An Overview of Deep Learning Based Small Sample Medical Imaging Classification

Deep Learning (DL) has been proven to be a promising technique for image analysis tasks such as image classification and object recognition. Compared with other fields, the accuracy of DL tasks in medical imaging depends heavily on the dataset volume. However, DL has been suffering from the problem of small sample datasets caused by a variety of ethical and financial reasons in medical imaging. Data augmentation and transfer learning are the two most commonly used approaches to enhance the practicability of the DL algorithms in medical imaging. This article discusses the data augmentation methods including image manipulation and generative adversarial networks. Feature-extracting and fine-tuning methods of transfer learning are also discussed. Finally, the paper mentions the real-life applications of many architectures, advantages and disadvantages, and future works.