Pith. sign in

REVIEW 4 major objections 5 minor 71 references

Deep Learning-Based Noninvasive Screening of Type 2 Diabetes with Chest X-ray Images and Electronic Health Records

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Combining chest X-ray images with time-series electronic health records in an end-to-end trained multimodal model screens for type 2 diabetes more accurately (AUROC 0.86) than chest X-rays alone, using far fewer training samples.

desk verdict First CXR+EHR T2DM screening study with a useful released pipeline; the 2.3% AUROC claim is not supported because the only CXR baseline comes from a different dataset and no same-data unimodal baseline is trained. read the letter →

arxiv 2412.10955 v1 pith:FMMCNCHH submitted 2024-12-14 cs.LG cs.CV

Sanjana Gundapaneni , Zhuo Zhi , Miguel Rodrigues This is my paper · ORCID
classification cs.LGcs.CV
keywords type2diabetesscreeningchestX-rayelectronichealthrecordsmultimodaldeeplearningtimeseriesEHRResNet-LSTMearlyfusiontransformernoninvasive
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that combining chest X-ray images with time-series electronic health records—two noninvasive data sources already captured in routine hospital care—can screen for type 2 diabetes more accurately than chest X-rays alone. The best-performing model, an end-to-end trained ResNet-LSTM that fuses a pre-trained ResNet50 image encoder with LSTM encoders for 48-hour vital-sign sequences, reaches an AUROC of 0.8616 (0.8592 without ECG), surpassing the CXR-only baseline's 0.84 by about 2.3 percentage points using only 9863 training episodes. The authors also compare an early-fusion multimodal transformer (ViLT), which reaches 0.8481 but is less robust to missing CXR inputs. If the claim holds, opportunistic, noninvasive T2DM screening could piggyback on chest radiographs that are already being performed, with no additional blood tests or patient visits.

What carries the argument

The load-bearing mechanism is the joint-fusion ResNet-LSTM encoder architecture: a pre-trained ResNet50 encodes each chest radiograph, while two single-layer LSTMs with 256 hidden units encode time-series EHR inputs (96 timesteps × 11 features, sampled every 30 minutes for 48 hours) and ECG matrices (100×12); the three representations are concatenated and passed to a fully connected binary classifier, with all encoders and the classifier trained synchronously. The paper contrasts this with the ViLT early-fusion transformer, which linearly projects EHR and ECG rows and CXR patches into 768-dimensional tokens that share a single transformer encoder. The end-to-end joint training, the authors argue, lets the model exploit cross-modal interactions dynamically instead of freezing pre-trained encoders, which is what they credit for the performance edge over the stage-wise trained variant.

What would settle it

Train and test the same models on a patient-level split (no patient in more than one partition) or on an external cohort; if the EHR+CXR AUROC drops to roughly the CXR-only level (0.84), the claimed 2.3% improvement is an artifact of episode-level data leakage rather than genuine multimodal signal.

Watch

Extended reading notes

Core claim

The paper's central discovery is that joint, end-to-end fusion of chest X-ray images with time-series EHR data improves T2DM screening over a strong image-only baseline. On the test set of 2819 episodes, the joint-fusion ResNet-LSTM attains AUROC 0.8592 (95% CI 0.8425–0.8751) with EHR+CXR, and 0.8616 (95% CI 0.8469–0.8757) when 12-lead ECG is added, against the CXR-only classifier's 0.84 (95% CI 0.83–0.85). These gains come with far fewer training samples (9863 versus 271,065 CXRs). Ablation experiments show that zeroing out the CXR in 30–70% of test samples degrades AUROC substantially (to 0.68–0.81), confirming the image modality carries signal, and that the early-fusion ViLT transformer is more noise-robust but overall weaker than the ResNet-LSTM.

Load-bearing premise

The evaluation assumes that randomly splitting ICU episodes into train, validation, and test sets keeps patients independent; the paper itself acknowledges that subsequent ICU stays of a patient are treated as separate samples even though they share the same CXR and ECG, so if the same patient appears in multiple splits, the reported AUROC is inflated.

Editorial extensions

If this is right

  • Hospitals that already perform chest radiographs could flag patients at high T2DM risk without extra procedures or blood tests.
  • The 2.3% AUROC gain over a baseline trained on 271,065 CXRs, obtained with 9863 episodes, suggests multimodal fusion is substantially more data-efficient for this screening task.
  • Adding ECG to the EHR+CXR combination yields only a small AUROC increment (0.8616 versus 0.8592), so ECG's marginal value for this screening use case is limited.
  • The released preprocessing pipeline allows other groups to reconstruct comparable multimodal T2DM datasets from public intensive-care databases, enabling external validation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The apparent gain over the image-only baseline may be driven largely by demographic and vital-sign features in the EHR (age, weight, heart rate) rather than by image content; the paper reports a missing-CXR ablation but no EHR-only baseline, so the unique contribution of imaging remains untested.
  • Because each patient's episodes share the same CXR and ECG, a random episode split lets the model see the exact same image in training and test; a patient-level split could shrink the reported advantage.
  • If patient-level validation confirms the result, the most natural deployment is in inpatient and ICU settings where CXRs are routine, before any outpatient screening program is considered.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes multimodal deep learning models for noninvasive screening of type 2 diabetes (T2DM) from chest X-ray (CXR) images, time-series electronic health records (EHR), and 12-lead ECG signals, using MIMIC-IV, MIMIC-CXR-JPG, and MIMIC-IV-ECG. Two fusion paradigms are evaluated: an early-fusion ViLT transformer and a joint-fusion ResNet-LSTM with two training strategies. The best model, ResNet-LSTMJoint, is reported to achieve AUROC 0.8616 on the EHR+CXR+ECG dataset and 0.8592 on the EHR+CXR dataset, and the paper claims a 2.3% AUROC improvement over a CXR-only baseline from the literature. Ablation studies examine lack of pre-training, noisy inputs, and missing CXR modality.

Significance. If the central claim were validated, the work would offer a useful contribution to opportunistic T2DM screening, given the ubiquity of CXR imaging. The paper has concrete strengths: it uses public benchmark databases, releases its preprocessing pipeline, reports bootstrapped confidence intervals, and includes three ablation studies that give insight into model behavior. However, the headline comparison is not a controlled experiment, and the evaluation protocol contains a patient-level data leakage mechanism. Because these issues directly affect the main claim, the current version does not provide reliable evidence for the stated improvement over CXR-only screening.

major comments (4)
  1. [§III-C and §V] The random split of episodes rather than patients violates the independence assumption for evaluation. Section III-C states that episodes are 'partitioned randomly' into train, validation, and test sets, and Section V acknowledges that 'subsequent ICU stays of a patient' are treated as separate samples sharing the same CXR and ECG. With 14,091 episodes from 7,861 patients, many patients will contribute episodes to more than one split, and the same CXR/ECG is reused for all episodes of a patient. This allows patient-specific features to leak across splits and inflates the reported AUROC. The authors should perform a patient-level split (grouping episodes by subject_id) and report test metrics under that protocol, together with the number of patients per split and the fraction of patients that appear in more than one split.
  2. [§IV-B-2, Table IV] The 2.3% improvement over the 'CXR-only baseline' is not a controlled comparison. The baseline is the externally published ResNet34 model of Pyrros et al. [20], trained on 271,065 CXRs from a different cohort and evaluated on that cohort's own test set. No CXR-only, EHR-only, or ECG-only model is trained on the DE+C or DE+C+G train splits and evaluated on the corresponding test splits. Consequently, the 2.3% gap could reflect differences in cohort composition, label prevalence, image acquisition, or evaluation protocol rather than the benefit of multimodal fusion. The authors should train unimodal baselines on the same data and splits (e.g., a ResNet50 CXR-only model, an LSTM EHR-only model, and an ECG-only model) and compare against the multimodal models under identical conditions.
  3. [§IV-C-3, Table V] The 'Missing CXR modality' ablation does not establish the diagnostic utility of CXR in the way the text claims. The experiment takes a model trained with CXR inputs and zeroes CXR at test time for 30, 50, or 70% of samples. This measures robustness to missing inputs, not the additive predictive contribution of CXR relative to a unimodal model. The statement that 'This ablation study also clearly demonstrates the diagnostic utility of CXRs' is not supported by the design, because the drop from AUROC 0.86 to 0.74 could also reflect a distribution shift introduced by zeroing a modality the model was trained to use. The paper needs a model trained without CXR at all, or a proper unimodal baseline, to support conclusions about CXR's contribution.
  4. [§III-B and §V] The label definition is not temporally aligned with the screening claim. T2DM status is assigned from T2DM ICD codes appearing anywhere in the patient's EHR, while the CXR, ECG, and EHR inputs are drawn from the same hospitalization period (e.g., the CXR is selected within a 30-day window around the first admission and final discharge). As a result, the model may be detecting an already-documented diagnosis rather than screening for undiagnosed T2DM. The paper should either restrict the label window to diagnoses occurring after the index date, or reframe the task as detection/classification of known T2DM rather than screening. This distinction is important for the clinical interpretation of the reported performance.
minor comments (5)
  1. [§III-B-2] The phrase 'resized to a width of 384 pixels on the shorter side' is ambiguous; it should probably read that the shorter side is resized to 384 pixels while maintaining the aspect ratio.
  2. [Table III] Table III reports only episode counts; patient counts per split would help readers assess the extent of patient overlap across splits.
  3. [§IV-A-2] The ResNet-LSTM training details mention 'early stopping is set to a patience level of 5 training epochs' and a learning-rate reduction factor of 0.5 on validation-loss plateau; specifying which of these triggered first in the reported runs would improve reproducibility.
  4. [Figures 1 and 2] The architecture diagrams are small and the text within them is difficult to read; higher-resolution figures or vector graphics would help.
  5. [References] Reference formatting is inconsistent: some entries include DOIs and some only URLs, and several arXiv papers lack the arXiv identifier in the visible text; aligning all references to the journal style would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the empirical screening claims are evaluated on held-out test data, and no load-bearing step reduces by construction to a fitted input or to a self-citation.

full rationale

This paper is an empirical machine-learning study rather than a derivation chain, so most circularity patterns do not apply. The central claim is that a jointly trained ResNet-LSTM model reaches AUROC 0.8616 on DE+C+G and 0.8592 on DE+C (Section IV-B-2, Table IV), with the reported 2.3% improvement measured against the published CXR-only baseline of Pyrros et al. [20] (AUROC 0.84). No equation in the paper defines the multimodal model's output in terms of that baseline, and the baseline value is an external benchmark, not a parameter fitted to the authors' data. The multimodal models are trained on a 70/10/20 random episode split and evaluated on the held-out test set, so the headline AUROC is an out-of-sample measurement rather than a quantity forced by construction. The 'Missing CXR modality' ablation (Section IV-C-3, Table V) zeroes out CXR inputs and is presented as a robustness study, not as the claimed unimodal baseline, so it does not create a fitted-input-called-prediction structure: the 2.3% comparison is simply under-controlled, which is a correctness and validity concern, not circularity. The paper's citations to prior preprocessing work [28] and MedFuse [63] are external and are used for pipeline adaptation and architecture inspiration, not to justify the paper's own conclusions; none of the cited works are authored by the present authors, so the self-citation-load-bearing pattern is absent. The acknowledged limitation that subsequent ICU stays of the same patient are treated as separate samples sharing the same CXR and ECG (Section V) is a data-leakage risk that may inflate absolute performance, but leakage is not a definitional reduction of the result to its inputs. There is no 'uniqueness theorem' imported from the authors' prior work, no ansatz smuggled in via self-citation, and no renaming of a known empirical pattern under new coordinates. Accordingly, the appropriate finding is no significant circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim depends on the MIMIC-derived labels and on the independence assumptions used in splitting. No new physical or biological entities are introduced. The listed free parameters are hyperparameters and data-construction choices tuned on the validation set; they affect the reported AUROC but are not a mechanistic model.

free parameters (5)
  • EHR sampling rate = 30 minutes
    Tuned over {15, 30, 60} minutes on the validation set.
  • EHR sampling duration = 48 hours
    Tuned over {24, 48, 72} hours on the validation set.
  • EHR imputation strategy = zero
    Tuned over {zero, mean, previous, next} on the validation set.
  • learning rate = 10^-4
    Tuned over {10^-3, 10^-4, 10^-5} on the validation set.
  • batch size = 256 for ViLT, 64 for ResNet-LSTM
    Tuned over {64, 128, 256, 512} on the validation set.
assumptions (4)
  • domain assumption ICD codes in MIMIC-IV correctly identify T2DM status
    Labels are derived from ICD-9-CM/ICD-10-CM codes (Section III-B-1); coding errors or undiagnosed patients would bias the ground truth.
  • ad hoc to paper Intensive care episodes are independent samples for splitting
    Section III-C splits episodes randomly, but one patient can contribute many episodes, so train and test share patients. This is the main load-bearing flaw.
  • ad hoc to paper A single CXR and ECG can represent all of a patient's episodes
    Section III-B-2 and III-B-3 pair the first CXR/ECG with all episodes for that patient, creating repeated inputs across samples.
  • domain assumption MIMIC-IV ICU patients generalize to the general screening population
    The cohort is single-center, ICU-based, and older (mean age about 61); the paper positions this as screening but does not validate externally.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Learning-Based Noninvasive Screening of Type 2 Diabetes with Chest X-ray Images and Electronic Health Records." pith.science (2026). https://pith.science/paper/FMMCNCHH

@misc{pith2026241210955,
  author       = {Pith},
  title        = {Pith review of: Deep Learning-Based Noninvasive Screening of Type 2 Diabetes with Chest X-ray Images and Electronic Health Records},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FMMCNCHH}},
  note         = {Machine review of arXiv:2412.10955}
}
read the original abstract

The imperative for early detection of type 2 diabetes mellitus (T2DM) is challenged by its asymptomatic onset and dependence on suboptimal clinical diagnostic tests, contributing to its widespread global prevalence. While research into noninvasive T2DM screening tools has advanced, conventional machine learning approaches remain limited to unimodal inputs due to extensive feature engineering requirements. In contrast, deep learning models can leverage multimodal data for a more holistic understanding of patients' health conditions. However, the potential of chest X-ray (CXR) imaging, one of the most commonly performed medical procedures, remains underexplored. This study evaluates the integration of CXR images with other noninvasive data sources, including electronic health records (EHRs) and electrocardiography signals, for T2DM detection. Utilising datasets meticulously compiled from the MIMIC-IV databases, we investigated two deep fusion paradigms: an early fusion-based multimodal transformer and a modular joint fusion ResNet-LSTM architecture. The end-to-end trained ResNet-LSTM model achieved an AUROC of 0.86, surpassing the CXR-only baseline by 2.3% with just 9863 training samples. These findings demonstrate the diagnostic value of CXRs within multimodal frameworks for identifying at-risk individuals early. Additionally, the dataset preprocessing pipeline has also been released to support further research in this domain.

Figures

Figures reproduced from arXiv: 2412.10955 by the authors.

Figure 1
Figure 1. FIGURE 1 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. FIGURE 2 [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

71 extracted references · 53 canonical work pages

  1. [20]

    Pyrros et al., ‘‘Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs,’’Nat

    A. Pyrros et al., ‘‘Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs,’’Nat. Commun., vol. 14, p. 4039, July 2023, doi: https://doi.org/10.1038/s41467-023-39631-x

    work page doi:10.1038/s41467-023-39631-x 2023

  2. [1]

    I. W. Campbell, ‘‘Type 2 diabetes mellitus: ‘the silent killer’,’’ Pract. Diabetes Int. , vol. 18, no. 6, pp. 187–191, Sept. 2001, doi: https://doi.org/10.1002/pdi.230

    work page doi:10.1002/pdi.230 2001

  3. [2]

    Nowakowska et al

    M. Nowakowska et al. , ‘‘The comorbidity burden of type 2 dia- betes mellitus: patterns, clusters and predictions from a large English primary care cohort,’’ BMC Med. , vol. 17, p. 145, July 2019, doi: https://doi.org/10.1186/s12916-019-1373-y

    work page doi:10.1186/s12916-019-1373-y 2019

  4. [3]

    (2007, June) The silent epidemic: An economic study of diabetes in developed and developing countries

    The Economist Intelligence Unit 2007. (2007, June) The silent epidemic: An economic study of diabetes in developed and developing countries. [Online]. Available: https://graphics.eiu.com/upload/portal/ diabetes_web.pdf

    work page 2007

  5. [4]

    Sun et al

    H. Sun et al. , ‘‘IDF Diabetes Atlas: Global, regional and country- level diabetes prevalence estimates for 2021 and projections for 2045,’’ Diabetes Res. Clin. Pract. , vol. 183, p. 109119, Jan. 2022, doi: https://doi.org/10.1016/j.diabres.2021.109119

    work page arXiv 2021

  6. [5]

    (2023, Dec.) Getting diagnosed - Type 2 diabetes

    NHS. (2023, Dec.) Getting diagnosed - Type 2 diabetes. [Online]. Avail- able: https://www.nhs.uk/conditions/type-2-diabetes/getting-diagnosed/

    work page 2023

  7. [6]

    American Diabetes Association, ‘‘Screening for Type 2 Diabetes,’’ Diabetes Care , vol. 27, no. suppl_1, pp. s11–s14, Jan. 2004, doi: https://doi.org/10.2337/diacare.27.2007.S11

    work page doi:10.2337/diacare.27.2007.s11 2004

  8. [7]

    Bergman et al

    M. Bergman et al. , ‘‘Current diagnostic criteria identify risk for type 2 diabetes too late,’’ Lancet Diabetes Endocrinol., vol. 11, no. 4, pp. 224– 226, Apr. 2023, doi: https://doi.org/10.1016/s2213-8587(23)00039-6

    work page doi:10.1016/s2213-8587(23)00039-6 2023

Show all 71 references
  1. [8]

    (2024, Feb.) Risk factors for pre-diabetes and undiagnosed type 2 diabetes in England: 2013 to 2019

    Office for National Statistics. (2024, Feb.) Risk factors for pre-diabetes and undiagnosed type 2 diabetes in England: 2013 to 2019. [Online]. Available: https://www.ons.gov.uk/peoplepopulationandcommunity/ healthandsocialcare/healthinequalities/bulletins/ riskfactorsforpredia...

    2024

  2. [9]

    (2021) IDF Diabetes Atlas 2021 – 10th edition

    International Diabetes Federation. (2021) IDF Diabetes Atlas 2021 – 10th edition. [Online]. Available: https://diabetesatlas.org/idfawp/ resource-files/2021/07/IDF_Atlas_10th_Edition_2021.pdf

    2021

  3. [10]

    Fregoso-Aparicio, J

    L. Fregoso-Aparicio, J. Noguez, L. Montesinos, and J. A. García-García, ‘‘Machine learning and deep learning predictive models for type 2 dia- betes: a systematic review,’’ Diabetol. Metab. Syndr ., vol. 13, no. 148, Dec. 2021, doi: https://doi.org/10.1186/s13098-021-00767-9

    2021 doi

  4. [11]

    Tuppad and S

    A. Tuppad and S. D. Patil, ‘‘Machine learning for diabetes clinical decision support: a review,’’Adv. Comput. Intell., vol. 2, no. 2, Apr. 2022, doi: https://doi.org/10.1007/s43674-022-00034-y

    2022 doi

  5. [12]

    W. C. Sleeman, R. Kapoor, and P . Ghosh, ‘‘Multimodal Clas- sification: Current Landscape, Taxonomy and Future Directions,’’ ACM Comput. Surv. , vol. 55, no. 7, pp. 1–31, Dec. 2022, doi: https://doi.org/10.1145/3543848

    2022 doi

  6. [13]

    F. A. C. Oliveira et al. , ‘‘Machine Learning-Based Diabetes De- tection Using Photoplethysmography Signal Features,’’ Aug. 2023, arXiv:2308.01930

    2023 arXiv

  7. [14]

    A. R. Kulkarni et al. , ‘‘Machine-learning algorithm to non-invasively detect diabetes and pre-diabetes from electrocardiogram,’’ BMJ Innov., vol. 9, pp. 32–42, 2023, doi: https://doi.org/10.1136/bmjinnov-2021- 000759

    2023 doi

  8. [15]

    Mohsen, H

    F. Mohsen, H. R. H. Al-Absi, N. A. Y ousri, N. El Hajj, and Z. Shah, ‘‘A scoping review of artificial intelligence-based methods for dia- betes risk prediction,’’ npj Digit. Med. , vol. 6, p. 197, Oct. 2023, doi: https://doi.org/10.1038/s41746-023-00933-5

    2023 doi

  9. [16]

    Tang et al

    Y . Tang et al. , ‘‘Prediction of Type II Diabetes Onset with Computed Tomography and Electronic Medical Records,’’ in Multimodal Learn. Clin. Decis. Support Clin. Image-Based Procedures , Oct. 2020, pp. 13– 23, doi: https://doi.org/10.1007/978-3-030-60946-7_2

    2020 doi

  10. [17]

    Y un et al

    J.-S. Y un et al. , ‘‘A deep learning model for screening type 2 diabetes from retinal photographs,’’ Nutr ., Metab. Cardiovasc. Dis. , vol. 32, no. 5, pp. 1218–1226, May 2022, doi: https://doi.org/10.1016/j.numecd.2022.01.010

    2022 doi

  11. [18]

    Zhang et al., ‘‘Deep-learning models for the detection and incidence prediction of chronic kidney disease and type 2 diabetes from retinal fundus images,’’Nat

    K. Zhang et al., ‘‘Deep-learning models for the detection and incidence prediction of chronic kidney disease and type 2 diabetes from retinal fundus images,’’Nat. Biomed. Eng., vol. 5, no. 6, pp. 533–545, June 2021, doi: https://doi.org/10.1038/s41551-021-00745-6

    2021 doi

  12. [19]

    (2023, May) Diagnostic Imag- ing Dataset Statistical Release

    NHS England. (2023, May) Diagnostic Imag- ing Dataset Statistical Release. [Online]. Avail- able: https://www.england.nhs.uk/statistics/wp-content/uploads/sites/2/ 2023/05/Statistical-Release-18th-May-2023-PDF-471KB-1.pdf

    2023

  13. [21]

    A. M. Sharma, ‘‘Mediastinal Fat, Insulin Resistance, and Hyperten- sion,’’ Hypertension, vol. 44, no. 2, pp. 117–118, July 2004, doi: https://doi.org/10.1161/01.hyp.0000137993.70745.82

    2004

  14. [22]

    Puttarajappa and A

    C. Puttarajappa and A. Dhoble, ‘‘Mediastinal lipomatosis as a cause of low voltage complexes on electrocardiogram and widened medi- astinum: A case report,’’ Cases J. , vol. 1, p. 171, Sept. 2008, doi: https://doi.org/10.1186/1757-1626-1-171

    2008 doi

  15. [23]

    S. Kota, S. Kota, S. Jammula, and P . R. Tripathy, ‘‘Me- diastinal widening in asymptomatic type 1 diabetes mellitus patient,’’ J. Diabetol. , vol. 3, no. 2, p. 1, May 2012. [Online]. Available: https://journals.lww.com/jodb/fulltext/2012/03020/ mediastinal_widening_in_asympto...

    2012

  16. [24]

    Bertrand, A

    A. Bertrand, A. Lewis, J. Camps, V . Grau, and B. Rodriguez, ‘‘Multi- modal characterisation of early-stage, subclinical cardiac deterioration in patients with type 2 diabetes,’’ Cardiovasc. Diabetol., vol. 23, p. 371, Oct. 2024, doi: https://doi.org/10.1186/s12933-024-02465-y

    2024 doi

  17. [25]

    Zhang et al

    K. Zhang et al. , ‘‘BiomedGPT: A Unified and Generalist Biomedical Generative Pre-trained Transformer for Vision, Language, and Multi- modal Tasks,’’ Jan. 2024, arXiv:2305.17100v2

    2024 arXiv

  18. [26]

    Hartsock and G

    I. Hartsock and G. Rasool, ‘‘Vision-Language Models for Medical Re- port Generation and Visual Question Answering: A Review,’’ Mar. 2024, arXiv:2403.02469

    2024 arXiv

  19. [27]

    P . Xu, X. Zhu, and D. A. Clifton, ‘‘Multimodal Learning With Transformers: A Survey,’’ IEEE Trans. Pattern Anal. Mach. Intell., vol. 45, no. 10, pp. 12 113–12 132, Oct. 2023, doi: https://doi.org/10.1109/TPAMI.2023.3275156

    2023

  20. [28]

    Harutyunyan, H

    H. Harutyunyan, H. Khachatrian, D. C. Kale, G. V . Steeg, and A. Galstyan, ‘‘Multitask learning and benchmarking with clini- cal time series data,’’ Sci. Data , vol. 6, p. 96, June 2019, doi: https://doi.org/10.1038/s41597-019-0103-9

    2019 doi

  21. [29]

    Noble, R

    D. Noble, R. Mathur, T. Dent, C. Meads, and T. Greenhalgh, ‘‘Risk models and scores for type 2 diabetes: systematic review,’’BMJ, vol. 343, Nov. 2011, doi: https://doi.org/10.1136/bmj.d7163

    2011 doi

  22. [30]

    Zheng, W

    T. Zheng, W. Xie, L. Xu, X. He, Y . Zhang, M. Y ou, G. Y ang, and Y . Chen, ‘‘A machine learning-based framework to identify type 2 di- abetes through electronic health records,’’ Int. J. Med. Inf. , vol. 97, pp. 120–127, Jan. 2017, doi: https://doi.org/10.1016/j.ijmedinf.2016.09.014

    2017 doi

  23. [31]

    Maniruzzaman et al

    M. Maniruzzaman et al. , ‘‘Comparative approaches for classifica- tion of diabetes mellitus data: Machine learning paradigm,’’ Comput. Methods Programs Biomed. , vol. 152, pp. 23–34, Dec. 2017, doi: https://doi.org/10.1016/j.cmpb.2017.09.004

    2017 doi

  24. [32]

    A. R. Olivera et al. , ‘‘Comparison of machine-learning algorithms to build a predictive model for detecting undiagnosed diabetes - ELSA- Brasil: accuracy study,’’Sao Paulo Med. J., vol. 135, no. 3, pp. 234–246, May 2017, doi: https://doi.org/10.1590/1516-3180.2016.0309010217

    2017

  25. [33]

    B. G. Choi, S.-W. Rha, S. W. Kim, J. H. Kang, J. Y . Park, and Y .-K. Noh, ‘‘Machine Learning for the Prediction of New-Onset Diabetes Mellitus during 5-Y ear Follow-up in Non-Diabetic Patients with Cardiovascular Risks,’’ Yonsei Med. J. , vol. 60, no. 2, pp. 191–199, Jan. 201...

    2019 doi

  26. [34]

    Z. Xie, O. Nikolayeva, J. Luo, and D. Li, ‘‘Building Risk Prediction Models for Type 2 Diabetes Using Machine Learning Techniques,’’ Prev. Chronic Dis. , vol. 16, Sept. 2019, doi: http://dx.doi.org/10.5888/pcd16.190109

    2019 doi

  27. [35]

    T. Zhu, K. Li, P . Herrero, and P . Georgiou, ‘‘Deep Learning for Diabetes: A Systematic Review,’’ IEEE J. Biomed. Health Inform., vol. 25, no. 7, pp. 2744–2757, July 2021, doi: https://doi.org/10.1109/jbhi.2020.3040225

    2021

  28. [36]

    Naz and S

    H. Naz and S. Ahuja, ‘‘Deep learning approach for diabetes prediction using PIMA Indian dataset,’’ J. Diabetes Metab. Disord. , vol. 19, pp. 391–403, June 2020, doi: https://doi.org/10.1007/s40200-020-00520-5

    2020 doi

  29. [37]

    Rahman, D

    M. Rahman, D. Islam, R. J. Mukti, and I. Saha, ‘‘A deep learn- ing approach based on convolutional LSTM for detecting dia- betes,’’ Comput. Biol. Chem. , vol. 88, p. 107329, Oct. 2020, doi: https://doi.org/10.1016/j.compbiolchem.2020.107329

    2020

  30. [38]

    M. T. García-Ordás, C. Benavides, J. A. Benítez-Andrades, H. Alaiz- Moretón, and I. García-Rodríguez, ‘‘Diabetes detection using deep learn- ing techniques with oversampling and feature augmentation,’’ Com- put. Methods Programs Biomed. , vol. 202, p. 105968, Apr. 2021, doi: h...

    2021

  31. [39]

    B. P . Nguyen et al. , ‘‘Predicting the onset of type 2 diabetes us- ing wide and deep learning with electronic health records,’’ Comput. Methods Programs Biomed. , vol. 182, p. 105055, Dec. 2019, doi: https://doi.org/10.1016/j.cmpb.2019.105055. 12 S. Gundapaneni et al.: Deep ...

    2019

  32. [40]

    Alhassan, A

    Z. Alhassan, A. S. McGough, R. Alshammari, T. Daghstani, D. Budgen, and N. Al Moubayed, ‘‘Type-2 Diabetes Mellitus Diagnosis from Time Series Clinical Data Using Deep Learning Models,’’ in Artif. Neural Netw. Mach. Learn. – ICANN 2018 , Sept. 2018, pp. 468–478, doi: https://do...

    2018 doi

  33. [41]

    Nasir and M

    T. Nasir and M. K. Malik, ‘‘SACDNet: Towards Early Type 2 Diabetes Prediction with Uncertainty for Electronic Health Records,’’ Jan. 2023, arXiv:2301.04844

    2023 arXiv

  34. [42]

    Peddinti et al

    G. Peddinti et al. , ‘‘Early metabolic markers identify potential targets for the prevention of type 2 diabetes,’’ Diabetologia, vol. 60, no. 9, pp. 1740–1750, June 2017, doi: https://doi.org/10.1007/s00125-017-4325-0

    2017 doi

  35. [43]

    Su et al., ‘‘Ten metabolites-based algorithm predicts the future devel- opment of type 2 diabetes in Chinese,’’J

    X. Su et al., ‘‘Ten metabolites-based algorithm predicts the future devel- opment of type 2 diabetes in Chinese,’’J. Adv. Res., vol. 64, pp. 131–142, Oct. 2024, doi: https://doi.org/10.1016/j.jare.2023.11.026

    2024 doi

  36. [44]

    Liu et al

    J. Liu et al. , ‘‘Metabolic and Genetic Markers Improve Prediction of Incident Type 2 Diabetes: A Nested Case-Control Study in Chinese,’’ J. Clin. Endocrinol. Metab., vol. 107, no. 11, pp. 3120–3127, Nov. 2022, doi: https://doi.org/10.1210/clinem/dgac487

    2022 doi

  37. [45]

    Ragab, A

    M. Ragab, A. S. A.-M. AL-Ghamdi, B. Fakieh, H. Choudhry, R. F. Man- sour, and D. Koundal, ‘‘Prediction of Diabetes through Retinal Images Using Deep Neural Network,’’ Comput. Intell. Neurosci. , vol. 2022, p. 7887908, June 2022, doi: https://doi.org/10.1155/2022/7887908

    2022 doi

  38. [46]

    Tallam, D

    H. Tallam, D. C. Elton, S. Lee, P . Wakim, P . J. Pickhardt, and R. M. Summers, ‘‘Fully Automated Abdominal CT Biomarkers for Type 2 Diabetes Using Deep Learning,’’ Radiology, vol. 304, no. 1, pp. 85–95, Apr. 2022, doi: https://doi.org/10.1148/radiol.211914

    2022 doi

  39. [47]

    Poplin et al

    R. Poplin et al. , ‘‘Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning,’’ Nat. Biomed. Eng. , vol. 2, pp. 158–164, Feb. 2018, doi: https://doi.org/10.1038/s41551-018-0195-0

    2018 doi

  40. [48]

    W. L. Alyoubi, W. M. Shalash, and M. F. Abulkhair, ‘‘Dia- betic retinopathy detection through deep learning techniques: A re- view,’’ Inform. Med. Unlocked , vol. 20, p. 100377, June 2020, doi: https://doi.org/10.1016/j.imu.2020.100377

    2020

  41. [49]

    D. S. W. Ting et al. , ‘‘Development and V alidation of a Deep Learning System for Diabetic Retinopathy and Related Eye Dis- eases Using Retinal Images From Multiethnic Populations With Di- abetes,’’ JAMA, vol. 318, no. 22, pp. 2211–2223, Dec. 2017, doi: https://doi.org/10.100...

    2017

  42. [50]

    T. Y . Wong et al. , ‘‘Rates of Progression in Diabetic Retinopathy Dur- ing Different Time Periods: A systematic review and meta-analysis,’’ Diabetes Care , vol. 32, no. 12, pp. 2307–2313, Dec. 2009, doi: https://doi.org/10.2337/dc09-0615

    2009 doi

  43. [51]

    V oigt et al., ‘‘Prevalence and Progression Rate of Diabetic Retinopa- thy in Type 2 Diabetes Patients in Correlation with the Duration of Diabetes,’’Exp

    M. V oigt et al., ‘‘Prevalence and Progression Rate of Diabetic Retinopa- thy in Type 2 Diabetes Patients in Correlation with the Duration of Diabetes,’’Exp. Clin. Endocrinol. Diabetes, vol. 126, no. 9, pp. 570–576, Nov. 2018, doi: https://doi.org/10.1055/s-0043-120570

    2018 doi

  44. [52]

    Johnson, L

    A. Johnson, L. Bulgarelli, T. Pollard, S. Horng, L. A. Celi, and R. Mark, ‘‘MIMIC-IV (version 2.2),’’ PhysioNet, Jan. 2023, doi: https://doi.org/10.13026/6mm1-ek67

    2023 doi

  45. [53]

    A. E. W. Johnson et al. , ‘‘MIMIC-IV , a freely accessible electronic health record dataset,’’ Sci. Data , vol. 10, p. 1, Jan. 2023, doi: https://doi.org/10.1038/s41597-022-01899-x

    2023 doi

  46. [54]

    Johnson et al

    A. Johnson et al. , ‘‘MIMIC-CXR-JPG - chest radiographs with structured labels (version 2.0.0),’’ PhysioNet, Nov. 2019, doi: https://doi.org/10.13026/8360-t248

    2019 doi

  47. [55]

    A. E. W. Johnson et al. , ‘‘MIMIC-CXR: A large publicly available database of labeled chest radiographs,’’ Jan. 2019, arXiv:1901.07042

    2019 arXiv

  48. [56]

    Gow et al

    B. Gow et al. , ‘‘MIMIC-IV -ECG: Diagnostic Electrocardiogram Matched Subset (version 1.0),’’ PhysioNet, Sept. 2023, doi: https://doi.org/10.13026/4nqg-sb35

    2023 doi

  49. [57]

    A. L. Goldberger et al. , ‘‘PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals,’’ Circulation, vol. 101, no. 23, pp. e215–e220, 2000, doi: https://doi.org/10.1161/01.CIR.101.23.e215

    doi

  50. [58]

    A. D. Association, ‘‘2. Classification and Diagnosis of Diabetes: Stan- dards of Medical Care in Diabetes—2020,’’ Diabetes Care, vol. 43, pp. S14–S31, Dec. 2019, doi: https://doi.org/10.2337/dc20-S002

    2020 doi

  51. [59]

    Y . T. Wondmkun, ‘‘Obesity, Insulin Resistance, and Type 2 Dia- betes: Associations and Therapeutic Implications,’’ Diabetes, Metab. Syndr . Obes.: Targets Ther ., vol. 13, pp. 3611–3616, Oct. 2020, doi: https://doi.org/10.2147/DMSO.S275898

    2020 doi

  52. [60]

    S. H. Ju and H.-S. Yi, ‘‘Implication of Sex Differences in Visceral Fat for the Assessment of Incidence Risk of Type 2 Diabetes Melli- tus,’’ Diabetes Met. J. , vol. 46, no. 3, pp. 414–416, May 2022, doi: https://doi.org/10.4093/dmj.2022.0089

    2022

  53. [61]

    Y ang, Y

    B. Y ang, Y . Zhu, X. Lu, and C. Shen, ‘‘A Novel Compos- ite Indicator of Predicting Mortality Risk for Heart Failure Pa- tients With Diabetes Admitted to Intensive Care Unit Based on Machine Learning,’’ Front. Endocrinol. , vol. 13, June 2022, doi: https://doi.org/10.3389/fen...

    2022

  54. [62]

    (2019, July) Chest X-ray Quality

    Lloyd-Jones, Graham. (2019, July) Chest X-ray Quality. [On- line]. Available: https://www.radiologymasterclass.co.uk/tutorials/chest/ chest_quality/chest_xray_quality_projection

    2019

  55. [63]

    Hayat, K

    N. Hayat, K. J. Geras, and F. E. Shamout, ‘‘MedFuse: Multi-modal fusion with clinical time-series data and chest X-ray images,’’ in Proc. 7th MLHC , vol. 182, Aug. 2022, pp. 479–503. [Online]. Available: https://proceedings.mlr.press/v182/hayat22a.html

    2022

  56. [64]

    (2021, Nov.) International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM)

    Centers for Disease Control and Prevention. (2021, Nov.) International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). [Online]. Available: https://archive.cdc.gov/www_cdc_ gov/nchs/icd/icd9cm.htm

    2021

  57. [65]

    (2024, June) International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM)

    ——. (2024, June) International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM). [Online]. Available: https://www.cdc.gov/nchs/icd/icd-10-cm/index.html

    2024

  58. [66]

    Kligfield et al

    P . Kligfield et al. , ‘‘Recommendations for the Standard- ization and Interpretation of the Electrocardiogram,’’ Cir- culation, vol. 115, no. 10, p. 1306–1324, Feb. 2007, doi: https://doi.org/10.1161/CIRCULA TIONAHA.106.180200

    2007 doi

  59. [67]

    W. Kim, B. Son, and I. Kim, ‘‘ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision,’’ in Proc. 38th Int. Conf. Mach. Learn., vol. 139, July 2021, pp. 5583–5594. [Online]. Available: https://proceedings.mlr.press/v139/kim21k.html

    2021

  60. [68]

    Dosovitskiy et al

    A. Dosovitskiy et al. , ‘‘An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,’’ in Int. Conf. Learn. Represent., Jan. 2021. [Online]. Available: https://openreview.net/ forum?id=YicbFdNTTy

    2021

  61. [69]

    V aswani et al

    A. V aswani et al. , ‘‘Attention is All you Need,’’ in Adv. Neural Inf. Process. Syst., vol. 30, June 2017. [Online]. Available: arXiv:1706.03762

    2017 arXiv

  62. [70]

    Harrigian, T

    K. Harrigian, T. Tang, A. Gonzales, C. X. Cai, and M. Dredze, ‘‘An Eye on Clinical BERT: Investigating Language Model Generalization for Diabetic Eye Disease Phenotyping,’’ Nov. 2023, arXiv:2311.08687

    2023 arXiv

  63. [71]

    Zhou et al., ‘‘Understanding The Robustness in Vision Transformers,’’ in Proc

    D. Zhou et al., ‘‘Understanding The Robustness in Vision Transformers,’’ in Proc. 39th Int. Conf. Mach. Learn. , vol. 162, July 2022, pp. 27 378–27 394. [Online]. Available: https://proceedings.mlr.press/v162/ zhou22m.html 13

    2022

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.