The proposed framework uses a shared-specific Mixture-of-LoRA architecture comprising one shared LoRA and six task-specific expert LoRAs, together with a two-stage training procedure, and achieves good results in the FLARE 2026 Task 3 test sets.
Abstract
Medical vision-language models (VLMs) allow a single model to perform clinical image analysis tasks ranging from diagnosis classification to report generation. However, joint adaptation is challenged by heterogeneous output formats, conflicting task gradients, and imbalanced training data. Hence, we present \textbf{Two-Stage Mixture-of-LoRA}, a framework built on MedGemma-1.5-4B. The framework uses a shared-specific Mixture-of-LoRA architecture comprising one shared LoRA and six task-specific expert LoRAs, together with a two-stage training procedure. In Stage 1, we jointly train the shared LoRA and all task-specific expert LoRAs on all tasks. In Stage 2, we first freeze the backbone, the shared LoRA, and all non-target experts, and refine one task expert at a time. Classification and regression then receive an additional modality-balanced continuation, in which smaller modality groups are repeated to match the largest group. In the FLARE 2026 Task 3 test sets, the proposed method achieves 0.85 balanced accuracy for classification, 0.48 micro-F1 for multi-label classification, 0.79 detection F1, and 17.39 regression MAE. Code is available at https://github.com/YuanYL03/MICCAI-FLARE-2026-Challenge-Task3-2D.
A shared backbones with task-specific linear heads for multi-task biomedical classification is evaluated, identifying a resolution domain shift between the MedMNIST API and evaluation environment and optimizing architecture-specific regularization substantially improved performance.
Ş. Gavril, Andrei Arhire, Adrian Iftene· arXiv.org· 0 citations
DeMoP, a language-model-guided mixture-of-experts framework that serializes structured patient profiles as natural-language sequences and learns adaptive prognostic representations from clinical variables, copy-number alterations, and gene descriptions, provides a unified approach to heterogeneous biomedical data integ...
Chen Tang, Lei Yu, Qi-Wei Li et al.· bioRxiv· 0 citations
Medical multimodal large language models (MLLMs) are increasingly expected not only to answer clinical questions, but also to localize the visual evidence behind their predictions. A common strategy connects a vision--language model (VLM) with SAM-style segmentation through a specialtoken, yet full-parameter training o...
Xu-Yang Cao, En-You Liu, Jun Zhao et al.· arXiv.org· 0 citations
MMOIR-VLM combines missing-modality-aware multimodal encoding, multi-task clinical prediction, and retrieval-grounded natural-language reasoning within a single modular framework.
Sohail Haresh Gidwani, Tamoghna Chattopadhyay, S. Thomopoulos et al.· Frontiers Comput. Neurosci.· 0 citations
Results suggest that LLM-guided distillation can effectively integrate clinical text reasoning with multimodal visual embeddings for robust brain tumor subtype classification and suggest that LLM-guided distillation can effectively integrate clinical text reasoning with multimodal visual embeddings for robust brain tum...
Sy-Duy Dang, D. Tran, Ngoc-Vuong Chu et al.· Conference and Labs of the E...· 1 citation
Alzheimer’s disease (AD) is a progressive neurode-generative disorder for which early and accurate diagnosis is critical for clinical management. Effective diagnosis depends on both disease staging and continuous cognitive monitoring through the Mini-Mental State Examination (MMSE), yet most existing deep learning-base...
Minh-Toan Le, Nhu-Y Tran-Van· International Conference on...· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduNov 20, 2026
Senior Research Scientist Jim Glass discusses how his MIT-IBM Computing Research Lab collaboration connects perception and language, and addresses the shortcoming of current models.
MIT News · Artificial Intelligence· news.mit.eduNov 20, 2026
Enforcing physics-based constraints prevents AI models from generating unreliable predictions for complex nonlinear systems in weather, climate, and turbulent flows.
An MIT and Dana-Farber Institute-led team used large language models to examine nearly 4,000 studies and hundreds of code repositories, revealing gaps in the information needed to scrutinize and reuse published research.
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.