Deep learning (DL) models enable the automatic extraction of imaging features to enhance the performance of a model for a given task. DL is a subfield of machine learning that directly processes raw data using artificial neural networks. Deep neural networks make predictive models possible from start to finish by automating all of the steps that go into making a traditional machine learning model, such as extracting features and learning. Deep neural networks are representation-learning algorithms that consist of a stack of processing layers with a finite number of nonlinear units (i.e., artificial neurons). In a network, the input and output layers are the first and last layers, while the hidden layers are stacked between them. Deep neural networks make predictive models possible as nonlinear function approximators by learning multiple representations of the input data at multiple abstraction levels. Depending on the number of layers and units per layer, a DL model can easily reach millions of trainable parameters that must be estimated during training. Consequently, DL models are prone to overfitting, especially when working with small training sets, and should only be used with datasets comprising thousands of images. To extract actionable insights from the provided data, data scientists use a variety of machine learning algorithms, the majority of which are supervised learning problems because you already know what you need to predict. You are given a wealth of information to assist you in achieving your goal. Unsupervised learning, on the other hand, is a difficult task, though with numerous advantages. It has gained a lot of traction in the machine learning and deep learning communities with its potential to resolve previously unsolvable problems.

Globally, tens of millions of contrast-enhanced magnetic resonance imaging (MRI) examinations are performed each year. To improve diagnostic accuracy, small, hydrophilic gadolinium (III)-based chelates are almost exclusively used as contrast agents. Concerns about the long-term safety of these compounds have prompted research into alternatives in recent years. Furthermore, efforts have been made to develop new contrast agents or agents that can detect molecularly targeted pathological changes in the immediate environment. This comprehensive review looks at the current status of clinically approved contrast agents as well as their action and safety concern mechanisms.

This Special Issue aims to provide an overview of approved MRI contrast agents, their applications, and the risks associated with them; to proffer an understanding of the molecular mechanisms that cause MR contrast from relaxation and exchange; and, lastly, to review the coordination chemistry of these agents as well as the factors that go into designing an agent safe for human use, employing unsupervised learning methodology. Then, in addition to molecularly targeted and activatable agents, we'll look at new gadolinium (III)-based and non-gadolinium (III)-based contrast agents. The goal of this Special Issue is to learn more about the current state of MRI contrast agents, the problems they face, and the unsupervised learning strategies that have been applied to address them. We welcome both original research and review articles.

Potential topics include but are not limited to the following:

• Promising applications of digital pathology and automated image diagnosis
• Artificial intelligence for the selection of low or high osmolar iodine-based contrast agents for MRI
• Unsupervised DL-based diagnostic support and personalized therapy
• Unsupervised DL-assisted automatic analysis and interpretation of MRI scans
• Contrast-induced nephropathy diagnosis through unsupervised DL
• Unsupervised DL-based micro investigative radiology
• Prediction of prognostics through clinical data and imaging integration