A Comprehensive Survey on Deep Learning in Abdominal Imaging: Datasets, Techniques, and Performance Metrics

Abstract

Integrating Deep Learning (DL) into abdominal imaging represents a significant leap forward in diagnosing and managing abdominal conditions, offering the potential to transform conventional medical practices. This comprehensive survey explores the application of DL techniques, such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Generative Adversarial Networks (GAN), across various domains of abdominal imaging, including liver, spleen, kidney, and other structures such as subcutaneous adipose tissue (SAT), muscle, viscera, and bone. It discusses the critical role of performance metrics in evaluating model efficacy and clinical applicability. Furthermore, the paper highlights emerging trends in DL, such as integrating multimodal data and exploring unsupervised and semi-supervised learning techniques, which promise to address current limitations and pave the way for future advancements. Ethical considerations, including algorithmic bias, transparency in model development, and equitable patient care, are thoroughly examined to underscore the importance of ethical practices in deploying Artificial Intelligence (AI) technologies in healthcare.