Using Meta-Transformers for Multimodal Clinical Decision Support and Evidence-Based Medicine
Bibliographic record
Abstract
Abstract The advancements in computer vision and natural language processing are keys to thriving modern healthcare systems and its applications. Nonetheless, they have been researched and used as separate technical entities without integrating their predictive knowledge discovery when they are combined. Such integration will benefit every clinical/medical problem as they are inherently multimodal - they involve several distinct forms of data, such as images and text. However, the recent advancements in machine learning have brought these fields closer using the notion of meta-transformers. At the core of this synergy is building models that can process and relate information from multiple modalities where the raw input data from various modalities are mapped into a shared token space, allowing an encoder to extract high-level semantic features of the input data. Nerveless, the task of automatically identifying arguments in a clinical/medical text and finding their multimodal relationships remains challenging as it does not rely only on relevancy measures (e.g. how close that text to other modalities like an image) but also on the evidence supporting that relevancy. Relevancy based on evidence is a normal practice in medicine as every practice is an evidence-based. In this article we are experimenting with meta-transformers that can benefit evidence based predictions. In this article, we are experimenting with variety of fine tuned medical meta-transformers like PubmedCLIP, CLIPMD, BiomedCLIP-PubMedBERT and BioCLIP to see which one provide evidence-based relevant multimodal information. Our experimentation uses the TTi-Eval open-source platform to accommodate multimodal data embeddings. This platform simplifies the integration and evaluation of different meta-transformers models but also to variety of datasets for testing and fine tuning. Additionally, we are conducting experiments to test how relevant any multimodal prediction to the published medical literature especially those that are published by PubMed. Our experimentations revealed that the BiomedCLIP-PubMedBERT model provide more reliable evidence-based relevance compared to other models based on randomized samples from the ROCO V2 dataset or other multimodal datasets like MedCat. In this next stage of this research we are extending the use of the winning evidence-based multimodal learning model by adding components that enable medical practitioner to use this model to predict answers to clinical questions based on sound medical questioning protocol like PICO and based on standardized medical terminologies like UMLS.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.015 | 0.059 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.010 | 0.004 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.005 | 0.007 |
| Open science | 0.002 | 0.004 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.011 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".