Giovanna Calabrese dos Santos
Giovanna Calabrese dos Santos has a total of 52 co-authors. Here are the top co-authors by number of shared publications:
- Daniela Giraldo-Roldán
Oral Diagnosis Department, Piracicaba Dental School, University of Campinas (UNICAMP), Av. Limeira, 901, Piracicaba, São Paulo, 13414-903, Brazil
6 shared publications - Matheus Cardoso Moraes
Institute of Science and Technology (ICT‐UNIFESP) Federal University of São Paulo São José dos Campos São Paulo Brazil
6 shared publications - Pablo Agustin Vargas
Department of Oral Diagnosis, Piracicaba Dental School, University of Campinas, Piracicaba, Brazil
6 shared publications - Alan Roger Santos-Silva
Department of Oral Diagnosis Piracicaba Dental School, University of Campinas (UNICAMP) Piracicaba São Paulo Brazil
5 shared publications - Anna Luíza Damaceno Araújo
Hospital Israelita Albert Einstein, São Paulo 05652-900, SP, Brazil
5 shared publications - Luiz Paulo Kowalski
Head and Neck Surgery Department University of São Paulo Medical School (FMUSP) São Paulo Brazil
5 shared publications
Latest Publications
- Open Access
- 26 June 2026
- Published by Springer Nature in Head and Neck Pathology
- Vol. 20 (1)
Pathology-informed Generative Adversarial Network Augmentation Improves Classification of Peripheral Nerve Sheath Tumors by Modeling Morphological Variability: A Pilot Investigation
Background: Peripheral nerve sheath tumors (PNSTs) of the head and neck (H&N) show histopathological overlap. Although convolutional neural networks (CNNs) have demonstrated feasibility in soft tissue tumor classification, limited intra-class variability related to perineurioma remains a critical constraint for rare tumor subtypes. Methods: This retrospective diagnostic accuracy study with internal validation included 30 patients diagnosed with PNSTs. Whole-slide images were digitized at 20× magnification and partitioned using a strict patient-wise split. Synthetic perineurioma patches were generated using a modified Pix2Pix-based Generative Adversarial Network (GAN) incorporating a bottleneck architecture and self-attention modules. Two morphology-driven augmentation strategies were evaluated: (1) intra-phenotypic expansion by cross-patient patch pairing within the sclerosing subtype and (2) inter-phenotypic interpolation by cross-phenotype patch pairing between sclerosing and intraneural variants. EfficientNetV2-B0 pre-trained on ImageNet was trained under three configurations: original dataset only, original + Experiment A synthetic patches, and original + Experiment B synthetic patches. All performance metrics were computed exclusively on an independent yet internal test set composed of original images. Results: GAN-based augmentation improved global classification performance compared with the baseline model trained on original images only (accuracy 0.733). Intra-phenotypic expansion increased accuracy to 0.767 and achieved the highest balanced accuracy (0.750) and macro-F1 (0.740). Inter-phenotypic interpolation yielded the highest overall accuracy and competitive multiclass agreement metrics. Perineurioma recall improved from 0.34 (baseline) to 0.51 with intra-phenotypic augmentation and 0.47 with inter-phenotypic interpolation, while specificity remained ≥ 0.999 across all strategies. Conclusions: Structured, pathology-informed GAN augmentation improved CNN classification of PNSTs, particularly for the morphologically heterogeneous perineurioma class. Intra-phenotypic expansion primarily improved rare-class sensitivity, whereas inter-phenotypic interpolation improved multiclass agreement and global robustness. These findings support morphology-driven synthetic enrichment as a clinically meaningful strategy to improve AI performance in underrepresented tumor entities and potentially support diagnostic decision-making in digital pathology environments.
- 26 June 2026
- Published by Springer Nature in Head and Neck Pathology
- Vol. 20 (1)
Open AccessPathology-informed Generative Adversarial Network Augmentation Improves Classification of Peripheral Nerve Sheath Tumors by Modeling Morphological Variability: A Pilot Investigation
Background: Peripheral nerve sheath tumors (PNSTs) of the head and neck (H&N) show histopathological overlap. Although convolutional neural networks (CNNs) have demonstrated feasibility in soft tissue tumor classification, limited intra-class variability related to perineurioma remains a critical constraint for rare tumor subtypes. Methods: This retrospective diagnostic accuracy study with internal validation included 30 patients diagnosed with PNSTs. Whole-slide images were digitized at 20× magnification and partitioned using a strict patient-wise split. Synthetic perineurioma patches were generated using a modified Pix2Pix-based Generative Adversarial Network (GAN) incorporating a bottleneck architecture and self-attention modules. Two morphology-driven augmentation strategies were evaluated: (1) intra-phenotypic expansion by cross-patient patch pairing within the sclerosing subtype and (2) inter-phenotypic interpolation by cross-phenotype patch pairing between sclerosing and intraneural variants. EfficientNetV2-B0 pre-trained on ImageNet was trained under three configurations: original dataset only, original + Experiment A synthetic patches, and original + Experiment B synthetic patches. All performance metrics were computed exclusively on an independent yet internal test set composed of original images. Results: GAN-based augmentation improved global classification performance compared with the baseline model trained on original images only (accuracy 0.733). Intra-phenotypic expansion increased accuracy to 0.767 and achieved the highest balanced accuracy (0.750) and macro-F1 (0.740). Inter-phenotypic interpolation yielded the highest overall accuracy and competitive multiclass agreement metrics. Perineurioma recall improved from 0.34 (baseline) to 0.51 with intra-phenotypic augmentation and 0.47 with inter-phenotypic interpolation, while specificity remained ≥ 0.999 across all strategies. Conclusions: Structured, pathology-informed GAN augmentation improved CNN classification of PNSTs, particularly for the morphologically heterogeneous perineurioma class. Intra-phenotypic expansion primarily improved rare-class sensitivity, whereas inter-phenotypic interpolation improved multiclass agreement and global robustness. These findings support morphology-driven synthetic enrichment as a clinically meaningful strategy to improve AI performance in underrepresented tumor entities and potentially support diagnostic decision-making in digital pathology environments.
- Open Access
- 19 February 2026
- Published by Springer Nature in Head and Neck Pathology
- Vol. 20 (1) , 24
Impact of Transfer Learning on Convolutional Neural Networks for Odontogenic Tumor Diagnosis
Objective: This study aimed to evaluate the coherence between data heterogeneity and model complexity by comparing seven convolutional neural network (CNN) architectures—trained with and without ImageNet pretraining—in a multiclass framework for the histopathological classification of three odontogenic tumors: adenomatoid odontogenic tumor, ameloblastoma, and ameloblastic carcinoma. The goal was to investigate how transfer learning influences performance and diagnostic reliability in a clinically relevant context characterized by overlapping histological patterns. Methods: An international, multicenter cross-sectional dataset of 64 hematoxylin- and eosin-stained whole slide images was analyzed, including adenomatoid odontogenic tumor ( n = 16), ameloblastoma ( n = 27), and ameloblastic carcinoma ( n = 21). Seven CNN models (DenseNet121, EfficientNetV2B0, InceptionV3, MobileNet, ResNet50, VGG16, and Xception) were trained and tested on 455,107 patches (224 × 224 pixels). Performance was assessed using accuracy, balanced accuracy, sensitivity, specificity, F1-score, and AUC. Results: Without ImageNet pretraining, DenseNet121 achieved the highest performance (accuracy = 0.73, balanced accuracy = 0.74, AUC = 0.78, specificity = 0.84, sensitivity = 0.65), followed by EfficientNetV2B0 (accuracy = 0.67, balanced accuracy = 0.68, sensitivity = 0.54). When ImageNet pretraining was applied, performance improved across all architectures. EfficientNetV2B0 reached the best overall results (accuracy = 0.79, balanced accuracy = 0.81, AUC = 0.91, specificity = 0.88, sensitivity = 0.74), while DenseNet121 maintained consistent performance (accuracy = 0.72, balanced accuracy = 0.74, AUC = 0.85, specificity = 0.84, sensitivity = 0.64). Conclusion: Transfer learning with ImageNet weights enhanced the performance of most CNNs, with EfficientNetV2B0 showing the greatest responsiveness to pretraining and DenseNet121 demonstrating intrinsic robustness to initialization. These results highlight the potential of CNN-based frameworks to support the differential diagnosis of odontogenic tumors—an inherently challenging task due to morphological overlap—while establishing reproducible methodological baselines that contribute to the global development of explainable, ensemble-based, and clinically reliable AI systems in oral pathology.
- 19 February 2026
- Published by Springer Nature in Head and Neck Pathology
- Vol. 20 (1) , 24
Open AccessImpact of Transfer Learning on Convolutional Neural Networks for Odontogenic Tumor Diagnosis
Objective: This study aimed to evaluate the coherence between data heterogeneity and model complexity by comparing seven convolutional neural network (CNN) architectures—trained with and without ImageNet pretraining—in a multiclass framework for the histopathological classification of three odontogenic tumors: adenomatoid odontogenic tumor, ameloblastoma, and ameloblastic carcinoma. The goal was to investigate how transfer learning influences performance and diagnostic reliability in a clinically relevant context characterized by overlapping histological patterns. Methods: An international, multicenter cross-sectional dataset of 64 hematoxylin- and eosin-stained whole slide images was analyzed, including adenomatoid odontogenic tumor ( n = 16), ameloblastoma ( n = 27), and ameloblastic carcinoma ( n = 21). Seven CNN models (DenseNet121, EfficientNetV2B0, InceptionV3, MobileNet, ResNet50, VGG16, and Xception) were trained and tested on 455,107 patches (224 × 224 pixels). Performance was assessed using accuracy, balanced accuracy, sensitivity, specificity, F1-score, and AUC. Results: Without ImageNet pretraining, DenseNet121 achieved the highest performance (accuracy = 0.73, balanced accuracy = 0.74, AUC = 0.78, specificity = 0.84, sensitivity = 0.65), followed by EfficientNetV2B0 (accuracy = 0.67, balanced accuracy = 0.68, sensitivity = 0.54). When ImageNet pretraining was applied, performance improved across all architectures. EfficientNetV2B0 reached the best overall results (accuracy = 0.79, balanced accuracy = 0.81, AUC = 0.91, specificity = 0.88, sensitivity = 0.74), while DenseNet121 maintained consistent performance (accuracy = 0.72, balanced accuracy = 0.74, AUC = 0.85, specificity = 0.84, sensitivity = 0.64). Conclusion: Transfer learning with ImageNet weights enhanced the performance of most CNNs, with EfficientNetV2B0 showing the greatest responsiveness to pretraining and DenseNet121 demonstrating intrinsic robustness to initialization. These results highlight the potential of CNN-based frameworks to support the differential diagnosis of odontogenic tumors—an inherently challenging task due to morphological overlap—while establishing reproducible methodological baselines that contribute to the global development of explainable, ensemble-based, and clinically reliable AI systems in oral pathology.
- 10 November 2025
- Published by Springer Nature in Head and Neck Pathology
- Vol. 19 (1) , 1-14
Evaluation of Deep Learning Convolutional Neural Networks for Classification of Carcinoma Ex Pleomorphic Adenoma and Pleomorphic Adenoma in Whole-Slide Images
Objective The study aimed to compare multiple convolutional neural networks architectures for their classification performance in distinguishing salivary gland tumors, pleomorphic adenoma and carcinoma ex pleomorphic adenoma, using whole-slide images. Methods A cross-sectional study using 107 hematoxylin and eosin stained whole-slide images from 83 patients diagnosed with pleomorphic adenoma (n = 41) and carcinoma ex pleomorphic adenoma (n = 42) was conducted. Eight convolutional neural networks models (ResNet50, InceptionV3, VGG16, Xception, MobileNet, DenseNet121, EfficientNetB0, and EfficientNetV2B0) were applied, trained, and evaluated. A total of 955,583 patches (224 × 224 pixels) were generated and not-randomly divided into training (80%), validation (10%), and testing (10%) subsets. Performance and generalization were assessed through analysis of training and validation accuracy and loss curves. Testing phase evaluation included multiple metrics—such as precision, sensitivity, specificity, and others. Results ResNet50 achieved the highest performance in 7 out of 9 metrics. DenseNet121 also delivered strong results, surpassing ResNet50 in specificity (94% vs. 93%) while matching its balanced accuracy (93%), precision (98%), and area under the receiver operating characteristic curve (0.97). Both exhibited comparable performance in loss (0.63 vs. 0.65), precision (98% vs. 98%), sensitivity (94% vs. 92%), and F1 score (0.96 vs. 0.95), demonstrating near-equivalent diagnostic capability. Conclusion This study demonstrates strong potential of convolutional neural networks for classifying salivary gland tumors, with ResNet50 and DenseNet121 showing notable performance. Future work should focus on expanding datasets, improving generalization, exploring ensemble methods, and incorporating interpretability to enhance clinical relevance with clinical and radiographic data.
- 10 November 2025
- Published by Springer Nature in Head and Neck Pathology
- Vol. 19 (1) , 1-14
Evaluation of Deep Learning Convolutional Neural Networks for Classification of Carcinoma Ex Pleomorphic Adenoma and Pleomorphic Adenoma in Whole-Slide Images
Objective The study aimed to compare multiple convolutional neural networks architectures for their classification performance in distinguishing salivary gland tumors, pleomorphic adenoma and carcinoma ex pleomorphic adenoma, using whole-slide images. Methods A cross-sectional study using 107 hematoxylin and eosin stained whole-slide images from 83 patients diagnosed with pleomorphic adenoma (n = 41) and carcinoma ex pleomorphic adenoma (n = 42) was conducted. Eight convolutional neural networks models (ResNet50, InceptionV3, VGG16, Xception, MobileNet, DenseNet121, EfficientNetB0, and EfficientNetV2B0) were applied, trained, and evaluated. A total of 955,583 patches (224 × 224 pixels) were generated and not-randomly divided into training (80%), validation (10%), and testing (10%) subsets. Performance and generalization were assessed through analysis of training and validation accuracy and loss curves. Testing phase evaluation included multiple metrics—such as precision, sensitivity, specificity, and others. Results ResNet50 achieved the highest performance in 7 out of 9 metrics. DenseNet121 also delivered strong results, surpassing ResNet50 in specificity (94% vs. 93%) while matching its balanced accuracy (93%), precision (98%), and area under the receiver operating characteristic curve (0.97). Both exhibited comparable performance in loss (0.63 vs. 0.65), precision (98% vs. 98%), sensitivity (94% vs. 92%), and F1 score (0.96 vs. 0.95), demonstrating near-equivalent diagnostic capability. Conclusion This study demonstrates strong potential of convolutional neural networks for classifying salivary gland tumors, with ResNet50 and DenseNet121 showing notable performance. Future work should focus on expanding datasets, improving generalization, exploring ensemble methods, and incorporating interpretability to enhance clinical relevance with clinical and radiographic data.
- 4 October 2025
- Published by Springer Nature
- p. 548-556
Assessment of the Influence of Batch Size on ResNet-50 Processing Applied to Histopathological Datasets
- 4 October 2025
- Published by Springer Nature
- p. 548-556
Assessment of the Influence of Batch Size on ResNet-50 Processing Applied to Histopathological Datasets
- 28 October 2024
- Published by Springer Nature in Head and Neck Pathology
- Vol. 18 (1) , 1-9
Deep Convolutional Neural Network for Accurate Classification of Myofibroblastic Lesions on Patch-Based Images
Objective This study aimed to implement and evaluate a Deep Convolutional Neural Network for classifying myofibroblastic lesions into benign and malignant categories based on patch-based images. Methods A Residual Neural Network (ResNet50) model, pre-trained with weights from ImageNet, was fine-tuned to classify a cohort of 20 patients (11 benign and 9 malignant cases). Following annotation of tumor regions, the whole-slide images (WSIs) were fragmented into smaller patches (224 × 224 pixels). These patches were non-randomly divided into training (308,843 patches), validation (43,268 patches), and test (42,061 patches) subsets, maintaining a 78:11:11 ratio. The CNN training was caried out for 75 epochs utilizing a batch size of 4, the Adam optimizer, and a learning rate of 0.00001. Results ResNet50 achieved an accuracy of 98.97%, precision of 99.91%, sensitivity of 97.98%, specificity of 99.91%, F1 score of 98.94%, and AUC of 0.99. Conclusions The ResNet50 model developed exhibited high accuracy during training and robust generalization capabilities in unseen data, indicating nearly flawless performance in distinguishing between benign and malignant myofibroblastic tumors, despite the small sample size. The excellent performance of the AI model in separating such histologically similar classes could be attributed to its ability to identify hidden discriminative features, as well as to use a wide range of features and benefit from proper data preprocessing.
- 28 October 2024
- Published by Springer Nature in Head and Neck Pathology
- Vol. 18 (1) , 1-9
Deep Convolutional Neural Network for Accurate Classification of Myofibroblastic Lesions on Patch-Based Images
Objective This study aimed to implement and evaluate a Deep Convolutional Neural Network for classifying myofibroblastic lesions into benign and malignant categories based on patch-based images. Methods A Residual Neural Network (ResNet50) model, pre-trained with weights from ImageNet, was fine-tuned to classify a cohort of 20 patients (11 benign and 9 malignant cases). Following annotation of tumor regions, the whole-slide images (WSIs) were fragmented into smaller patches (224 × 224 pixels). These patches were non-randomly divided into training (308,843 patches), validation (43,268 patches), and test (42,061 patches) subsets, maintaining a 78:11:11 ratio. The CNN training was caried out for 75 epochs utilizing a batch size of 4, the Adam optimizer, and a learning rate of 0.00001. Results ResNet50 achieved an accuracy of 98.97%, precision of 99.91%, sensitivity of 97.98%, specificity of 99.91%, F1 score of 98.94%, and AUC of 0.99. Conclusions The ResNet50 model developed exhibited high accuracy during training and robust generalization capabilities in unseen data, indicating nearly flawless performance in distinguishing between benign and malignant myofibroblastic tumors, despite the small sample size. The excellent performance of the AI model in separating such histologically similar classes could be attributed to its ability to identify hidden discriminative features, as well as to use a wide range of features and benefit from proper data preprocessing.
- 27 October 2024
- Published by Wiley in Head & Neck
- Vol. 47 (3) , 832-838
Development and Evaluation of a Convolutional Neural Network for Microscopic Diagnosis Between Pleomorphic Adenoma and Carcinoma Ex‐Pleomorphic Adenoma
Aims: To develop a model capable of distinguishing carcinoma ex‐pleomorphic adenoma from pleomorphic adenoma using a convolutional neural network architecture.Methods and Results: A cohort of 83 Brazilian patients, divided into carcinoma ex‐pleomorphic adenoma (n = 42) and pleomorphic adenoma (n = 41), was used for training a convolutional neural network. The whole‐slide images were annotated and fragmented into 743 869 (carcinoma ex‐pleomorphic adenomas) and 211 714 (pleomorphic adenomas) patches, measuring 224 × 224 pixels. Training (80%), validation (10%), and test (10%) subsets were established. The Residual Neural Network (ResNet)‐50 was chosen for its recognition and classification capabilities. The training and validation graphs, and parameters derived from the confusion matrix, were evaluated. The loss curve recorded 0.63, and the accuracy reached 0.93. Evaluated parameters included specificity (0.88), sensitivity (0.94), precision (0.96), F1 score (0.95), and area under the curve (0.97).Conclusions: The study underscores the potential of ResNet‐50 in the microscopic diagnosis of carcinoma ex‐pleomorphic adenoma. The developed model demonstrated strong learning potential, but exhibited partial limitations in generalization, as indicated by the validation curve. In summary, the study established a promising baseline despite limitations in model generalization. This indicates the need to refine methodologies, investigate new models, incorporate larger datasets, and encourage inter‐institutional collaboration for comprehensive studies in salivary gland tumors.
- 27 October 2024
- Published by Wiley in Head & Neck
- Vol. 47 (3) , 832-838
Development and Evaluation of a Convolutional Neural Network for Microscopic Diagnosis Between Pleomorphic Adenoma and Carcinoma Ex‐Pleomorphic Adenoma
Aims: To develop a model capable of distinguishing carcinoma ex‐pleomorphic adenoma from pleomorphic adenoma using a convolutional neural network architecture.Methods and Results: A cohort of 83 Brazilian patients, divided into carcinoma ex‐pleomorphic adenoma (n = 42) and pleomorphic adenoma (n = 41), was used for training a convolutional neural network. The whole‐slide images were annotated and fragmented into 743 869 (carcinoma ex‐pleomorphic adenomas) and 211 714 (pleomorphic adenomas) patches, measuring 224 × 224 pixels. Training (80%), validation (10%), and test (10%) subsets were established. The Residual Neural Network (ResNet)‐50 was chosen for its recognition and classification capabilities. The training and validation graphs, and parameters derived from the confusion matrix, were evaluated. The loss curve recorded 0.63, and the accuracy reached 0.93. Evaluated parameters included specificity (0.88), sensitivity (0.94), precision (0.96), F1 score (0.95), and area under the curve (0.97).Conclusions: The study underscores the potential of ResNet‐50 in the microscopic diagnosis of carcinoma ex‐pleomorphic adenoma. The developed model demonstrated strong learning potential, but exhibited partial limitations in generalization, as indicated by the validation curve. In summary, the study established a promising baseline despite limitations in model generalization. This indicates the need to refine methodologies, investigate new models, incorporate larger datasets, and encourage inter‐institutional collaboration for comprehensive studies in salivary gland tumors.
- 4 June 2024
- Published by Wiley in Journal of Oral Pathology & Medicine
- Vol. 53 (7) , 444-450
Feasibility study of ResNet‐50 in the distinction of intraoral neural tumors using histopathological images
Background: Neural tumors are difficult to distinguish based solely on cellularity and often require immunohistochemical staining to aid in identifying the cell lineage. This article investigates the potential of a Convolutional Neural Network for the histopathological classification of the three most prevalent benign neural tumor types: neurofibroma, perineurioma, and schwannoma.Methods: A model was developed, trained, and evaluated for classification using the ResNet‐50 architecture, with a database of 30 whole‐slide images stained in hematoxylin and eosin (106, 782 patches were generated from and divided among the training, validation, and testing subsets, with strategies to avoid data leakage).Results: The model achieved an accuracy of 70% (64% normalized), and showed satisfactory results for differentiating two of the three classes, reaching approximately 97% and 77% as true positives for neurofibroma and schwannoma classes, respectively, and only 7% for perineurioma class. The AUROC curves for neurofibroma and schwannoma classes was 0.83%, and 0.74% for perineurioma. However, the specificity rate for the perineurioma class was greater (83%) than in the other two classes (neurofibroma with 61%, and schwannoma with 60%).Conclusion: This investigation demonstrated significant potential for proficient performance with a limitation regarding the perineurioma class (the limited feature variability observed contributed to a lower performance).
- 4 June 2024
- Published by Wiley in Journal of Oral Pathology & Medicine
- Vol. 53 (7) , 444-450
Feasibility study of ResNet‐50 in the distinction of intraoral neural tumors using histopathological images
Background: Neural tumors are difficult to distinguish based solely on cellularity and often require immunohistochemical staining to aid in identifying the cell lineage. This article investigates the potential of a Convolutional Neural Network for the histopathological classification of the three most prevalent benign neural tumor types: neurofibroma, perineurioma, and schwannoma.Methods: A model was developed, trained, and evaluated for classification using the ResNet‐50 architecture, with a database of 30 whole‐slide images stained in hematoxylin and eosin (106, 782 patches were generated from and divided among the training, validation, and testing subsets, with strategies to avoid data leakage).Results: The model achieved an accuracy of 70% (64% normalized), and showed satisfactory results for differentiating two of the three classes, reaching approximately 97% and 77% as true positives for neurofibroma and schwannoma classes, respectively, and only 7% for perineurioma class. The AUROC curves for neurofibroma and schwannoma classes was 0.83%, and 0.74% for perineurioma. However, the specificity rate for the perineurioma class was greater (83%) than in the other two classes (neurofibroma with 61%, and schwannoma with 60%).Conclusion: This investigation demonstrated significant potential for proficient performance with a limitation regarding the perineurioma class (the limited feature variability observed contributed to a lower performance).
- 4 January 2024
- Published by Springer Nature
- p. 78-85
Evidence of Cortical Activation Alterations of Paraplegics with Pelvic Nerves Stimulation on Resting State: A Case Series
- 4 January 2024
- Published by Springer Nature
- p. 78-85
Evidence of Cortical Activation Alterations of Paraplegics with Pelvic Nerves Stimulation on Resting State: A Case Series