Comparative Analysis of Lightweight Deep Learning Models for Classification of Medical Images

Document Type : Original Article

Authors

1 Scientific Computing Department, Faculty of Computer and Information Science, Ain Shams University, Cairo, Egypt

2 Scientific Computing Department, Faculty of Computer & Information Sciences, Ain Shams University, Cairo, Egypt

Abstract

Accurate medical imaging analysis is crucial for clinical decision-making and effective diagnosis. While deep learning has shown impressive results in different vision tasks, including medical image classification tasks, many of these models are designed to be computationally intensive and come with large number of parameters and high computational cost, making them impractical for deployment on resource-constrained and edge devices. Recent advances have introduced efficient lightweight models that can achieve comparable results, while being resource efficient and suitable for mobile and embedded applications. In this paper, we perform a comprehensive comparison of recent state-of-the art lightweight models that fall under three different categories, including Convolutional neural networks (CNNs), Vision Transformers (ViTs), and hybrid approaches that combine the strengths of both paradigms. These models are evaluated on multiple medical imaging tasks. Specifically, we conduct experiments using HAM-10000 skin lesion dataset and brain tumor dataset for skin and brain cancer classification tasks, respectively.

Keywords