Abstract:The application of deep learning-based architecture has seen a tremendous rise in recent years. For example, medical image classification using deep learning achieved breakthrough results. Convolutional Neural Networks (CNNs) are implemented predominantly in medical image classification and segmentation. On the other hand, transfer learning has emerged as a prominent supporting tool for enhancing the efficiency and accuracy of deep learning models. This paper investigates the development of CNN architectures using transfer learning techniques in the field of medical image classification using a timeline mapping model for key image classification challenges. Our findings help make an informed decision while selecting the optimum and state-of-the-art CNN architectures.
What problem does this paper attempt to address?
The main problems that this paper attempts to solve are: in the field of medical image classification, how to select the optimal convolutional neural network (CNN) architecture, and explore the use of transfer learning techniques to improve model efficiency and accuracy. Specifically, the paper aims to sort out the development of CNN architectures in the past eight years and their applications in medical image classification challenges through the timeline mapping model, thus helping researchers make more informed choices when facing multiple medical imaging modalities.
### Main problem decomposition:
1. **Selecting the optimal CNN architecture**:
- With the development of deep - learning technology, especially the wide application of CNN in medical image classification, a key problem faced by researchers is: among numerous CNN architectures, how to select the model most suitable for a specific task.
- The paper provides a detailed comparison of the performance of different architectures by reviewing and analyzing the performance of different CNN architectures in challenges such as ImageNet and BraTS, to help researchers select the optimal CNN architecture.
2. **Improving model efficiency and accuracy**:
- Transfer learning, as an effective tool, can help improve the generalization ability and accuracy of the model, especially in the case of limited data.
- The paper explores the application of transfer learning in CNN architectures, shows its potential in improving model performance, and summarizes the best practices of transfer learning.
3. **Dealing with multiple medical imaging modalities**:
- Medical image classification involves multiple imaging modalities (such as MRI, CT, X - ray, etc.), and each modality has its own unique characteristics and challenges.
- The paper provides researchers with a comprehensive perspective by analyzing the performance of CNN architectures under different modalities, helping them select appropriate models and methods when facing different types of medical images.
### Summary:
The core objective of this paper is to provide clear guidance for researchers in the field of medical image classification by systematically reviewing and analyzing the development of CNN architectures in the past eight years, especially in combination with the application of transfer learning techniques, helping them select the optimal CNN architecture and improve the efficiency and accuracy of the model.