Due to increasingly ageing populations and a rising demand on quality of life hip replacement became one of the most common operations in orthopedics and trauma surgery over the last decades and even continues to gain importance. However, in more than 10 percent of cases, loosening of the implanted prosthesis occurs within the first 15 years after surgical implantation. As a result, the prosthesis usually has to be completely replaced in a complex operation which often leads to complications. The later the prosthesis loosening is detected, the more difficult the initial situation for a successful and complication-free prosthesis replacement is. In addition to early diagnosis, it is essential to distinguish between purely mechanical (aseptic) loosening and loosening caused by bacterial infection (septic). Due to a lack of accuracy, conventional diagnostic methods such as projection radiography cannot reliably diagnose loosening in its early stages or detect whether it is associated with the formation of a biofilm in the bone-implant interface.

In this work, we developed a non-invasive ultrasound-based measurement procedure for quantifying the thickness of the layer between the bone and the stem of a hip prosthesis as a correlate to loosening. In principle, it also allows for material characterization of the interface.

The introduced method is based on a well-known analytical model of the reflection of sound waves in a three-layer system being mainly used for characterizing the thickness of thin lubricant films. For the desired medical application of characterizing the bone-implant interface which comes along with rough surfaces, inhomogeneous materials and the impossiblity of reference measurements, we adapted and extended this model and developed suitable data processing algorithms for analyzing the interface.

With respect to determining the layer thickness, the procedure was experimentally validated at different idealized test systems and a more realistic bone-implant system within in the range of approx. 200 µm to 2 mm. Consequently, our procedure is able to quantify the thickness of thin interlayers even at rough, porous and inhomogeneous materials and without a reference measurement.

After further optimization and adaptation, as well as further experimental tests, the procedure offers great potential to significantly improve the diagnosis of prosthesis loosening at an early stage and may also bei applicable to detect the formation of a biofilm.

Objective
Federated Learning (FL) enables collaborative training of artificial intelligence (AI) models from multiple data sources without directly sharing data. Due to the large amount of sensitive data in dentistry, FL may be particularly relevant for oral and dental research and applications. This study, for the first time, employed FL for a dental task, automated tooth segmentation on panoramic radiographs.
Methods
We employed a dataset of 4,177 panoramic radiographs collected from nine different centers (n = 143 to n = 1881 per center) across the globe and used FL to train a machine learning model for tooth segmentation. FL performance was compared against Local Learning (LL), i.e., training models on isolated data from each center (assuming data sharing not to be an option). Further, the performance gap to Central Learning (CL), i.e., training on centrally pooled data (based on data sharing agreements) was quantified. Generalizability of models was evaluated on a pooled test dataset from all centers.
Results
For 8 out of 9 centers, FL outperformed LL with statistical significance (p<0.05); only the center providing the largest amount of data FL did not have such an advantage. For generalizability, FL outperformed LL across all centers. CL surpassed both FL and LL for performance and generalizability.
Conclusion
If data pooling (for CL) is not feasible, FL is shown to be a useful alternative to train performant and, more importantly, generalizable deep learning models in dentistry, where data protection barriers are high.
Clinical Significance
This study proves the validity and utility of FL in the field of dentistry, which encourages researchers to adopt this method to improve the generalizability of dental AI models and ease their transition to the clinical environment.

For the investigation of moisture and salt content in historic masonry, destructive drilling samples followed by a gravimetric investigation is still the preferred method. In order to prevent the destructive intrusion into the building substance and to enable a large-area measurement, a nondestructive and easy-to-use measuring principle is needed. Previous systems for moisture measurement usually fail due to a strong dependence on contained salts. In this work, a ground penetrating radar (GPR) system was used to determine the frequency-dependent complex permittivity in the range between 1 and 3 GHz on salt-loaded samples of historical building materials. By choosing this frequency range, it was possible to determine the moisture in the samples independently of the salt content. In addition, it was possible to make a quantitative statement about the salt level. The applied method demonstrates that with ground penetrating radar measurements in the frequency range selected here, a salt-independent moisture determination can be carried out.

Material characterization using contactless excited and detected guided acoustic waves is a well proven, but overall expensive approach, since a cheap method of excitation is missing so far. Ignition-spark-excitation of Lamb waves could be such a method but has not shown to be suitable for material characterization yet. Covering various thicknesses and materials of metal plates, systematic dependencies of the spectral amplitude of the ignition-spark-excited Lamb waves are presented in this work. This enables the approach to be an alternative for exciting Lamb waves for material characterization.

Material characterization using contactless excited and detected guided acoustic waves is a well proven, but overall expensive approach, since a cheap method of excitation is missing so far. Ignition-spark-excitation of Lamb waves could be such a method but has not shown to be suitable for material characterization yet. Covering various thicknesses and materials of metal plates, systematic dependencies of the spectral amplitude of the ignition-spark-excited Lamb waves are presented in this work. This enables the approach to be an alternative for exciting Lamb waves for material characterization.

Non-coding RNA (ncRNA) classes take over important housekeeping and regulatory functions and are quite heterogeneous in terms of length, sequence conservation and secondary structure. High-throughput sequencing reveals that the expressed novel ncRNAs and their classification are important to understand cell regulation and identify potential diagnostic and therapeutic biomarkers. To improve the classification of ncRNAs, we investigated different approaches of utilizing primary sequences and secondary structures as well as the late integration of both using machine learning models, including different neural network architectures. As input, we used the newest version of RNAcentral, focusing on six ncRNA classes, including lncRNA, rRNA, tRNA, miRNA, snRNA and snoRNA. The late integration of graph-encoded structural features and primary sequences in our MncR classifier achieved an overall accuracy of \textgreater97%, which could not be increased by more fine-grained subclassification. In comparison to the actual best-performing tool ncRDense, we had a minimal increase of 0.5% in all four overlapping ncRNA classes on a similar test set of sequences. In summary, MncR is not only more accurate than current ncRNA prediction tools but also allows the prediction of long ncRNA classes (lncRNAs, certain rRNAs) up to 12.000 nts and is trained on a more diverse ncRNA dataset retrieved from RNAcentral.