PALM-MTL: Progressive Adaptive LoRA-based Multi-Task Learning for Pathological Myopia Analysis

doi:10.1016/j.eswa.2026.134245

PALM-MTL: Progressive Adaptive LoRA-based Multi-Task Learning for Pathological Myopia Analysis

2026-09-02

0

PRE

AI

DOI：10.1016/j.eswa.2026.134245

Original

Original request for help

Save

Abstract

En 中文

• Constructed the first PMMA dataset for pathological myopia multitask analysis. • Proposed PALM-MTL, a SAM-based parameter-efficient multitask framework. • ELT dynamically adjusts LoRA ranks to reduce redundancy and overfitting. • HPHA enhances hierarchical multitask feature representation of lesions. • PALM-MTL achieves superior performance on classification, segmentation, and VQA.

Keywords:

Pathological Myopia

Multi-Task Learning

Medical Image Analysis

Low-Rank Adaptation

PALM-MTL: Progressive Adaptive LoRA-based Multi-Task Learning for Pathological Myopia Analysis

Abstract

Journal

Organization