions 'ages 0 to 90,' but no age distribution, age-stratified sample sizes, or coverage information is visible. Reporting the age distribution by sex and diabetes status would help assess the reliability of lifespan trends at the extremes.
- [Full text / tables] The supplied full text is not decipherable in the provided encoding, and most table and figure captions are unreadable. The authors should ensure that the submitted version has readable text and that all figures/tables include self-contained captions, especially for the GAMLSS model specifications and the modality comparison results.
- [Abstract, Conclusions] The conclusion 'the pancreas is smaller in type 2 diabetes' appears in the abstract, but no quantitative magnitude (e.g., adjusted mean volume difference or percentage difference) is reported. A numeric effect size would make the claim more informative and actionable.
Circularity Check
No significant circularity; the T2D-versus-control aging comparison is a direct case-control statistical contrast, not a prediction that reduces to its fitted inputs.
full rationale
The abstract's central claim is that, after confounder adjustment, aging trends for 10 of 13 morphological features differ significantly between T2D patients and non-diabetic controls (p < 0.05 after multiple-comparison correction). This is a standard case-control statistical inference: the GAMLSS model is fitted on a cohort of 675 T2D patients and 675 matched controls, and the reported differences are model contrasts between the two groups. No fitted parameter is renamed as a 'prediction,' and the T2D deviation is not defined as the same quantity that was fitted. Even if the 'normative' reference is derived from the 675 controls within the same 1,350-patient cohort, comparing the other group against that reference is an ordinary epidemiological design, not circularity-by-equivalence. The result may need external validation and may be affected by the acknowledged CT-versus-MRI measurement difference ('MRI appeared to yield different pancreas measurements than CT using our AI-based method'), but that is a validity/confounding concern, not a circular reduction. No load-bearing self-citation or uniqueness-importing argument is visible in the readable portions of the manuscript, and no specific equation or fitted parameter can be quoted that reduces the reported aging-trend differences to their own inputs. Therefore, under the stated hard rules, the appropriate finding is no significant circularity.
Assumptions & free parameters
free parameters (1)
- GAMLSS model parameters (mu, sigma, nu, tau smoothing) =
not disclosed in abstract
assumptions (2)
- domain assumption Automated segmentation of pancreas from CT and MRI at 3mm isotropic resampling is sufficiently accurate for morphological measurement.
- domain assumption Matching on age, sex, and type 2 diabetes status sufficiently controls confounders.
Cite this review
Pith. "Pith review of Lifespan Pancreas Morphology for Control vs Type 2 Diabetes using AI on Largescale Clinical Imaging." pith.science (2026). https://pith.science/paper/PY2RA5SV
@misc{pith2026250814878,
author = {Pith},
title = {Pith review of: Lifespan Pancreas Morphology for Control vs Type 2 Diabetes using AI on Largescale Clinical Imaging},
year = {2026},
howpublished = {\url{https://pith.science/paper/PY2RA5SV}},
note = {Machine review of arXiv:2508.14878}
}
read the original abstract
Purpose: Understanding how the pancreas changes is critical for detecting deviations in type 2 diabetes and other pancreatic disease. We measure pancreas size and shape using morphological measurements from ages 0 to 90. Our goals are to 1) identify reliable clinical imaging modalities for AI-based pancreas measurement, 2) establish normative morphological aging trends, and 3) detect potential deviations in type 2 diabetes. Approach: We analyzed a clinically acquired dataset of 2533 patients imaged with abdominal CT or MRI. We resampled the scans to 3mm isotropic resolution, segmented the pancreas using automated methods, and extracted 13 morphological pancreas features across the lifespan. First, we assessed CT and MRI measurements to determine which modalities provide consistent lifespan trends. Second, we characterized distributions of normative morphological patterns stratified by age group and sex. Third, we used GAMLSS regression to model pancreas morphology trends in 1350 patients matched for age, sex, and type 2 diabetes status to identify any deviations from normative aging associated with type 2 diabetes. Results: When adjusting for confounders, the aging trends for 10 of 13 morphological features were significantly different between patients with type 2 diabetes and non-diabetic controls (p < 0.05 after multiple comparisons corrections). Additionally, MRI appeared to yield different pancreas measurements than CT using our AI-based method. Conclusions: We provide lifespan trends demonstrating that the size and shape of the pancreas is altered in type 2 diabetes using 675 control patients and 675 diabetes patients. Moreover, our findings reinforce that the pancreas is smaller in type 2 diabetes. Additionally, we contribute a reference of lifespan pancreas morphology from a large cohort of non-diabetic control patients in a clinical setting.
Reference graph
Works this paper leans on
-
[1]
���������� ����������� ���������� ���� ���� ���������� ���� ����� ������ �������� ��� � ������ ��� ��� � �������� ���� ������� ���� ��������� ���� ����� �� ���� ������� ��� �������� �� ���� ��� ��������� ���� ����������� ��� ������������ ���� � ��������� ��������� ������������ ����������� ��������� ����� ��������� ����������� �������� ����� ������� ������...
arXiv 2025
Reviewed August 5, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.