We studied the formation process of a mayenite structure from hydroxide precursors in different gas media. According to X-ray diffraction data, this method allows a well-crystallized mayenite (Ca₁₂Al₁₄O₃₃ or C12A7) phase to be obtained at low (500–900 °C) temperatures with an insignificant impurity of CaO. It was shown that the lattice parameters for C12A7 obtained in an inert atmosphere (Ar) were lower when compared with similar samples in the air. These results can be explained by the different levels of oxygen nonstoichiometry in the resulting phase. We noted that sintering and crystallization of mayenite proceeds at lower temperatures in Ar than in the air medium. We found the presence of donor and acceptor active sites on the surface of mayenite, which was detected by the spin probe method. The specific (per unit surface) concentration of such sites (2.5 × 10¹⁶ m⁻² and 1.5 × 10¹⁵ m⁻² for donor and acceptor sites, respectively) is comparable to that of γ-Al₂O₃, which is traditionally used as catalyst support. This allows it to be used in adsorption and catalytic technologies, taking into account its high specific surface area (~30–50 m²/g at a low synthesis temperature).

The evolution of the structure and the phase composition of a dispersed mayenite at its interaction with metallic aluminum was studied in a temperature range from 900 to 1400 °C in both argon and air atmospheres. The aluminum loading was varied from 0 to 50 wt%. It was found that the addition of aluminum significantly affects the stability of the mayenite and other calcium aluminate phases within the studied temperature range. The formation of the electride state registered by the appearance of a characteristic electron paramagnetic resonance (EPR) signal from F⁺-like centers (g~1.994) in an argon atmosphere was shown to take place already at 1150 °C due to an aluminothermic reduction of this material. The super-narrow (H_p-p < 0.5 G) EPR spectra from F⁺-like centers, which were recently observed for the core–shell structures of the C12A7@C type only, were registered for mayenite for the first time. The results obtained in the present study testify firstly towards the possibility of significantly diminishing the temperatures required for the formation of the electride state in such systems and secondly towards the ability to stabilize the size of small electride nanoparticles within the synthesized calcium aluminate matrix.

The development of foundation models for brain MRI depends critically on the scale, diversity, and consistency of available data, yet systematic assessments of these factors remain scarce. In this study, we analyze 54 publicly accessible brain MRI datasets encompassing over 538,031 scans to provide a structured, multi-level overview tailored to foundation model development. At the dataset level, we characterize modality composition, disease coverage, and dataset scale, revealing strong imbalances between large healthy cohorts and smaller clinical populations. At the image level, we quantify voxel spacing, orientation, and intensity distributions across 14 representative datasets, demonstrating substantial heterogeneity that can influence representation learning. We then perform a quantitative evaluation of preprocessing variability, examining how intensity normalization, bias field correction, skull stripping, spatial registration, and interpolation alter voxel statistics and geometry. While these steps improve within-dataset consistency, residual differences persist between datasets. Finally, a feature-space case study using a 3D DenseNet121 shows measurable residual covariate shift after standardized preprocessing, confirming that harmonization alone cannot eliminate inter-dataset bias. Together, these analyses provide a unified characterization of variability in public brain MRI resources and emphasize the need for preprocessing-aware and domain-adaptive strategies in the design of generalizable brain MRI foundation models.

Background: In this study, we examined the effectiveness of transfer learning in improving automatic segmentation of brain metastases on magnetic resonance imaging scans, with potential applications in preventive exams and remote diagnostics. Methods: We trained three deep learning models on a public dataset from the ASNR-MICCAI Brain Metastasis Challenge 2024, fine-tuned them on a small private dataset, and compared their performance to models trained from scratch. Results: Results showed that models using transfer learning performed better than scratch-trained models, though the improvement was not statistically substantial. The custom Tversky and Binary Cross-Entropy loss function helped manage class imbalance and reduce false negatives, limiting missed tumor regions. Medical experts noted that, while fine-tuned models worked well with larger, well-defined tumors, they struggled with tiny, scattered tumors in complex cases. Conclusions: This study highlights the potential of transfer learning and tailored loss functions in medical imaging, while also pointing out the models’ limitations in detecting very small tumors in challenging cases.