Enhancing zero-shot brain tumor subtype classification via fine-grained patch-text alignment

Lubin Gan

doi:10.1016/j.eswa.2025.130161

Enhancing zero-shot brain tumor subtype classification via fine-grained patch-text alignment

2025-10-27

0

PRE

AI

DOI：10.1016/j.eswa.2025.130161

Original

Original request for help

Save

Abstract

En 中文

• FG-PAN enables zero-shot, fine-grained classification of brain tumor subtypes. • Local window attention and gated fusion improve patch-level feature discrimination. • LLM-generated pathology text prompts enhance visual-semantic alignment. • FG-PAN achieves robust, state-of-the-art results across diverse pathology datasets. • Fine-grained descriptions boost class separability and model adaptability.

Enhancing zero-shot brain tumor subtype classification via fine-grained patch-text alignment

Abstract

Journal

Organization