Perramon-Llussà, Joan, Jiménez-Sánchez, Amelia, Skorupko, Grzegorz, Avgoustidis, Fotis, Martín-Isla, Carlos, Lekadir, Karim, Gkontra, Polyxeni · Mar 11, 2026
arXiv cs.AI: Med-DualLoRA: Local Adaptation of Foundation Models for 3D Cardiac MRI
Markdown · JSON · Original source
The authors of Med-DualLoRA propose a way to adapt a for disease detection in 3D cardiac MRI across multiple sites, while keeping each site’s local adaptations private. Their method separates adaptations into a global component, shared and aggregated across sites, and local components that stay at each site. The authors argue this design can personalize the model while reducing communication cost, addressing two challenges they identify: clinical data may not be centralized because of privacy constraints, and data from different sites can vary. In experiments using ACDC and combined M&Ms datasets, with each vendor treated as a federated client, they report improved performance over other federated PEFT baselines, including balanced accuracy of 0.768 and specificity of 0.612. They also report that adapting only two transformer blocks preserves performance and further improves efficiency. These findings describe the paper’s evaluation; the abstract does not establish how the method performs in routine clinical use or beyond the evaluated datasets and setup.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Mar 11, 2026 (2026-03-11).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-10T21:16:33.095Z. First observed: 2026-11-10T08:39:34.005Z. Retrieved: 2026-11-10T21:06:39.738Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The abstract reports experiments on ACDC and combined M&Ms datasets, with each vendor treated as a federated client; it does not establish performance beyond that evaluated setup. The abstract describes disease-detection experiments, not performance in routine clinical use.
Open question
How does Med-DualLoRA perform on cardiac MRI data beyond the ACDC and combined M&Ms datasets and vendor-based setup evaluated?
Source-stated
The authors propose Med-DualLoRA, which separates shared and local adaptations; only the shared component is exchanged and aggregated across sites.
only the global component is shared and aggregated across sites, keeping local adapters private.
Source-stated
The authors say this design improves personalization and reduces communication cost.
This design improves personalization while significantly reducing communication cost
Source-stated
In the reported evaluation, Med-DualLoRA achieved balanced accuracy of 0.768 and specificity of 0.612, outperforming other federated PEFT baselines.
Med-DualLoRA achieves statistically significant improved performance (balanced accuracy 0.768, specificity 0.612) compared to other federated PEFT baselines
Source-stated
The authors report that adapting only two transformer blocks preserved performance and further improved efficiency.
adapting only two transformer blocks preserves performance while further improving efficiency.
Source-stated
The evaluation used ACDC and combined M&Ms datasets and treated each vendor as a federated client.
using ACDC and combined M\&Ms datasets, treating each vendor as a federated client.
Source and contributor context
Perramon-Llussà, Joan, Jiménez-Sánchez, Amelia, Skorupko, Grzegorz, Avgoustidis, Fotis, Martín-Isla, Carlos, Lekadir, Karim, Gkontra, Polyxeni · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Foundation model
A large model trained on broad data that can be adapted to many different tasks.
Statistical significance
A test of whether a result is unlikely to be due to chance alone; it does not say how large or important an effect is.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 0b8b184f-ab54-4af0-8e8f-ee1d96b55283.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-10T21:16:33.095Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 860617ba2522343859cda4e33b898b4f048ddac8d73a31bac6f4738fb2b4ff0b. Source text hash: 02990231dda4f28cad5c436264f7605ce2fbbad41da4649c31ba784387b24e38.
Revision and source-check history
Version 1 · 2026-11-10T21:16:33.095Z
arXiv cs.AI: Med-DualLoRA: Local Adaptation of Foundation Models for 3D Cardiac MRI
The authors of Med-DualLoRA propose a way to adapt a for disease detection in 3D cardiac MRI across multiple sites, while keeping each site’s local adaptations private. Their method separates adaptations into a global component, shared and aggregated across sites, and local components that stay at each site. The authors argue this design can personalize the model while reducing communication cost, addressing two challenges they identify: clinical data may not be centralized because of privacy constraints, and data from different sites can vary. In experiments using ACDC and combined M&Ms datasets, with each vendor treated as a federated client, they report improved performance over other federated PEFT baselines, including balanced accuracy of 0.768 and specificity of 0.612. They also report that adapting only two transformer blocks preserves performance and further improves efficiency. These findings describe the paper’s evaluation; the abstract does not establish how the method performs in routine clinical use or beyond the evaluated datasets and setup.
Original source
cs.AI · Article
Source details
Original source date: Mar 11, 2026 (2026-03-11).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-10T21:16:33.095Z. First observed: 2026-11-10T08:39:34.005Z. Retrieved: 2026-11-10T21:06:39.738Z.
Limits
The abstract reports experiments on ACDC and combined M&Ms datasets, with each vendor treated as a federated client; it does not establish performance beyond that evaluated setup. The abstract describes disease-detection experiments, not performance in routine clinical use.
Open question
How does Med-DualLoRA perform on cardiac MRI data beyond the ACDC and combined M&Ms datasets and vendor-based setup evaluated?
Source-stated
The authors propose Med-DualLoRA, which separates shared and local adaptations; only the shared component is exchanged and aggregated across sites.
only the global component is shared and aggregated across sites, keeping local adapters private.
Source-stated
The authors say this design improves personalization and reduces communication cost.
This design improves personalization while significantly reducing communication cost
Source-stated
In the reported evaluation, Med-DualLoRA achieved balanced accuracy of 0.768 and specificity of 0.612, outperforming other federated PEFT baselines.
Med-DualLoRA achieves statistically significant improved performance (balanced accuracy 0.768, specificity 0.612) compared to other federated PEFT baselines
Source-stated
The authors report that adapting only two transformer blocks preserved performance and further improved efficiency.
adapting only two transformer blocks preserves performance while further improving efficiency.
Source-stated
The evaluation used ACDC and combined M&Ms datasets and treated each vendor as a federated client.
using ACDC and combined M\&Ms datasets, treating each vendor as a federated client.
Source and contributor context
Perramon-Llussà, Joan, Jiménez-Sánchez, Amelia, Skorupko, Grzegorz, Avgoustidis, Fotis, Martín-Isla, Carlos, Lekadir, Karim, Gkontra, Polyxeni · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Foundation model
A large model trained on broad data that can be adapted to many different tasks.
Statistical significance
A test of whether a result is unlikely to be due to chance alone; it does not say how large or important an effect is.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 0b8b184f-ab54-4af0-8e8f-ee1d96b55283.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-10T21:16:33.095Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 860617ba2522343859cda4e33b898b4f048ddac8d73a31bac6f4738fb2b4ff0b. Source text hash: 02990231dda4f28cad5c436264f7605ce2fbbad41da4649c31ba784387b24e38.
passed · 2026-11-10T21:16:33.095Z · anthropic/claude-opus-5.5
Publication: 0b8b184f-ab54-4af0-8e8f-ee1d96b55283
attribution — pass: The reading attributes the method, its design rationale and the results to the paper's authors. It does not present them as established fact or as Escape Meridian's view. The 'arXiv cs.AI' label matches the publisher feed; the paper is primarily cs.CV and cross-listed in cs.AI.
Authors:Joan Perramon-Llussà, Amelia Jiménez-Sánchez, Grzegorz Skorupko, Fotis Avgoustidis, Carlos Martín-Isla, Karim Lekadir, Polyxeni Gkontra
summary — pass: The summary faithfully states the proposal: shared global and private local adaptations, with only the global part aggregated. It covers the privacy and site-variation motivations, the results and the two-block finding, and the question follows from the evidence. A minor omission is the source's point that single-site adaptation risks bias. The acronym PEFT (parameter-efficient fine-tuning), federated client and LoRA would benefit from term-library tooltips, but the core idea remains understandable.
adaptation using single-site data may lead to suboptimal performance and increased model bias
date — pass: originalPublished is 2026-03-11, which matches the v1 submission date. It is not the v2 revision date (17 Mar 2026) or the later feed date.
[Submitted on 11 Mar 2026 (v1), last revised 17 Mar 2026 (this version, v2)]
support — pass: Each claim matches the abstract verbatim in substance and remains framed as author-reported. The limitations correctly note that the evaluation is confined to the stated datasets and setup and does not show routine clinical use. No unsupported benefits, costs or forecasts are added.
Med-DualLoRA achieves statistically significant improved performance (balanced accuracy 0.768, specificity 0.612) compared to other federated PEFT baselines
media — pass: The selected media is the verified arXiv abstract page for this paper. It is the direct original source and the most relevant access point.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A large model trained on broad data that can be adapted to many different tasks.
Definition referenceA test of whether a result is unlikely to be due to chance alone; it does not say how large or important an effect is.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA large model trained on broad data that can be adapted to many different tasks.
Definition referenceA test of whether a result is unlikely to be due to chance alone; it does not say how large or important an effect is.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv