Multimodal deep learning for biomedical data fusion: a review
Open accessThe DOI matches its Crossref record; 107 of the 117 references it cites were found in a registry or an index; 60 of the 96 sources its claims cite were checked for support.
Abstract
Biomedical data are becoming increasingly multimodal and thereby capture the underlying complex relationships among biological processes. Deep learning (DL)-based data fusion strategies are a popular approach for modeling these nonlinear relationships. Therefore, we review the current state-of-the-art of such methods and propose a detailed taxonomy that facilitates more informed choices of fusion strategies for biomedical applications, as well as research on novel methods. By doing so, we find that deep fusion strategies often outperform unimodal and shallow approaches. Additionally, the proposed subcategories of fusion strategies show different advantages and drawbacks. The review of current methods has shown that, especially for intermediate fusion strategies, joint representation learning is the preferred approach as it effectively models the complex interactions of different levels of biological organization. Finally, we note that gradual fusion, based on prior biological knowledge or on search strategies, is a promising future research path. Similarly, utilizing transfer learning might overcome sample size limitations of multimodal data sets. As these data sets become increasingly available, multimodal DL approaches present the opportunity to train holistic models that can learn the complex regulatory dynamics behind health and disease.
Overview
CoChat Summary
TL;DR
This review surveys deep learning-based multimodal biomedical data fusion and proposes a taxonomy of early, intermediate, and late fusion strategies. It finds that deep fusion often outperforms unimodal and shallow approaches, with intermediate joint representation learning preferred, while gradual fusion and transfer learning remain underexplored.
Key findings
- The paper proposes a taxonomy dividing fusion into early fusion (direct modeling, autoencoder subcategories), intermediate fusion (homogeneous/heterogeneous designs; marginal/joint representations), and late fusion (averaging equal/weighted, meta-learning).
- Deep fusion strategies frequently outperform unimodal and shallow methods; joint intermediate fusion is the preferred approach because it effectively models complex interactions across levels of biological organization.
- Early fusion is simple but may miss cross-modal relationships apparent only at higher abstraction; intermediate fusion offers flexibility for heterogeneity, missing modalities, and dimensionality imbalance; late fusion combines decisions but cannot learn feature-level cross-modal interactions.
- Gradual fusion guided by prior biological knowledge or search strategies and transfer learning are promising future paths; reviewed applications included sample sizes as low as 96 patients.
Methods
- Narrative literature review of deep learning-based data fusion strategies in the biomedical domain, taking a representation learning perspective.
- Proposes a detailed taxonomy of early, intermediate, and late fusion and their subcategories, exemplified with published biomedical applications.
- No primary dataset, systematic search protocol, or quantitative meta-analysis is described; synthesis is conceptual and qualitative.
Limitations
- The paper states the literature is most likely skewed toward reporting positive results, performance of fusion strategies is problem- and data-specific, and more theoretical knowledge is needed to specify when strategies excel.
- The paper states multimodal DL faces general challenges including data volume, quality, interpretability, temporality, missing entire modalities, and overfitting/generalizability due to small sample sizes and many parameters.
- Evident from the text: as a narrative review, it does not report a systematic search strategy or quantitative comparison across studies.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures
![Figure 1 Development of technologies and multimodal deep learning (DL). ‘Omics’ and ‘multi-omics’ data become increasingly relevant in the scientific literature. To fully utilize the growing number of multimodal data sets, data fusion methods based on DL are evolving into an important approach in the biomedical field. This unprecedented generation of data has been made possible by high-throughtput technologies like microarrays and next-generation sequencing [7]. The development of bulk RNA-seq was followed by several related sequencing technologies, such as single-cell RNA-seq and ATAC-seq [8]. Currently, spatial transcriptomics [9] and single-cell multi-omics [10] are being increasingly used.](https://figures.cochat.ai/paper-figures/5c1e0ae86d784af8937add9f4a2f8906/multimodal-deep-learning-for-biomedical-data-fusion-a-review-figure-1.jpg)
![Figure 2 DL-based fusion strategies. Layers marked in blue are shared between modalities and learn joint representations. (a) Early fusion strategies take as input a concatenated vector. No marginal representations are learned. (b) Intermediate fusion strategies first learn marginal representations and fuse these later inside the network. This can occur in one layer or gradually. (c) Late fusion strategies combine decisions by sub-models for each modality. Figure adapted from [2].](https://figures.cochat.ai/paper-figures/73c64526f30e41a695c45c40cae87d27/multimodal-deep-learning-for-biomedical-data-fusion-a-review-figure-2.jpg)


Verification
Claims the paper makes
40 Valid · 5 Not valid · 23 Not checked · 87 Own results and interpretations
Some cited sources support this claim and some don’t — read each source’s verdict below.
In the paper
“Despite the aforementioned drawbacks, some reviewed papers have shown that early fusion AEs can perform on par with intermediate strategies [38, 41]”
§ Multimodal deep learning for biomedical data fusion: a review > Early fusion > Discussion of early fusion strategies
What the sources say
- Disputedcites [38] Islam, 2020, An integrative deep learning framework for classifying molecular subtypes of breast cancer, Comput Struct Biotechnol J sourceExperimental — accuracy not yet measured
The Methods section describes an early-fusion stacked autoencoder that takes concatenated CNA and gene-expression data as input, alongside a DCNN_Concat model that combines features learned separately from each source. The Results section (page unavailable) says DCNN_Concat performed better than the autoencoder-based DNN_SE, rather than reporting that they performed on par. (LLM, conf=0.86)
Checked by a reasoning model — may be wrong.
- Supportedcites [41] Simidjievski, 2019, Variational autoencoders for cancer data integration: design principles and computational practice, Front Genet source
The Methods section describes CNC-VAE as combining data at the input and the other architectures as integrating data in hidden layers. In Results > Quality of the Learned Representations (page unavailable), H-VAE and X-VAE mostly produced more accurate predictions, but the differences among the four architectures were not significant. This supports the claim that an early-fusion AE can perform on par with intermediate strategies in the tasks studied. (LLM, conf=0.89)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
Every source the paper cites for this claim was checked and supports it.
In the paper
“Questioning this assumption is relevant as many methods have shown unequal contribution of different modalities to the predictive performance (see for example [36, 42]).”
§ Multimodal deep learning for biomedical data fusion: a review > Late fusion
What the sources say
- Supportedcites [36] Chai, 2021, Integrating multi-omics data through deep learning for accurate cancer prognosis prediction, Comput Biol Med source
The Conclusion and Discussion reports an ablation study in which mRNA performed best, followed by miRNA and methylation, while copy-number variation contributed least. Table 3 also shows different C-index values when individual omics types are used or excluded, supporting unequal contributions to predictive performance. (LLM, conf=0.92)
- Supportedcites [42] Albaradei, 2021, Meta cancer: a deep learning-based pan-cancer metastasis prediction model developed using multi-omics data, Comput Struct Biotechnol J source
In “Evaluating the contribution of different omics data types” (page not supplied), MetaCancer reports different predictive contributions from mRNA, DNA methylation and microRNA. The section also describes similar differences in other studies, supporting its use as an example of the claim. (LLM, conf=0.91)
boostedstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“However, different modalities can add information to different degrees (e.g. [36, 44]).”
§ Multimodal deep learning for biomedical data fusion: a review > Early fusion > Discussion of early fusion strategies
What the sources say
- Supportedcites [36] Chai, 2021, Integrating multi-omics data through deep learning for accurate cancer prognosis prediction, Comput Biol Med source
The ‘Patients’ risks estimation by multi-omics data’ section compares the contributions of individual omics types: mRNA contributed most to prognosis prediction, while copy number variation contributed least. This supports the claim that different modalities add information to different degrees. (LLM, conf=0.93)
- Supportedcites [44] Lee, 2019, Alzheimer’s disease neuroimaging initiative. Predicting Alzheimer’s disease progression using multi-modal deep learning approach, Sci Rep source
The Results section, in its comparison of single-modality and multimodal predictions, reports that cognitive-performance data predicted conversion more accurately than the other individual modalities, while demographic data performed too poorly to include in the figure. Table 5 also shows different accuracies across modalities and higher accuracy for the combined model. The supplied passages have no page numbers. (LLM, conf=0.91)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Vale-Silva et al. [74] compared these methods, although they did not find large differences in performance for predicting long-term cancer survival.”
§ Multimodal deep learning for biomedical data fusion: a review > Intermediate fusion > Heterogeneous network design
What the sources say
- Partly supportedcites [74] Vale-Silva, 2021, Long-term cancer survival prediction using multimodal deep learning, Sci Rep sourceExperimental — accuracy not yet measured
The Methods section (“Multimodal data fusion layer”; page unavailable) says MultiSurv used a row-wise maximum to combine modality representations and tested alternative fusion schemes without improving performance. It does not identify those alternatives in our copy or establish that performance differences among them were small. (LLM, conf=0.82)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Particularly in cancer patient survival subtyping with multi-omics modalities, this sequence of steps has become popular [25–31, 34].”
§ Multimodal deep learning for biomedical data fusion: a review > Early fusion > Latent representation with multimodal AEs
What the sources say
- Supportedcites [25] Chaudhary, 2018, Deep learning-based multi-omics integration robustly predicts survival in liver cancer, Clin Cancer Res source
In Results, the paper describes integrating three omics data types with an autoencoder, selecting bottleneck features using univariate Cox-PH regression, and inferring survival-subtype labels through K-means clustering. It therefore exemplifies the sequence for which it is cited; the citation need not establish popularity on its own. (LLM, conf=0.92)
- Partly supportedcites [26] Lv, 2020, Survival prediction in patients with colon adenocarcinoma via multiomics data integration using a deep learning algorithm, Biosci Rep sourceExperimental — accuracy not yet measured
The Results section ('Identification of two survival risk groups in TCGA multiomics data'; page unavailable) describes the stated sequence: an autoencoder transformed multiomics features, univariable Cox-PH regression selected survival-associated features, and K-means clustering inferred two groups. This establishes that the paper uses the sequence, but not that it has become popular; the Introduction instead describes deep-learning applications in prognostic modeling as being in their infancy. (LLM, conf=0.85)
- Partly supportedcites [27] Zhang, 2018, Deep learning-based multi-omics data integration reveals two prognostic subtypes in high-risk neuroblastoma, Front Genet sourceExperimental — accuracy not yet measured
The Materials and methods sections (pages unavailable) describe multi-omics features generated by an autoencoder, subsequent Cox regression to select prognostic features, and K-means clustering to infer survival-risk labels. The paper does not establish that this full sequence has become popular, and the passages do not specify that the post-autoencoder Cox modeling was univariate. (LLM, conf=0.83)
- Partly supportedcites [28] Yu, 2020, A model for predicting prognosis in patients with esophageal squamous cell carcinoma based on joint representation learning, Oncol Lett sourceExperimental — accuracy not yet measured
The Materials and methods section (page unavailable) describes joint multimodal autoencoder features, univariate Cox-PH feature selection, and K-means clustering to infer survival-risk subgroup labels. The paper therefore exemplifies the sequence, but it does not establish that the full sequence had become popular in multi-omics cancer survival subtyping. (LLM, conf=0.83)
- Supportedcites [29] Asada, 2020, Uncovering prognosis-related genes and pathways by multi-omics analysis in lung cancer, Biomol Ther source
The Methods section (2.2) describes an autoencoder followed by univariate Cox-PH feature selection and K-means clustering to infer labels. The Discussion (section 4) identifies this as the paper’s multi-omics pipeline for lung-cancer survival subtyping, so the paper supports its use as an example of the cited sequence. (LLM, conf=0.90)
- Supportedcites [30] Zhang, 2021, Robust prognostic subtyping of muscle-invasive bladder cancer revealed by deep learning-based multi-omics data integration, Front Oncol source
The Results section, “The Identification of OS-Related Subtypes Based on TCGA Multi-Omics Data,” describes the claimed sequence: autoencoder-derived multi-omics features, univariate Cox selection of survival-related features, and K-means clustering to assign subtypes. The paper therefore serves as an example of the approach cited alongside other studies; it does not independently establish how popular the approach is. (LLM, conf=0.89)
- Not checked — no full text of the sourcecites [31] Lee, 2020, Incorporating deep learning and multi-omics autoencoding for analysis of lung adenocarcinoma prognostication, Comput Biol Chem source
- Supportedcites [34] Zhao, 2020, Deep learning-based model for predicting progression in patients with head and neck squamous cell carcinoma, Cancer Biomark source
Section 3.1 describes combining three omics modalities in an autoencoder, selecting survival-associated features by univariate Cox-PH regression, and using K-means clustering to identify patient subtypes. Section 2.3 explicitly lays out the autoencoder → Cox-PH selection → clustering sequence. The paper therefore exemplifies the approach for which it is cited. (LLM, conf=0.91)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
91 more claims · 26 with their cited sources checked
Create a free account to see all 155 claims and to verify the 4 references still open.
- It may state two separate claims as one. 19
- Probably not a substantial claim. 43
- The quote only partly states this claim. 2
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source doesn’t support the claim as written.
In the citing paper
“This method has consistently demonstrated superior performance compared to early and late fusion strategies across diverse clinical applications14–16.”
From MoMA: a mixture-of-multimodal-agents architecture for enhancing clinical prediction modelling (2025) · cites this paper as [15]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“However, each of these public datasets, as well as the private datasets used in included studies, contains no more than 280 subjects (with sample sizes ranging from 24 to 275), which hampers the ability to identify clinical biomarkers universally and reliably, especially with DL approaches132,133.”
From AI-assisted multi-modal information for the screening of depression: a systematic review and meta-analysis (2025) · cites this paper as [132]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“DL architectures are well-suited for intermediate fusion as they enable an easy connection of marginal representations to a shared layer, facilitating the correspondence of hierarchical representations to real-world phenomena (Stahlschmidt et al. 2022).”
From A review of multimodal deep learning methods for genomic-enabled prediction in plant breeding (2024) · cites this paper as Stahlschmidt et al. 2022
The cited source says what the claim attributes to it.
In the citing paper
“This approach assumes an equal contribution from each sub-model, as there is no weighting of their outputs (Stahlschmidt et al. 2022).”
From A review of multimodal deep learning methods for genomic-enabled prediction in plant breeding (2024) · cites this paper as Stahlschmidt et al. 2022
The cited source says what the claim attributes to it.
In the citing paper
“Stahlschmidt et al. summarized multimodal fusion methods in biomedical analysis [31]”
From Challenges in AI-driven Biomedical Multimodal Data Fusion and Analysis (2025) · cites this paper as [31]
1 more claim about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
LeCun, Y., Bengio, Y., & Hinton, G. (2015) · Nature · doi:10.1038/nature14539 · Cited by 84,126
Regression Models and Life-Tables
Cox, D. R. (1972) · Journal of the Royal Statistical Society Series B: Statistical Methodology · doi:10.1111/j.2517-6161.1972.tb00899.x · Cited by 39,793
A fast learning algorithm for deep belief nets
Hinton, G. E., Osindero, S., & Teh, Y.-W. (2006) · Neural Computation · doi:10.1162/neco.2006.18.7.1527 · Cited by 16,336 · cited 1×
“Alternatively, the representations can serve as a useful and computationally efficient initialization of a DNN, which is fine-tuned with more expensive supervised algorithms such as back propagation to learn p(y|x) [94].”
MethodRepresentation Learning: A Review and New Perspectives
Bengio, Y., Courville, A., & Vincent, P. (2013) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2013.50 · Cited by 13,001 · cited 3×
“Earlier layers learn simple abstractions of the data, whereas deeper layers combine these into more abstract representations that are informative for the learning task [13].”
BackgroundInfluentialA Comprehensive Survey on Graph Neural Networks
Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C., & Yu, P. S. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2020.2978386 · Cited by 11,000
Extracting and composing robust features with denoising autoencoders
Vincent, P., Larochelle, H., Bengio, Y., & Manzagol, P.-A. (2008) · Proceedings of the 25th international conference on Machine learning - ICML '08 · doi:10.1145/1390156.1390294 · Cited by 7,436 · cited 1×
“For the same clinical task, a similar workflow has been adapted by researchers, but applying denoising AEs (DAEs) [91] instead [35, 36].”
MethodGrad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2020) · International Journal of Computer Vision · doi:10.1007/s11263-019-01228-7 · Cited by 6,909 · cited 2×
“[79] enabled modality-specific interpretability by applying Grad-CAM [97] for WSI and integrated gradient [98] for cell graph and genomic modalities using convolutional, graph convolutional [14] and fully connected branches.”
MethodLarge-Scale Video Classification with Convolutional Neural Networks
Karpathy, A., Toderici, G., Shetty, S., Leung, T., Sukthankar, R., & Fei-Fei, L. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.223 · Cited by 6,354 · cited 1×
“Despite the aforementioned drawbacks, some reviewed papers have shown that early fusion AEs can perform on par with intermediate strategies [38, 41], though other structured investigations inside [28] and outside [93] the biomedical domain have shown superiority of intermediate fusion over early fusion strategies.”
BackgroundMultimodal Machine Learning: A Survey and Taxonomy
Baltrusaitis, T., Ahuja, C., & Morency, L.-P. (2019) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2018.2798607 · Cited by 5,018 · cited 5×
“Thereby, coordinated representations [11] of each modality could be learned in an unsupervised way, which resulted in encodings of patterns between modalities.”
BackgroundInfluentialVisualization and analysis of gene expression in tissue sections by spatial transcriptomics
Ståhl, P. L., Salmén, F., Vickovic, S., Lundmark, A., Navarro, J. F., Magnusson, J., Giacomello, S., Asp, M., Westholm, J. O., Huss, M., et al. (2016) · Science · doi:10.1126/science.aaf2403 · Cited by 3,834 · cited 1×
“Currently, spatial transcriptomics [9] and single-cell multi-omics [10] are being increasingly used.”
Method
Research works printed but not linked: 6
- Springer Nature. Milestones in Genomic Sequencing
- Goodfellow, 2016, Deep Learning
- Ballard, 1987, Modular learning in neural networks. In: Proceedings of the Sixth National Conference on Artificial Intelligence, 279
- Kingma, 2014, Proceedings of the International Conference on Learning Representations (ICLR)
- Alemi, 2017, 5th International Conference on Learning Representations
- Sundararajan, 2017, 34th International Conference on Machine Learning, 5109
Cited by
Items 701–749 of 749
A Multi-Modal Framework for Fraud Detection and Weakly Supervised Risk Stratification
Peng, Y., & Noman, S. M. (2026) · Journal of Organizational and End User Computing · doi:10.4018/joeuc.419519
Opportunities and future directions of neurological disorders diagnosis research
Saini, A., Dadhich, A., Maheshwari, S., Sharma, V. K., & Sharma, R. (2026) · Elsevier eBooks · doi:10.1016/b978-0-44-330291-6.00020-2
Zhu, F., Huo, Y., Rong, Y., Jiang, Y., Dang, R., Hu, B., & Huang, Y. (2026) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2026.111340
Tang, J. (2026) · International Conference on Bioinformatics and Intelligent Computing (BIC) · doi:10.1145/3809986.3810060
“Recent studies indicate that integrating multi-modal data can effectively capture the complex heterogeneity of metabolic diseases and enhance predictive performance [ 29 , 30 ] .”
Fujiwara, M., Yoshida, S., Banerjee, I., Arita, Y., & Fujii, Y. (2026) · Abdominal Radiology · doi:10.1007/s00261-026-05723-7
Suresh, S., Parvathy, A. S., Raj, M., S., K. G., Priya, M. L., Biswas, R., Ramaiah, S., & Anbarasu, A. (2026) · Frontiers in Digital Health · doi:10.3389/fdgth.2026.1872286 · Cited by 1
“The representation can be generated via similar neural networks, yielding a homogeneous network, or via various networks that serve heterogeneous models (59).”
Liang, L., Xu, J., & Zhang, Y. (2026) · Physics in Medicine and Biology · doi:10.1088/1361-6560/ae9e80
Explainable AI for multi-omics in precision medicine: a systematic review
Hasnain, M., Hameed, S., Azim, K., & Waseem, M. (2026) · Exploration of Digital Health Technologies · doi:10.37349/edht.2026.1011102
Huang, X., bao, Q., Sun, Y., Yang, X., & Zhang, X. (2026) · Medicine · doi:10.1097/md.0000000000050391
“[15,16] Multiple studies have hitherto utilized machine learning algorithms to predict risk factors for diseases such as stroke, [17–21] yet few have applied these approaches to derive sex-specific risk nomograms within a long-term nationwide cohort.”
3D-FuseNet: An innovative multimodal fusion network for enhancing survival prediction
Chaddad, A., Feng, J.-J., Daqqaq, T. S., Katib, Y., & Jiang, L. (2026) · Computerized Medical Imaging and Graphics · doi:10.1016/j.compmedimag.2026.102808
Sun, L., Peng, Q., Xiao, X., et al. (2026) · Journal of Imaging Informatics in Medicine · doi:10.1007/s10278-026-02184-8
Agbavor, F., & Liang, H. (2026) · JMIR Aging · doi:10.2196/93279
Rahmani, A. H., & Sarwar, T. (2026) · Diagnostics · doi:10.3390/diagnostics16172799 · Cited by 1
Time and frequency characteristics of various noninvasive heartbeat sensors
Charlier, P., Jeanne, M., Hureau, M., & Jonckheere, J. D. (2026) · PLoS ONE · doi:10.1371/journal.pone.0357922
Principles of multimodal fusion
Rakhmatulin, I., & Naik, G. R. (2026) · Elsevier eBooks · doi:10.1016/b978-0-443-49276-1.00010-4
Yan, F., Ma, K., Li, C., & Liang, X. (2026) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2026.111409
Deep Learning-Based Multi-Source Data Fusion Methods for Intelligent Information Processing
Lin, G. (2026) · International Conference on Artificial Intelligence, Virtual Reality and Interaction Design · doi:10.1145/3820915.3820932
“[3] [4] [5] [6] Beyond centralized training, multi-source fusion is increasingly studied in distributed and privacy-constrained settings, where federated learning aims to learn shared representations without exposing raw data.”
Jiang, X., Wang, Z., Wang, Y., Mu, Y., Yang, X., Tan, L. H., & Wei, T. (2026) · Biotechnology Advances · doi:10.1016/j.biotechadv.2026.109038
AI-assisted multimodal data integration for precision oncology
Chera, A., Chirilaș, A.-M., El-Sabeh, A., Cretu-Stancu, M., & Bucur, O. (2026) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2026.1917442
Explainable AI for Multi-modal Medical Image Analysis and Radio-Genomics: A Comprehensive Review
Singh, A., & Singh, N. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-30438-4_30
Li, Y., Wang, P., Zou, X., Zhang, Q., & Wang, W. (2026) · Magnetic Resonance Imaging · doi:10.1016/j.mri.2026.110796
Tahmin, N., Chinthala, L. K., Baye, T. M., Davis, R. L., & Khojandi, A. (2026) · medRxiv · doi:10.64898/2026.09.03.26362179
Artificial intelligence and target trial emulation: toward scalable and credible real-world evidence
Liu, T., & Wang, Y. (2026) · Journal of Clinical Epidemiology · doi:10.1016/j.jclinepi.2026.112501
Daniel, J., & Sundar Rajan, R. (2026) · Franklin Open · doi:10.1016/j.fraope.2026.100766
Huang, P., Huang, Y., Chen, Y., Yeung, M., Jiang, Y., & Yu, L. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-37432-5_13
Reliability-aware disentangled adaptive network for multimodal sentiment analysis
Xu, J., Zhao, X., Jia, L., Zhong, Z., & Zhong, X. (2026) · International Journal of Data Science and Analytics · doi:10.1007/s41060-026-01280-w · Cited by 1
Nguyen, B.-D., Hoang, V.-D., & Nguyen, H. D. (2026) · Network Modeling Analysis in Health Informatics and Bioinformatics · doi:10.1007/s13721-026-00871-8
“…strategies difficult to generalize and motivate adap - tive mechanisms that can selectively weight, route, or discard modality contributions while accounting for the distinct characteristics, reliability, and clinical relevance of each modality (Stahlschmidt et al. 2022 ; Lipkova et al. 2022 ).”
Unifying multimodal single-cell data with a mixture-of-experts β-variational autoencoder framework
Ashford, A. J., Enright, T., Somers, J., Nikolova, O. H., & Demir, E. (2026) · Genome Research · doi:10.1101/gr.281431.125 · Cited by 1
“Deep generative models extend this landscape by coupling probabilistic encoders/decoders with modality-aware likelihoods (Stahlschmidt et al. 2022): scVI provides a principled framework for scRNA-seq, whereas totalVI and MultiVI generalize to CITE-seq and RNA – ATAC, respectively (Lopez et al.…”
Tian, Y. (2026) · International Conference on Biomedicine and Bioinformatics Engineering (ICBBE) · doi:10.1145/3822183.3822287
“" Combining ESM-2 structural embeddings Δ 𝐸 with genomic context (e.g., replication timing and tissue specificity) proves both computationally effective and biologically sound [3, 13].”
Sun, X., Li, D., Song, X., Li, J., Hu, X., Sun, J., & Wang, L. (2026) · Talanta · doi:10.1016/j.talanta.2026.130625
Lv, Z., Yang, J., Chen, L., Ji, R., Li, Y., Wu, M., Huang, S., Yuan, J., Jiang, Z., & Li, T. (2026) · Computers and Electronics in Agriculture · doi:10.1016/j.compag.2026.112452
Gweon, H., & Li, S. (2026) · Astin Bulletin · doi:10.1017/asb.2026.10116
Multimodal AI for cardiovascular care must learn when not to predict
Han, Y., Li, Q., Zhao, T., Cai, S., Li, B., Kadier, K., Gu, T. F., Sun, Z., Xu, P., Liu, D., et al. (2026) · Medicine Discovery · doi:10.1016/j.medd.2026.100004
Mehri-Kakavand, G., Mdletshe, S., Pursamimi, M., & Wang, A. (2026) · Journal of Imaging Informatics in Medicine · doi:10.1007/s10278-026-02282-7
Skaramagkas, V., Boura, I., Karamanis, G., Pentari, A., Michou, E., Kostas Marias, A., Fotiadis, D. I., Kefalopoulou, Z. M., Spanaki, C., & Tsiknakis, M. (2026) · Frontiers in Digital Health · doi:10.3389/fdgth.2026.1937343
“…(78.62% f1-score compared to unimodal baselines); however, performance deteriorated upon the indiscriminate inclusion of speech or upper limb signals, reflecting established constraints of naïve LF, which overlooks signal-level interactions and assumes uniform reliability across modalities (43).”
InfluentialHerrera, D., Almhdie-Imjabbar, A., Ibrahim-Nasser, N., Toumi, H., & Lespessailles, É. (2026) · Scientific Reports · doi:10.1038/s41598-026-68440-7 · Cited by 1
“Furthermore, the intermediate fusion strategy, may have reduced the contribution of MRI features, as balancing feature representations with significantly different dimensionalities can limit the amount of discriminative information retained from the higher-dimensional modality 39 .”
Combinatorial AI and nanotechnological approaches for advancing medical sensors
Kapur, U., Adhikari, D., Tripathi, H., & Gabrani, R. (2026) · Elsevier eBooks · doi:10.1016/b978-0-443-45321-2.00003-5
Martinez, J., & Lopez, C. (2026) · Journal of Health Informatics and Digital Systems · doi:10.68159/k105027051
Hojeij, B., Tchetverikov, I., Vasileiou, E., Apostolidis, G., Dimitroulas, T., Mytilinaiou, M., Gonçalves, C., Rodrigues, A. M., Coates, L. C., Konstantinidis, D., et al. (2026) · BMJ Open · doi:10.1136/bmjopen-2025-115903
Strategies for recycled plastics content verification
Zabihi, O., Brüll, R., Pfaendner, R., Malz, F., Dagley, I., & Naebe, M. (2026) · Advanced Industrial and Engineering Polymer Research · doi:10.1016/j.aiepr.2026.100284
A multimodal wearable sensor patch with data-driven analysis for postoperative wound monitoring
Li, M., Liu, N., Zheng, X. T., Liang, J., Zhang, T., Lin, R., Wen, T., Xu, X., Su, X., & Li, P. (2026) · Sensors and Actuators B Chemical · doi:10.1016/j.snb.2026.140987
Xiao, H., Wang, J., & Wan, S. (2026) · Journal of Alzheimer s Disease · doi:10.1177/13872877261491975 · Cited by 5
Wang, Z., Li, J., Zhang, C., & Zhang, J. (2026) · International Conference on Advanced Algorithms and Neural Networks (AANN) · doi:10.1109/aann71145.2026.11709238
Sun, H., Li, B., Jiang, X., et al. (2026) · INTERNATIONAL CONFERENCE ON MULTIMODAL INTERACTION · doi:10.1145/3776574.3831150
Korutla, R., Garg, T., Pol, S. S., Shivandi, V., Gnanam, S., Padari, K. S., & Amal, S. (2026) · Artificial Intelligence in the Life Sciences · doi:10.1016/j.ailsci.2026.100183 · Cited by 1
Al‐Quraishi, T., Zaboon, W. A., Abawajy, J., & Nowrozy, R. (2026) · Discover Artificial Intelligence · doi:10.1007/s44163-026-02333-6
“This multimodal integration is likely to produce models that are more biologically robust and better suited to practical applications in structural biology, functional annotation, and drug discovery [93, 114].”
Li, Q. (2026) · Lecture notes on data engineering and communications technologies · doi:10.1007/978-981-95-7622-7_12
When One View Isn’t Enough: Multimodal Cardiac Representation Fusion for Sudden Cardiac Arrest RiskPreprint
Prity, J. (2026) · Research Square · doi:10.21203/rs.3.rs-11261393/v1
Fu, L., Chen, J., Zhang, Y., Wang, Z., & Huang, X. (2026) · Analytica Chimica Acta · doi:10.1016/j.aca.2026.346347
Metadata
- Authors
- Stahlschmidt, S. R., Ulfenborg, B., & Synnergren, J.
- Affiliations
- University of Skövde (SE)
- Corresponding author
- Stahlschmidt, Sören Richard
- Venue
- Briefings in Bioinformatics
- Publisher
- Oxford University Press (OUP)
- Published
- 2022-01-28
- Volume
- 23
- Issue
- 2
- ISSN
- 1467-5463 (ISSN-L)
- 1477-4054
- PMC
- PMC8921642
- OpenAlex
- W4206042814
- Cited by
- 766
- References
- 117
- Citations by year
- 2026: 241 · 2025: 245 · 2024: 175 · 2023: 82 · 2022: 23
- FWCI
- 7.34
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- Briefings in Bioinformatics (publishedVersion, cc-by-nc)
- PubMed Central (publishedVersion, cc-by-nc)
- PubMed (publishedVersion)
- KTH Publication Database DiVA (KTH Royal Institute of Technology) (submittedVersion, other-oa)
- Funders
- University of Skövde, Sweden (20170302, 20200014)
- Topics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
- Keywords
- multimodal fusion, biomedical data, deep learning, joint representation learning, intermediate fusion, transfer learning, biological knowledge integration
- MeSH terms
- Deep Learning
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.