Dynamic Sub-Cluster-Aware Network for Few-Shot Skin Disease Classification

TL;DR

This paper introduces SCAN, a novel network that improves few-shot skin disease classification by capturing sub-clustered structures within classes, leading to better diagnostic accuracy for rare skin conditions.

Contribution

The paper proposes a dual-branch framework with a cluster loss and a purity loss to learn sub-clustered representations within each skin disease class for few-shot learning.

Findings

Outperforms state-of-the-art methods by 2-5% on SD-198 and Derm7pt datasets.

Effectively captures sub-cluster structures within classes, improving classification metrics.

Demonstrates robustness in diagnosing rare skin diseases with limited data.

Abstract

This paper addresses the problem of few-shot skin disease classification by introducing a novel approach called the Sub-Cluster-Aware Network (SCAN) that enhances accuracy in diagnosing rare skin diseases. The key insight motivating the design of SCAN is the observation that skin disease images within a class often exhibit multiple sub-clusters, characterized by distinct variations in appearance. To improve the performance of few-shot learning, we focus on learning a high-quality feature encoder that captures the unique sub-clustered representations within each disease class, enabling better characterization of feature distributions. Specifically, SCAN follows a dual-branch framework, where the first branch learns class-wise features to distinguish different skin diseases, and the second branch aims to learn features which can effectively partition each class into several groups so as to preserve the sub-clustered structure within each class. To achieve the objective of the second branch, we present a cluster loss to learn image similarities via unsupervised clustering. To ensure that the samples in each sub-cluster are from the same class, we further design a purity loss to refine the unsupervised clustering results. We evaluate the proposed approach on two public datasets for few-shot skin disease classification. The experimental results validate that our framework outperforms the state-of-the-art methods by around 2% to 5% in terms of sensitivity, specificity, accuracy, and F1-score on the SD-198 and Derm7pt datasets.