MEDFORM: A Foundation Model for Contrastive Learning of CT Imaging and Clinical Numeric Data in Multi-Cancer Analysis

Daeun Jung; Jaehyeok Jang; Sooyoung Jang; Yu Rang Park

REVIEW 4 major objections 5 minor 30 references

MEDFORM: A Foundation Model for Contrastive Learning of CT Imaging and Clinical Numeric Data in Multi-Cancer Analysis

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A two-stage contrastive pretraining scheme can align CT scans with clinical numeric data, improving cancer classification and few-shot performance across breast, colorectal, and lung cancer tasks.

desk verdict Adapts TANGLE to CT + clinical data, but the reported evidence is undercut by a likely train/test overlap and undisclosed clinical features. read the letter →

arxiv 2501.13277 v1 pith:MCRSLPFP submitted 2025-01-22 cs.CV

Daeun Jung , Jaehyeok Jang , Sooyoung Jang , Yu Rang Park This is my paper · ORCID

classification cs.CV

keywords medicalfoundationmodelcomputedtomographycontrastivelearningmultipleinstancecross-modalalignmentcancerclassificationfew-shotclinicaltabulardata

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

MEDFORM tests a two-stage recipe for building a medical foundation model from CT images and tabular clinical records without expert annotations. First, a CT slice encoder is pretrained with self-supervised contrastive learning on unlabeled slices; then the slice embeddings are pooled by an attention-based multiple-instance-learning module into a patient-level representation, and that representation is aligned with a clinical-data embedding through a symmetric cross-modal contrastive loss. The paper reports that this aligned representation outperforms a unimodal CT encoder and a feature-concatenation baseline across three cancer-specific tasks, and that the gains survive with only a handful of labeled examples. If the result holds, large archives of paired CT scans and routine clinical records could serve as cheap supervision for label-efficient cancer classifiers.

What carries the argument

The load-bearing object is the aligned embedding space produced by the dual pretraining schedule. Stage one trains a ResNet-50 slice encoder with SimCLR-style contrastive learning on 2D axial CT slices, using circular masking and Hounsfield-unit windowing as preprocessing. Stage two computes a patient-level CT vector by attention-based multiple-instance learning pooling over the slice embeddings, then minimizes a symmetric cross-modal contrastive loss against an MLP embedding of clinical numeric features, so matched CT-clinical pairs have high cosine similarity in a shared space. At inference, only the CT side is used for downstream linear-probe classification; the clinical encoder is present to shape the representation.

What would settle it

Audit the clinical feature columns used in the alignment experiment and remove any column that is the target label or a deterministic transform of it; if the AUROC gap over the unimodal CT baseline collapses, the reported advantage rests on label information in the clinical input rather than on cross-modal alignment.

Watch

Extended reading notes

Core claim

The central claim is that guiding CT image representation learning with clinical numeric data, rather than simply concatenating the two modalities at decision time, produces a shared embedding with more usable signal for cancer classification. Using the dual pretraining strategy, MEDFORM reports higher AUROC than both baselines on all three tasks, with breast T-stage rising from 0.6390 to 0.7042, colorectal bilobar disease from 0.4825 to 0.5336, and lung histologic stage from 0.6000 to 0.6658. The paper interprets these results as showing that contrastive cross-modal alignment integrates imaging and clinical information more effectively than feature concatenation, and that the learned CT representation retains useful structure when only a few labeled examples are available.

Load-bearing premise

The computer's clinical input contains no hidden copy of the exact answer, such as tumor stage or bilobar status, that it is later asked to predict.

Editorial extensions

If this is right

Because the CT side alone is used for downstream classification, the pretraining promises a label-efficient pipeline: a hospital with CT archives and matching clinical records could align representations once, then reuse them for several downstream tasks.
The attention-based multiple-instance aggregation removes the need for slice-level annotations, so the same encoder can consume volumetric CTs of arbitrary slice count.
The reported few-shot numbers imply that with 5 to 10 labeled examples per task, the aligned representation still gives usable accuracy, which could reduce expert annotation cost for new cancer staging tasks.
The gap between contrastive alignment and feature concatenation suggests that learning a shared space changes the representation itself, not just the input to the final classifier.

Reading between the lines

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

The paper never lists the clinical variables entering the encoder, so the strongest check on the result is to audit those columns; if T-stage, bilobar status, or histologic stage appear as inputs, the contrastive alignment could simply copy the label into the CT embedding and the reported multimodal gain would be circular.
The experiments are restricted to the three cancers used in pretraining, so the phrase foundation model should be read narrowly; a genuine transfer test on an unseen cancer type or an unseen task remains an open extension.
A direct ablation that removes all outcome-related clinical columns and reruns the Table 1 comparisons would separate genuine multimodal signal from label leakage; the authors do not report such an ablation.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: this is a straightforward adaptation of TANGLE to CT slices plus clinical numeric data, not a new mechanism. The paper is worth a look because the problem is real — using tabular clinical data as weak supervision for CT representation learning is a sensible idea — and the authors put together multi-cancer data and released code. But the experimental section, as written, does not support the headline claim that MEDFORM “consistently outperforms all baselines.”

What’s good: the MIL treatment of volumetric CT is appropriate, the combination of SimCLR pretraining with cross-modal contrastive alignment is clearly described, and the choice to benchmark against feature concatenation is the right baseline. The code availability is a plus.

Where it falls down. First, the data split description in Section 4.1 strongly suggests that the same set of patients/slices was used both for multimodal alignment training and for testing. If so, the linear probe evaluates on data whose representations were aligned during pretraining, which can inflate results through memorization even without label leakage. That is a load-bearing flaw. Second, the clinical feature list is never given, so the possibility that the target label (T-stage, histologic stage, bilobar disease) appears among the input columns remains live. That would make the multimodal gains circular. Third, Table 1 itself contradicts the “consistently outperforms” claim: on NSCLC histologic stage, feature concatenation beats contrastive learning on accuracy (0.8420 vs 0.8208), and most AUROC differences are within one standard deviation. The zero standard deviation for that baseline is also a red flag. Fourth, Table 2 is labeled as the authors’ model but the model column says TANGLE, which suggests a copy-paste error and leaves the few-shot results ambiguous.

In short, the core idea is plausible but the evidence is not. The paper needs a corrected train/test split, a full disclosure of the clinical variables, and a more honest reading of Table 1. The novelty is limited, but the flaws, not the novelty, are what sink the current version.

Who this is for: someone working on multimodal medical imaging pre-training, especially those considering clinical tabular data as a supervision source. They might learn from the method’s design and from the pitfalls this paper illustrates.

Recommendation for peer review: send it out. The methodological concerns are serious but could be addressed in revision, and the paper touches a question worth refereeing. My own verdict would be reject in the present form, but with an invitation to resubmit if the data split and variable disclosure are fixed.

Referee Report

4 major / 5 minor

Summary. The paper proposes MEDFORM, a multimodal pre-training framework that combines CT imaging with clinical numeric data for multi-cancer classification. The method has three components: a ResNet-50 slice encoder trained with SimCLR, an attention-based multiple instance learning (ABMIL) module that aggregates slice embeddings into a volumetric CT representation, and a cross-modal contrastive learning objective (symmetric CLIP-style loss) that aligns CT representations with representations of clinical numeric features encoded by an MLP. The authors evaluate on three cancer datasets from TCIA: breast cancer T-stage, colorectal cancer bilobar disease, and NSCLC histologic stage, comparing unimodal CT, feature concatenation, and their contrastive model. They also report few-shot learning results. The central claim is that the dual pre-training strategy 'consistently outperforms all baselines' (Section 4.2) and 'maintains robust performance in few-shot learning scenarios' (Abstract).

Significance. If the claims were valid, MEDFORM would be a useful contribution to multimodal medical foundation models for CT and tabular clinical data, and the use of MIL for volumetric CT aggregation is a reasonable design choice. The paper makes code available and uses public TCIA data, which are positive features. However, the reported experimental evaluation is not reliable: the same data are used for multimodal alignment training and for testing, the clinical variables are undisclosed so label leakage cannot be ruled out, and the paper's own Table 1 contradicts the 'consistently outperforms' claim. As a result, the current evidence does not support the central claims, and the significance of the contribution cannot be assessed from the reported results.

major comments (4)

[Section 4.1] The data split is not a held-out split. For each cancer type, the text states that 'the rest ... was used in training multimodal alignment and as test set' (NSCLC, breast, and colorectal descriptions). This means the same patients/slices used to train the cross-modal alignment are also used to evaluate the downstream linear probe in Section 4.2. Any reported AUROC/ACC can therefore reflect memorization of the training data rather than generalization. This is a load-bearing flaw: without a proper train/validation/test split, none of the performance numbers in Tables 1 and 2 support the paper's claims.
[Sections 3.2 and 4.1] The paper never lists the clinical variables used as input to the clinical encoder. The downstream tasks are T-stage, bilobar disease, and histologic stage. If any of these target labels (or direct proxies, such as stage or grade) appear as a column in the clinical data, the cross-modal contrastive objective lets the CT embedding learn to encode the label from the paired clinical vector. The downstream linear probe could then read the label from the CT representation, making the multimodal gain circular. The authors must disclose the full clinical feature list and explicitly exclude label columns, or demonstrate that no such leakage is possible.
[Table 1 and Section 4.2] The claim that the contrastive model 'consistently outperforms all baselines' is contradicted by the reported numbers. For NSCLC histologic stage, the feature concatenation model achieves ACC 0.8420 while the contrastive model achieves ACC 0.8208. Moreover, most AUROC differences in Table 1 are within one standard deviation, and the colorectal unimodal AUROC of 0.4825 is below the chance level of 0.5. The paper provides no statistical significance testing. These inconsistencies undermine the central claim even before considering the split and leakage issues.
[Table 2] The few-shot experiments report only the proposed contrastive model; there are no unimodal or feature-concatenation baselines for the same few-shot protocol, and the evaluation again uses the overlapping data split. Without a comparison baseline and a proper held-out split, the statement that the model 'maintains robust performance in few-shot learning scenarios' is not supported.

minor comments (5)

[Abstract and Section 3.1] The number of colorectal CT slices used for CT-SimCLR pre-training is reported as 10,393 in the Abstract and Section 4.1 but as 10,323 in Section 3.1; these should be reconciled.
[Equation (1)] Equation (1) is not typeset correctly: the temperature τ appears inside the exponent without a clear definition of normalized cosine similarity. The authors should specify that embeddings are L2-normalized and define the temperature parameter explicitly.
[Throughout] There are numerous typographical errors, including 'CT slide' instead of 'CT slice' in Section 3.3 and grammatical issues in Sections 4.1 and 5. A careful language edit is needed.
[Table 2] The few-shot table does not report standard deviations for the k=1 breast cancer AUROC or provide a baseline such as training a linear probe on the unimodal CT representation; adding these would help interpret the results.
[Section 3.1] The description of the CT-SimCLR pre-training dataset is inconsistent: the paper says '141,171 NSCLC slices, 8,100 breast cancer slices, and 10,323 colorectal cancer images' in Section 3.1, but Section 4.1 reports different cohort totals. The exact composition of the pre-training set should be clarified.

Circularity Check

2 steps flagged · score 6.0 of 10

Test-set contamination in multimodal alignment training and undisclosed clinical features make the reported contrastive gains non-independent.

fitted input called prediction [Section 4.1 (Dataset split descriptions for NSCLC, breast, and colorectal cancer)]
"Among these, 101,117 slices from NSCLC Radiomics, COVID-19-AR, MIDRC-RICORD-1B were used for CT-SimCLR pre-training and the rest from NSCLC Radiogenomics were used in training multimodal alignment and as test set. ... part of ACRIN-FLT-Breast was used in CT-SimCLR pretraining, and rest of it was used in training multimodal alignment and as test set."

The same patient/slice set that is called the 'test set' is also used to train the cross-modal alignment in Eq. (1). The downstream linear probe therefore evaluates on representations whose alignment was learned from those exact test examples. The contrastive model's reported AUROC/ACC are not independent generalization estimates; the model was fitted to the test set before evaluation, while the unimodal baseline did not receive this test-set alignment pretraining. This makes the central claim that the contrastive model 'consistently outperforms all baselines' an artifact of the data split.
self definitional [Section 3.2-3.3, Eq. (1); task definitions in Section 4.2]
"Given a set of clinical datasets, denoted as t_i ∈ R^{N_C}, where N_C represents the dimension of clinical features, we employ a multilayer perceptron (MLP) for encoding ... To align these cross-modal representations is shared embedding space, we minimize the following symmetric contrastive loss function."

The paper never enumerates the clinical features in t_i. The downstream tasks are breast cancer T-stage classification, colorectal cancer bilobar disease assessment, and NSCLC histologic staging, which are standard clinical variables. If any of these target labels (or a direct proxy) is a column of t_i, then the clinical embedding c_i encodes the label; Eq. (1) trains the CT embedding s_i to match c_i, so the label is transferred into the CT representation. The downstream linear probe then reads from an input that already contains the target label. Since the paper does not state that label columns are excluded from the clinical data, the reported multimodal improvement is circular by construction under the stated method.

full rationale

The central comparison is compromised by the data split described in Section 4.1: the same 'rest' set is used both for training the multimodal alignment and as the test set, so the contrastive model's test predictions are fitted to the test data. This alone undermines the claim that multimodal contrastive learning 'consistently outperforms all baselines.' Additionally, the clinical feature list is never disclosed, and because the prediction targets (T-stage, bilobar disease, histologic stage) are themselves typical clinical variables, the cross-modal alignment can in principle make the CT embedding mirror the label through the clinical encoder, which would make the multimodal gain circular. No self-citation chain or imported uniqueness theorem is present; the SimCLR and MIL components are standard external methods. The absence of any stated exclusion of label columns or separate alignment/test split means the central claim is only partially supported by independent evidence, so a score of 6 reflects partial circularity rather than full definitional equivalence.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new physical or conceptual entities. The free parameters are standard hyperparameters. The critical axioms are the clean separation of labels from clinical features and clean patient-level data splits, neither of which is verified.

free parameters (2)

temperature tau in Eq. (1) = not reported
The symmetric contrastive loss uses a temperature scaling factor tau, a hand-chosen hyperparameter, but the paper does not specify its value.
embedding dimension d = not reported
The dimension of the aligned CT and clinical embeddings is not stated; it is a design choice.

assumptions (3)

domain assumption Clinical features are informative about the target label without containing the label itself
The whole method relies on clinical data providing complementary supervision, but the paper never lists the features or excludes the target labels (T-stage, bilobar disease, histologic stage). This is load-bearing.
domain assumption Patient-level split between pretraining and evaluation
Section 4.1 is ambiguous; 'the rest' could mean the same cohort is used for both training and test, and there is no explicit statement that patients are partitioned. If patients overlap, the SSL pretraining can memorize test patients.
domain assumption ResNet50 trained with SimCLR on 2D slices transfers to volumetric CT via MIL pooling
The authors assume that slice-level features learned via SimCLR are a good basis for patient-level aggregation with ABMIL, which is a standard but unproven assumption in this setup.

how reviews work

0 comments

Cite this review

Pith. "Pith review of MEDFORM: A Foundation Model for Contrastive Learning of CT Imaging and Clinical Numeric Data in Multi-Cancer Analysis." pith.science (2026). https://pith.science/paper/MCRSLPFP

@misc{pith2026250113277,
  author       = {Pith},
  title        = {Pith review of: MEDFORM: A Foundation Model for Contrastive Learning of CT Imaging and Clinical Numeric Data in Multi-Cancer Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MCRSLPFP}},
  note         = {Machine review of arXiv:2501.13277}
}

read the original abstract

Computed tomography (CT) and clinical numeric data are essential modalities for cancer evaluation, but building large-scale multimodal training datasets for developing medical foundation models remains challenging due to the structural complexity of multi-slice CT data and high cost of expert annotation. In this study, we propose MEDFORM, a multimodal pre-training strategy that guides CT image representation learning using complementary information from clinical data for medical foundation model development. MEDFORM efficiently processes CT slice through multiple instance learning (MIL) and adopts a dual pre-training strategy: first pretraining the CT slice feature extractor using SimCLR-based self-supervised learning, then aligning CT and clinical modalities through cross-modal contrastive learning. Our model was pre-trained on three different cancer types: lung cancer (141,171 slices), breast cancer (8,100 slices), colorectal cancer (10,393 slices). The experimental results demonstrated that this dual pre-training strategy improves cancer classification performance and maintains robust performance in few-shot learning scenarios. Code available at https://github.com/DigitalHealthcareLab/25MultiModalFoundationModel.git

Figures

Figures reproduced from arXiv: 2501.13277 by the authors.

**Figure 1.** Overview of MEDFORM (S+C) pre-training We adapted TANGLE [22], a transcriptomics-guided representation learning framework, for joint pretraining of CT images and clinical data through contrastive learning. The proposed framework consists of three main components: (1) a vision encoder that encodes slice-level embeddings from sequential CT slices and is pooled by a module to aggregate them into a unified CT represent… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 29 canonical work pages

[1]

A simple framework for contrastive learning of visual representations

Chen, T., et al. A simple framework for contrastive learning of visual representations. in International conference on machine learning. 2020. PMLR

work page 2020
[2]

Nature biomedical engineering, 2021

Lu, M.Y ., et al., Data-efficient and weakly supervised computational pathology on whole- slide images. Nature biomedical engineering, 2021. 5(6): p. 555-570

work page 2021
[3]

Arik, S.O. and T. Pfister, Tabnet: Attentive interpretable tabular learning. arXiv 2019. arXiv preprint arXiv:1908.07442, 1908

arXiv 2019
[4]

Nature Medicine, 2022

Acosta, J.N., et al., Multimodal biomedical AI. Nature Medicine, 2022. 28(9): p. 1773-1784

work page 2022
[5]

Nature Medicine, 2024

V olinsky-Fremond, S., et al., Prediction of recurrence risk in endometrial cancer with multimodal deep learning. Nature Medicine, 2024. 30(7): p. 1962-1973

work page 2024
[6]

Moor, M., et al., Foundation models for generalist medical artificial intelligence. Nature,

work page
[7]

medRxiv, 2024: p

Huang, S.-C., et al., Multimodal Foundation Models for Medical Imaging-A Systematic Review and Implementation Guidelines. medRxiv, 2024: p. 2024.10. 23.24316003

work page 2024
[8]

npj Precision Oncology, 2024

Truhn, D., et al., Large language models and multimodal foundation models for precision oncology. npj Precision Oncology, 2024. 8(1): p. 72

work page 2024

Show all 30 references

[9]

Yuan, and K

Qiu, J., W. Yuan, and K. Lam, The application of multimodal large language models in medicine. The Lancet Regional Health–Western Pacific, 2024. 45

2024
[10]

Lathrop, and T

Dietterich, T.G., R.H. Lathrop, and T. Lozano-Pérez, Solving the multiple instance problem with axis-parallel rectangles. Artificial Intelligence, 1997. 89(1): p. 31-71

1997
[11]

Tomczak, and M

Ilse, M., J. Tomczak, and M. Welling. Attention-based deep multiple instance learning. in International conference on machine learning. 2018. PMLR

2018
[12]

Efficient multiple instance metric learning using weakly supervised data

Law, M.T., et al. Efficient multiple instance metric learning using weakly supervised data. in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017

2017
[13]

Li, and K.W

Li, B., Y . Li, and K.W. Eliceiri. Dual-stream multiple instance learning network for whole slide image classification with self-supervised contrastive learning. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021

2021
[14]

Smooth attention for deep multiple instance learning: Application to ct intracranial hemorrhage detection

Wu, Y ., et al. Smooth attention for deep multiple instance learning: Application to ct intracranial hemorrhage detection. in International Conference on Medical Image Computing and Computer-Assisted Intervention. 2023. Springer

2023
[15]

Medical & Biological Engineering & Computing, 2022

Frade, J., et al., Multiple instance learning for lung pathophysiological findings detection using CT scans. Medical & Biological Engineering & Computing, 2022. 60(6): p. 1569-1584

2022
[16]

IEEE transactions on medical imaging, 2020

Han, Z., et al., Accurate screening of COVID-19 using attention-based deep 3D multiple instance learning. IEEE transactions on medical imaging, 2020. 39(8): p. 2584-2594

2020
[17]

Momentum contrast for unsupervised visual representation learning

He, K., et al. Momentum contrast for unsupervised visual representation learning. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020

2020
[18]

Advances in neural information processing systems, 2019

Li, C., et al., Cross-modal learning with adversarial samples. Advances in neural information processing systems, 2019. 32

2019
[19]

Hassan, and G.K

Zhou, K., F.H. Hassan, and G.K. Hoon, The State of the Art for Cross-Modal Retrieval: A Survey. IEEE Access, 2023

2023
[20]

Crossclr: Cross-modal contrastive learning for multi-modal video representations

Zolfaghari, M., et al. Crossclr: Cross-modal contrastive learning for multi-modal video representations. in Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021

2021
[21]

Contig: Self-supervised multimodal contrastive learning for medical imaging with genetics

Taleb, A., et al. Contig: Self-supervised multimodal contrastive learning for medical imaging with genetics. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022

2022
[22]

Transcriptomics-guided slide representation learning in computational pathology

Jaume, G., et al. Transcriptomics-guided slide representation learning in computational pathology. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024

2024
[23]

Learning transferable visual models from natural language supervision

Radford, A., et al. Learning transferable visual models from natural language supervision. in International conference on machine learning. 2021. PMLR

2021
[24]

Menten, and D

Hager, P., M.J. Menten, and D. Rueckert. Best of both worlds: Multimodal contrastive learning with tabular and imaging data. in Proceedings of the IEEE/CVF Conference on 8 Computer Vision and Pattern Recognition. 2023

2023
[25]

Contrastive learning of medical visual representations from paired images and text

Zhang, Y ., et al. Contrastive learning of medical visual representations from paired images and text. in Machine Learning for Healthcare Conference. 2022. PMLR

2022
[26]

Nature Biomedical Engineering, 2022: p

Lee, Y ., et al., Derivation of prognostic contextual histopathological features from whole-slide images of tumours via graph deep learning. Nature Biomedical Engineering, 2022: p. 1-15

2022
[27]

IEEE transactions on medical imaging, 2020

Wang, X., et al., A weakly-supervised framework for COVID-19 classification and lesion localization from chest CT. IEEE transactions on medical imaging, 2020. 39(8): p. 2615-2625

2020
[28]

Visual language pretrained multiple instance zero-shot transfer for histopathology images

Lu, M.Y ., et al. Visual language pretrained multiple instance zero-shot transfer for histopathology images. in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023

2023
[29]

Behrad, F. and M. Saniee Abadeh, An overview of deep learning methods for multimodal medical data mining. Expert Systems with Applications, 2022. 200: p. 117006

2022
[2023]

616(7956): p. 259-265

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

A simple framework for contrastive learning of visual representations

Chen, T., et al. A simple framework for contrastive learning of visual representations. in International conference on machine learning. 2020. PMLR

work page 2020

[2] [2]

Nature biomedical engineering, 2021

Lu, M.Y ., et al., Data-efficient and weakly supervised computational pathology on whole- slide images. Nature biomedical engineering, 2021. 5(6): p. 555-570

work page 2021

[3] [3]

Arik, S.O. and T. Pfister, Tabnet: Attentive interpretable tabular learning. arXiv 2019. arXiv preprint arXiv:1908.07442, 1908

arXiv 2019

[4] [4]

Nature Medicine, 2022

Acosta, J.N., et al., Multimodal biomedical AI. Nature Medicine, 2022. 28(9): p. 1773-1784

work page 2022

[5] [5]

Nature Medicine, 2024

V olinsky-Fremond, S., et al., Prediction of recurrence risk in endometrial cancer with multimodal deep learning. Nature Medicine, 2024. 30(7): p. 1962-1973

work page 2024

[6] [6]

Moor, M., et al., Foundation models for generalist medical artificial intelligence. Nature,

work page

[7] [7]

medRxiv, 2024: p

Huang, S.-C., et al., Multimodal Foundation Models for Medical Imaging-A Systematic Review and Implementation Guidelines. medRxiv, 2024: p. 2024.10. 23.24316003

work page 2024

[8] [8]

npj Precision Oncology, 2024

Truhn, D., et al., Large language models and multimodal foundation models for precision oncology. npj Precision Oncology, 2024. 8(1): p. 72

work page 2024