Accurate fetal brain tissue segmentation and biometric measurement are essential for monitoring neurodevelopment and detecting abnormalities in utero. The Fetal Tissue Annotation (FeTA) Challenges have established robust multi-center benchmarks for evaluating state-of-the-art segmentation methods. This paper presents the results of the 2024 challenge edition, which introduced three key innovations. First, we introduced a topology-aware metric based on the Euler characteristic difference (ED) to overcome the performance plateau observed with traditional metrics like Dice or Hausdorff distance (HD), as the performance of the best models in segmentation surpassed the inter-rater variability. While the best teams reached similar scores in Dice (0.81-0.82) and HD95 (2.1-2.3 mm), ED provided greater discriminative power: the winning method achieved an ED of 20.9, representing roughly a 50% improvement over the second- and third-ranked teams despite comparable Dice scores. Second, we introduced a new 0.55T low-field MRI test set, which, when paired with high-quality super-resolution reconstruction, achieved the highest segmentation performance across all test cohorts (Dice=0.86, HD95=1.69, ED=6.26). This provides the first quantitative evidence that low-cost, low-field MRI can match or surpass high-field systems in automated fetal brain segmentation. Third, the new biometry estimation task exposed a clear performance gap: although the best model reached a mean average percentage error (MAPE) of 7.72%, most submissions failed to outperform a simple gestational-age-based linear regression model (MAPE=9.56%), and all remained above inter-rater variability with a MAPE of 5.38%. Finally, by analyzing the top-performing models from FeTA 2024 alongside those from previous challenge editions, we identify ensembles of 3D nnU-Net trained on both real and synthetic data with both image- and anatomy-level augmentations as the most effective approaches for fetal brain segmentation. Our quantitative analysis reveals that acquisition site, super-resolution strategy, and image quality are the primary sources of domain shift, informing recommendations to enhance the robustness and generalizability of automated fetal brain analysis methods.

Background: Total marrow (lymphoid) irradiation (TMI/TMLI) is a radiotherapy treatment used to selectively target the bone marrow and lymph nodes in conditioning regimens for allogeneic hematopoietic stem cell transplantation. A complex field geometry is needed to cover the large planning target volume (PTV) of TMI/TMLI with volumetric modulated arc therapy (VMAT). Five isocenters and ten overlapping fields are needed for the upper body, while, for patients with large anatomical conformation, two specific isocenters are placed on the arms. The creation of a field geometry is clinically challenging and is performed by a medical physicist (MP) specialized in TMI/TMLI.Purpose: To develop convolutional neural networks (CNNs) for automatically generating the field geometry of TMI/TMLI.Methods: The dataset comprised 117 patients treated with TMI/TMLI between 2011 and 2023 at our Institute. The CNN input image consisted of three channels, obtained by projecting along the sagittal plane: (1) average CT pixel intensity within the PTV; (2) PTV mask; (3) brain, lungs, liver, bowel, and bladder masks. This “averaged” frontal view combined the information analyzed by the MP when setting the field geometry in the treatment planning system (TPS). Two CNNs were trained to predict the isocenters coordinates and jaws apertures for patients with (CNN‐1) and without (CNN‐2) isocenters on the arms. Local optimization methods were used to refine the models output based on the anatomy of the patient. Model evaluation was performed on a test set of 15 patients in two ways: (1) by computing the root mean squared error (RMSE) between the CNN output and ground truth; (2) with a qualitative assessment of manual and generated field geometries—scale: 1 = not adequate, 4 = adequate—carried out in blind mode by three MPs with different expertise in TMI/TMLI. The Wilcoxon signed‐rank test was used to evaluate the independence of the given scores between manual and generated configurations (p < 0.05 significant).Results: The average and standard deviation values of RMSE for CNN‐1 and CNN‐2 before/after local optimization were 15 ± 2/13 ± 3 mm and 16 ± 2/18 ± 4 mm, respectively. The CNNs were integrated into a planning automation software for TMI/TMLI such that the MPs could analyze in detail the proposed field geometries directly in the TPS. The selection of the CNN model to create the field geometry was based on the PTV width to approximate the decision process of an experienced MP and provide a single option of field configuration. We found no significant differences between the manual and generated field geometries for any MP, with median values of 4 versus 4 (p = 0.92), 3 versus 3 (p = 0.78), 4 versus 3 (p = 0.48), respectively. Starting from October 2023, the generated field geometry has been introduced in our clinical practice for prospective patients.Conclusions: The generated field geometries were clinically acceptable and adequate, even for an MP with high level of expertise in TMI/TMLI. Incorporating the knowledge of the MPs into the development cycle was crucial for optimizing the models, especially in this scenario with limited data.

Accurate fetal brain tissue segmentation and biometric analysis are essential for studying brain development in utero. The FeTA Challenge 2024 advanced automated fetal brain MRI analysis by introducing biometry prediction as a new task alongside tissue segmentation. For the first time, our diverse multi-centric test set included data from a new low-field (0.55T) MRI dataset. Evaluation metrics were also expanded to include the topology-specific Euler characteristic difference (ED). Sixteen teams submitted segmentation methods, most of which performed consistently across both high- and low-field scans. However, longitudinal trends indicate that segmentation accuracy may be reaching a plateau, with results now approaching inter-rater variability. The ED metric uncovered topological differences that were missed by conventional metrics, while the low-field dataset achieved the highest segmentation scores, highlighting the potential of affordable imaging systems when paired with high-quality reconstruction. Seven teams participated in the biometry task, but most methods failed to outperform a simple baseline that predicted measurements based solely on gestational age, underscoring the challenge of extracting reliable biometric estimates from image data alone. Domain shift analysis identified image quality as the most significant factor affecting model generalization, with super-resolution pipelines also playing a substantial role. Other factors, such as gestational age, pathology, and acquisition site, had smaller, though still measurable, effects. Overall, FeTA 2024 offers a comprehensive benchmark for multi-class segmentation and biometry estimation in fetal brain MRI, underscoring the need for data-centric approaches, improved topological evaluation, and greater dataset diversity to enable clinically robust and generalizable AI tools.

Text-to-image diffusion models can generate high-quality images from individual prompts, yet it remains unclear how reliably they can fuse two distinct concepts into a single coherent image in a strictly zero-shot setting. Inspired by human combinational creativity, we operationalize visual concept blending here as conditioning a diffusion model on a pair of text prompts and ask whether the resulting image preserves recognizable traits of both concepts while remaining visually coherent. We study four inference-time blending strategies that intervene at different stages of the diffusion process, including embedding-space interpolation, mid-denoising prompt switching, timestep-wise prompt alternation, and layer-wise conditioning within the denoising network. We evaluate these strategies across diverse blend categories, spanning object–object blends, compound concepts, style–content combinations, and architectural landmarks, and we complement the analysis with a user study involving 100 participants. Results indicate that no single strategy dominates across all categories: timestep-scheduling approaches are often preferred for producing recognisable and well-integrated hybrids, while embedding- and layer-based interventions exhibit characteristic strengths and failure modes in specific settings. Finally, we analyse practical control factors such as prompt ordering, blend ratio, and random seed, showing how they can substantially alter outcomes and providing concrete levers for steering blends more predictably in creative applications.

Most datasets encountered in computer vision and medical applications present symmetries that should be taken into account in classification tasks. A typical example is the symmetry by rotation and/or scaling in object detection. A common way to build neural networks that learn the symmetries is to use data augmentation. In order to avoid data augmentation and build more sustainable algorithms, we present an alternative method to mod out symmetries based on the notion of section of a principal fiber bundle. This framework allows to use simple metrics on the space of objects in order to measure dissimilarities between orbits of objects under the symmetry group. Moreover, the section used can be optimized to maximize separation of classes. We illustrate this methodology on a dataset of contours of objects for the groups of translations, rotations, scalings and reparameterizations. In particular, we present a 2-parameter family of canonical parameterizations of curves, containing the constant-speed parameterization as a special case, which we believe is interesting in its own right. We hope that this simple application will serve to convey the geometric concepts underlying this method, which have a wide range of possible applications.