.png){kind=icon}
.png){kind=icon}
Return
Medical hierarchical image classification via dual-geometry image–text learning
.png){kind=icon}
OA
AI
L
A
E
Z
Z
Y
DOI：10.1016/j.media.2026.104120.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose a dual-geometry imagetext framework, termed H CL, which appends a lightweight classifier head atop standard backbones and integrates features from both Euclidean and hyperbolic spaces for hierarchical medical image classification. • We introduce a group CL strategy to capture multi-level label dependencies, and adapt hyperbolic entailment from MERU to a two-level hierarchical imagetext alignment setting, extending single-level textimage entailment to a nested parent child structure that jointly constrains text and image representations across hierarchy levels. • Extensive experiments on the HiCervix microscopy dataset, the MoleMap dermoscopy dataset, and the UIdataGB ultrasound dataset demonstrate the superior performance of our framework with 16% higher accuracy compared to advanced methods.
Keywords:
dual-geometry
hierarchical classification
medical image analysis
hyperbolic space
image-text learning
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
11.8
Papers:
3.7K
Citations:
2.4W
Organization
Cited Papers