A Review on Deep Learning Techniques for Medical Image Segmentation and Classification

The intersection of artificial intelligence (AI) with medical imaging has advanced at an unprecedented rate over the last decade, enabling the diagnosis of disease, the planning of treatment, and monitoring of patients. Deep learning, specifically convolutional neural networks and their derivatives, has been employed to interpret medical images with specialized methods and proven high accuracy in working with complex forms of medical imaging. This review summarizes the most well-known of deep learning architectures in the implementation of medical image analysis, including CNNs, U-Net, ResNet, DenseNet, GANs, and transformer architecture-based models. We discuss the most common applications (e.g., tumor detection, organ segmentation of medical images and image enhancement) associated with each of the architectures. Even with successful applications of AI analyses in medical imaging, specific challenges remain that limit adoption of AI tools by the clinical community that range from lack of annotated data to questions of interpretability and ultimately clinical applicability. We highlight the current work and future directions intended to mitigate main challenges, thus maximizing the potential of AI in clinical practice. This article serves as a detailed overview of AI and medical imaging, and will serve as a deep resource to the reader that wishes to engage in the discussion of AI and its applications in medical imaging.