Escarcitys: A framework for enhancing medical image classification performance in scarcity of trainable samples scenarios

Tianxiang Wang, Qun Dai, Wei Xiong

Abstract: Highlights•EScarcityS Framework: Introduces a novel model-level strategy to augment data generation methods, significantly enhancing classification accuracy while reducing data dependency.•Component Architecture: Proposes a multi-component architecture comprising a Multi-Granularity Vision Transformer (MGViT), a Disease Probability Map Fusion module, and a DPM-guided data generator.•MGVit Innovation: Develops a progressive granularity transition in MGViT for integrated feature learning.•Disease Probability Map Fusion: Fuses detailed disease probability maps to boost interpretability.•Clinical Interpretability: Ensures clinical relevance by producing meaningful, interpretable outputs that bridge computational models and medical practice.