Cristina Alfaro Vergara

Dr. Cristina Alfaro Vergara


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 August 2025 in Journal of Imaging Informatics in Medicine
Verified
Journal Article
Open access
CVCristina Alfaro Vergara
Cristina Alfaro Vergara
NCNicolás Araya Caro
Nicolás Araya Caro
DMDomingo Mery
Domingo Mery
CPClaudia Prieto
Claudia Prieto
The scarcity of publicly available digital breast tomosynthesis (DBT) datasets significantly limits the development of robust deep learning (DL) models for breast tumor segmentation. In this exploratory proof-of-concept study, we assess the viability of in silico-generated DBT data as a training source for tumor segmentation. A dataset of 230 two-dimensional (2D) regions of interest (ROIs) derived from FDA-cleared software and encompassing a spectrum of breast densities and tumor complexities, was used to train 13 DL models, including U-Net, FCN, DeepLabv3, and DeepLabv3 + architectures. Each model was trained either from scratch or fine-tuned using COCO-pretrained weights (ResNet50/101 backbones). Performance was evaluated using F1-score, intersection over union (IoU), precision, and recall. Among all models, U-Net trained from scratch and DeepLabv3 + fine-tuned with ResNet50 achieved the highest and most consistent results (F1-scores of 82.52% and 84.98%, and per-image IoUs of 78.49% and 83.77%, respectively). No statistically significant differences were found using the Wilcoxon signed-rank test and post hoc Bonferroni correction ( α > 0.0042). To evaluate generalization across domains, the baseline U-Net model was retrained from scratch on a hybrid dataset combining in silico and real-world DBT ROIs, yielding promising results (F1-score of 79%). Despite the domain shift, these findings support the utility of in silico DBT as a complementary resource for training and benchmarking DL models, particularly in data-limited environments. This study provides foundational experimental evidence for integrating computationally generated in silico data into AI-based DBT tumor segmentation research workflows.

20 September 2026 in Diagnostics
Verified
Journal Article
Open access
PDF file
CVCristina Alfaro Vergara
Cristina Alfaro Vergara
GGGabriel Guerra
Gabriel Guerra
CPClaudia Prieto
Claudia Prieto
DMDomingo Mery
Domingo Mery
Background/Objectives: Digital breast tomosynthesis (DBT) provides quasi-three-dimensional breast imaging, but volumetric segmentation model development is constrained by limited expert-annotated clinical data. This proof-of-concept study evaluated the feasibility of synthetic-to-clinical breast tumor segmentation using synthetic model development, limited clinical fine-tuning, and heterogeneous ensemble learning, and aimed to establish a reproducible low-resource framework for volumetric DBT segmentation. Methods: Three architectures—3D U-Net, nnU-Net, and Attention U-Net—were evaluated using publicly available synthetic and clinical DBT datasets. Models initialized from the Mixed-size synthetic configuration were fine-tuned using 10 clinical development cases with five-fold cross-validation; 10 additional clinical cases were reserved for independent testing. Ensemble weights and threshold were evaluated from development-set out-of-fold predictions. Performance was assessed using Dice, Intersection-over-Union (IoU), precision, and recall, with paired non-parametric comparisons on the clinical test cohort. Results: nnU-Net achieved the highest mean Dice on the Large Tumor (0.864) and Mixed-size (0.841) synthetic test sets, while performance was lower in the Small Tumor configuration (0.560). On the independent clinical cohort, the ensemble achieved the highest mean Dice (0.518) and recall (0.630), compared with nnU-Net (Dice = 0.482), Attention U-Net (0.372), and 3D U-Net (0.367). After Holm correction, ensemble Dice was significantly higher than Attention U-Net and 3D U-Net, but not nnU-Net. Conclusions: Synthetic DBT data provided a useful source-domain foundation for volumetric segmentation under constrained annotation conditions, although a synthetic-to-clinical gap remained. Heterogeneous ensemble integration achieved the highest mean clinical Dice without demonstrating superiority over fine-tuned nnU-Net. Larger and more diverse clinical cohorts are required to establish generalizability.

5 December 2024 in medRxiv
Verified
Preprint
Open access
PDF file
SMStephen James Morrell
Stephen James Morrell
MHMichael Huetel
Michael Huetel
OLOeslle de Lucena
Oeslle de Lucena
CVCristina Alfaro Vergara
Cristina Alfaro Vergara
GZGeorgiana Zamfir
Georgiana Zamfir
ECEmily F Conant
Emily F Conant
Structured Purpose: To compare Digital Breast Tomosynthesis (DBT) tissue matching errors with and without artificial intelligence (AI) assistance to typical screen-detected breast tumor sizes, evaluating whether AI ameliorates lesion mislocalization beyond tumor boundaries, especially for nonexpert radiologists. The technology category is deep learning. Materials and Methods: This multicenter retrospective feasibility study conducted in April 2022 – July 2023 included 12 radiologists (mean age, 42 years ± 8) interpreting 94 lesion regions of interest in 30 women. Readers performed annotations with and without AI assistance after a minimum four-week washout period. The root mean square errors (RMSE) and maximum distance errors (MDE) were measured relative to consensus references. Stratifications included radiologist expertise ( ≥ 5 vs < 5 years), lesion abnormal-ity, and AI warnings. The Wilcoxon signed-rank test was used to assess statistical significance. Results: Across all abnormal lesions, mean RMSE was 32% higher without AI (11.70 mm vs 8.88 mm , p = .049), and mean maximum distance errors were 37.5% higher (20.68 mm vs 15.08 mm , p = .036). Non-expert radiologists showed the largest benefit: for abnormal lesions without AI warnings, RMSE was 61.9% higher without AI (12.20 mm vs 7.57 mm , p = .010) and maximum distance error was 67.5% higher (15.76 mm vs 9.47 mm , p = .028). These reductions are clinically relevant given typical screen-detected breast tumor sizes (median, 13 mm [IQR: 9–20]). Conclusion: AI-assisted tissue matching significantly reduced DBT localization errors, particularly for non-experts handling challenging cases. By keeping errors below typical tumor dimensions, AI may improve diagnostic precision and reduce risks of missed or mischaracterized lesions. Summary Statement: Dynamic artificial intelligence assisted tissue matching in digital breast tomosynthesis improves localization accuracy for non-expert radiologists, with errors in abnormal cases significantly larger (67.5%, p ≤ 0.05) without assistance. Key Points: In this multicenter retrospective study of 94 Regions of Interest (ROI) analyzed by 12 radiologists across 5 hospitals, manual tissue matching was found to have errors 32% higher in abnormal cases than AI-assisted tissue matching ( p < 0.05). For non-expert radiologists interpreting abnormal cases (excluding cases with AI warnings) without AI assistance, the root mean squared distance errors (RMSE) were found to be 61.9% higher (12.20 mm vs 7.57 mm , p < 0.01) and the maximum distance errors (MDE) was 67.6% higher (15.76 mm vs 9.47 mm , p < 0.05) than when using AI assistance. For challenging cases, many non-expert readers’ MDE without AI assistance (75th percentile: 20.21 mm ) exceeded the largest tumor dimensions (75th percentile: 20 mm ), while AI-assisted errors (75th percentile: 11.94 mm , p < 0.05) remained within median tumor sizes (12 mm ), potentially preventing correlation with non-lesion tissue.