سامانه همایش های دانشگاه آزاد اسلامی واحد نجف آباد

صفحه اصلی / سومین کنفرانس بین المللی مهندسی و علوم کامپیوتر و اولین کنفرانس ملی سامانه های هوشمند، کلان داده و رایانش پیشرفته

A Survey of Deep Learning Approaches for Medical Image Analysis: From CNNs to Hybrid CNN-Transformer Models

نویسندگان :

Pourya Pouryeganeh¹ Farzaneh Kaviani²

1- دانشگاه آزاد خوراسگان 2- دانشگاه آزاد خوراسگان

کلمات کلیدی :

Medical Image Segmentation،Convolutional Neural Networks،Vision Transformer

چکیده :

Medical image segmentation is a critical task for accurate diagnosis, treatment planning, and disease monitoring in various healthcare applications. Traditionally, convolutional neural networks (CNNs) have dominated this field, excelling at local feature extraction. However, CNNs face limitations in capturing global context and long-term dependencies in images. The emergence of Vision Transformers (ViTs), initially successful in natural language processing, has provided a new approach, leveraging self-attention mechanisms to learn global relationships. This paper reviews recent advancements in applying ViTs and hybrid architectures combining CNNs and ViTs for medical image segmentation, drawing upon the provided sources. We discuss various proposed models, their architectural components designed to enhance feature representation and efficiency, and their evaluation on diverse medical imaging datasets. The review highlights the superior performance of ViT-based and hybrid models compared to traditional CNNs, particularly in capturing both local and global information and improving segmentation accuracy. Challenges, such as data requirements and efficiency, and future research directions are also explored.