Background: The use of machine learning in medical diagnosis and treatment has grown significantly in recent years with the development of computer-aided diagnosis systems, often based on annotated medical radiology images. However, the lack of large annotated image datasets remains a major obstacle, as the annotation process is time-consuming and costly. This study aims to overcome this challenge by proposing an automated method for annotating a large database of medical radiology images based on their semantic similarity. Results: An automated, unsupervised approach is used to create a large annotated dataset of medical radiology images originating from the Clinical Hospital Centre Rijeka, Croatia. The pipeline is built by data-mining three different types of medical data: images, DICOM metadata and narrative diagnoses. The optimal feature extractors are then integrated into a multimodal representation, which is then clustered to create an automated pipeline for labelling a precursor dataset of 1,337,926 medical images into 50 clusters of visually similar images. The quality of the clusters is assessed by examining their homogeneity and mutual information, taking into account the anatomical region and modality representation. Conclusions: The results indicate that fusing the embeddings of all three data sources together provides the best results for the task of unsupervised clustering of large-scale medical data and leads to the most concise clusters. Hence, this work marks the initial step towards building a much larger and more fine-grained annotated dataset of medical radiology images.

Predicting the optimal recovery time following a soccer player’s injury is a complex task with heavy implications on team performance. While most current decision-based models rely on the physician’s perspective, this study proposes a machine learning (ML)-based approach to predict recovery duration using three modeling techniques: linear regression, decision tree, and extreme gradient boosting (XGB). Performance is compared between the models, against the expert, and together with the expert. The results demonstrate that integrating the expert’s predictions as a feature improves the performance of all models, with XGB performing best with a mean $R^2$ score of $0.72$, outperforming the expert’s predictions with an $R^2$ score of $0.62$. This approach has significant implications for sports medicine, as it could help teams make better decisions on the return-to-play of their players, leading to improved performance and reduced risk of re-injury.

Digital Imaging and Communication in Medicine (DICOM) is a standardized format for storing medical images enriched with descriptive data. It consists of a number of informative header tags. E.g., these tags contain information concerning imaging technique, patient weight, patient age and so on. In machine learning, these tags can, among other things, be used for categorization, classification, and general manipulation of DICOM data. However, because of their number and variety, using them for automation undoubtedly impacts computational performance. To tackle this problem, the possibility of using auto encoders is explored in this manuscript. It was hypothesized that clustering data compressed with autoencoders can achieve the same results as when working with raw - uncompressed data. To support this claim, clustering of compressed and uncom- pressed data was performed on a dataset of 25, 000 DICOM files from the Clinical Hospital Centre Rijeka PACS. The results show no significant difference between compressed and uncompressed data (p > 0.05), thus confirming that equally good clustering results can be accomplished using a smaller representation.