Maria Baldeon-Calisto

Noisy social text can be transformed into semantic representations, but evaluating the resulting artifacts is difficult when source records, ontology activations, graph resources, conformance reports, and model inputs are considered separately. The evaluation problem is whether their connections remain inspectable as these artifacts are generated and used. This study examines the above continuity using 1614 Reddit posts about artificial intelligence and technology. Ontology activations produce sibling Resource Description Framework (RDF) and numeric representations linked through stable identifiers. Explicit correspondence checks and controlled fault tests examine these connections, while Shapes Constraint Language (SHACL) checks assess conformance to declared graph constraints. Existing verification checks detect 15 of 22 injected continuity faults. The remaining seven are detected only by supplementary diagnostics. Separate human assessment of 120 validation texts evaluates concept presence and local polarity against the source text, distinguishing extraction quality from artifact continuity. Downstream classification provides secondary evidence of modest mean macro-F1 gains in agreement with inherited lexical labels, varying by model across three seeds on one fixed split. The contribution is an evaluation of governed semantic artifact continuity at the post and variable levels, making the connections among these artifacts and their failure conditions clear and inspectable.

Objective. Vision foundation models (FMs) have shown strong potential for adapting to diverse downstream tasks without specialized fine-tuning. However, their performance in head and neck cancer (HNC) tumor segmentation remains insufficiently explored. This work provides a comprehensive evaluation of zero-shot and few-shot performance of five prominent FMs for HNC tumor segmentation. Approach. The FMs tested include SAM, MedSAM, nnInteractive, UniverSeg, and SegGPT. The evaluation of SAM, MedSAM, and nnInteractive utilized a bounding box prompting strategy. The influence of prompt precision is assessed by systematically enlarging the prompt bounding box until it reaches twice the size of the tightest bounding box enclosing the region of interest. For UniverSeg and SegGPT, the prompt consisted of example image-label pairs, and the effect of increasing the size of this support set is tested. Two-factor repeated measures ANOVAs were used for statistical analysis, confirming that both the FM architecture and the prompting strategy have a statistically significant effect on segmentation performance for both primary tumors (GTVp) and metastatic lymph nodes (GTVn). Main results. For SAM, MedSAM, and nnInteractive larger bounding boxes lead to notable performance degradation. Moreover, MedSAM outperformed SAM and nnInteractive for GTVp segmentation, whereas SAM provided slightly better results than the competitors for GTVn segmentation. For UniverSeg, performance improved as more image-label pairs are added to the support set, while SegGPT remained unaffected by this factor. Overall, UniverSeg and SegGPT demonstrate limited few-shot capability, with low mean dice scores across all scenarios. Comparisons against supervised convolutional neural network models trained specifically for HNC segmentation showed that conventional architectures remain more stable and clinically reliable. Significance. These findings highlight the ongoing challenges faced by current FMs in achieving reliable zero- and few-shot medical segmentation. The results emphasize the critical need for improved medical-domain adaptation and enhanced prompt robustness to advance the clinical utility of FMs in oncology.

This work introduces the MassSeg-Framework, a fully automatic two-stage pipeline for breast mass analysis in mammography that integrates YOLOv11-based detection with Chan–Vese ACM refinement to achieve accurate mass localization and segmentation with a lightweight computational footprint. The framework was trained and evaluated on two publicly available datasets using consistent experimental protocols. In the detection stage, YOLOv11-nano was the most effective architecture, with a confidence threshold of 0.4, achieving statistically significant mAP50 values of 0.862 and 0.709 on the dINbreast and dCBIS datasets, respectively. These results confirm that a moderate threshold preserves clinically relevant true-positive candidates, which is particularly important for screening-oriented settings where missed lesions are costly. In the segmentation stage, the proposed framework achieved mean DICE scores of 0.721 and 0.700 on the test sets of the same datasets, demonstrating consistent overlap with expert annotations. Compared with state-of-the-art approaches that commonly assume lesion-centered ROIs or rely on heavier backbones, the proposed pipeline addresses a more realistic scenario by performing automatic detection followed by segmentation while maintaining substantially lower computational requirements. This balance between performance and efficiency makes the MassSeg-Framework a promising tool for scalable mammography analysis, particularly in resource-constrained environments or high-throughput screening workflows that require rapid processing.

The segmentation of head and neck cancer (HNC) tumors is a critical step in radiotherapy treatment planning. The development of automatic segmentation algorithms has the potential to streamline the radiation oncology process. In this work, we develop an ensemble of LinkNet networks for HNC tumor segmentation as part of the HNTS-MRG 2024 Grand Challenge. A single LinkNet network, pretrained on the Imagenet dataset, was trained for 200 epochs on the HNC dataset provided by the challenge. Eight good performing weights from the internal validation set were selected to create an ensemble of 2D networks. Specifically, each selected weight was used to generate a LinkNet architecture, resulting in eight networks whose predictions were averaged to produce the final predicted segmentation. Our experiments demonstrate that the ensemble network performs better than each individual architecture, leveraging the benefits of ensemble learning without the computational cost of training each network from scratch. In the challenge’s test set, the LinkNet Ensemble (team ECU) achieved an aggregated Dice score of 64.60% and 49.53% for metastatic lymph nodes and primary gross tumor segmentation, respectively, and a mean score of 57.06%.

Content analysis of political manifestos is necessary to understand the policies and proposed actions of a party. However, manually labeling political texts is time-consuming and labor-intensive. Transformer networks have become essential tools for automating this task. Nevertheless, these models require extensive datasets to achieve good performance. This can be a limitation in manifesto classification, where the availability of publicly labeled datasets can be scarce. To address this challenge, in this work, we developed a Transformer network for the classification of manifestos using a cross-domain training strategy. Using the database of the Comparative Manifesto Project, we implemented a fractional factorial experimental design to determine which Spanish-written manifestos form the best training set for Ecuadorian manifesto labeling. Furthermore, we statistically analyzed which Transformer architecture and preprocessing operations improve the model accuracy. The results indicate that creating a training set with manifestos from Spain and Uruguay, along with implementing stemming and lemmatization preprocessing operations, produces the highest classification accuracy. In addition, we found that the DistilBERT and RoBERTa transformer networks perform statistically similarly and consistently well in manifesto classification. Using the cross-context training strategy, DistilBERT and RoBERTa achieve 60.05% and 57.64% accuracy, respectively, in the classification of the Ecuadorian manifesto. Finally, we investigated the effect of the composition of the training set on performance. The experiments demonstrate that training DistilBERT solely with Ecuadorian manifestos achieves the highest accuracy and F1-score. Furthermore, in the absence of the Ecuadorian dataset, competitive performance is achieved by training the model with datasets from Spain and Uruguay.

The utilization of Artificial Intelligence (AI) and Generative AI (GenAI) in higher education has increased importantly in the last years. Studies show that AI holds promise in enhancing the learning experiences for both students and educators, offering personalized learning and assessment opportunities. This study conducts a systematic review on the application of AI within Latin American higher education. To this end, we synthesized 25 papers published between 2021 and 2023, encompassing AI's utilization in Mexico, Colombia, Ecuador, Brazil, Peru, Chile, Argentina, and Bolivia. The analysis addresses three key inquiries: the prevalent applications of AI in Latin American education, the perceptions of AI and GenAI models among educators and students, and the particular challenges encountered by Latin American institutions in AI implementation. This systematic review offers an updated understanding of AI's role in Latin American higher education, with a particular emphasis on the latest AI technologies.