In the context of bone-related injuries, radiography is the most commonly used technique for noninvasive diagnosis. Standard procedure involves a technician who obtains a high-quality X-ray image and a radiologist who sets a diagnosis based on the obtained image. To make diagnostic procedures easier, faster and more accurate, computer-aided diagnostics (CADx) systems are being developed to support the radiologists in their decision-making process. In recent years, machine learning (ML) has become the focus of CADx systems development and research because it is capable of seamlessly capturing highly complex distributions. This is especially important in pediatric radiology, where the variations in the data are often very demanding for modeling. This doctoral dissertation first investigates preprocessing of pediatric X-ray images and the detection of targeted features related to wrist fractures on wrist radiographs of children. Inspecting the dataset of pediatric X-ray images provided by the Division of Pediatric Radiology, Department of Radiology, Medical University of Graz, Austria, the issue of X-ray image alignment and orientation arose. Therefore, the alignment and orientation of X-ray images became the focus of the first developed ML model. Following image alignment and orientation, targeted features extraction is needed, such that would help radiologists detect wrist fractures on pediatric X-ray images. Two separate ML models for pediatric wrist fracture detection were developed. The first developed model was based on local-entropy bone segmentation, while the second model utilized a YOLOv4 convolutional neural network to cope with the shortcomings of the first developed model. Besides fracture detection, it is helpful to estimate the age of the fracture. Therefore, another deep learning-based system was developed for fracture age estimation. The developed multimodal system based on fusing multiple X-ray projections (of the same case) with a patient’s age and gender information provides uncertainty in its decision. By estimating uncertainty in its decisions, the system becomes more trustworthy for the experts who use it. Finally, to enhance the visibility of the fractures and tissue obstructed by the cast during the wrist fracture healing monitoring, a CycleGAN-based system was trained for cast suppression in X-ray images. Also, in order to appropriately evaluate the developed system, rigorous quantitative and qualitative evaluation was proposed. All things considered, merging all developed models into one system creates a backbone of a CADx system capable of providing crucial information about pediatric wrist fractures that could improve diagnostics and make the whole process less labored for radiologists.