Roberta Galdiero
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Roberta Galdiero has a total of 55 co-authors. Here are the top co-authors by number of shared publications:
- Roberta Fusco
Division of Radiology, Istituto Nazionale Tumori IRCCS Fondazione Pascale—IRCCS di Napoli, 80131 Naples, Italy
5 shared publications - Vincenza Granata
Radiology Division, Istituto Nazionale Tumori-IRCCS-Fondazione G. Pascale, 80131 Naples, Italy
5 shared publications - Antonella Petrillo
Dipartimento di Ingegneria, Isola C4 Centro Direzionale, Università degli Studi di Napoli “Parthenope”, 80143 Napoli, Italy
3 shared publications - Roberto Grassi
University of Campania “Luigi Vanvitelli
3 shared publications - Carmine Picone
Istituto Nazionale Tumori di Napoli, IRCCS “G. Pascale”, Via Mariano Semmola, 80131 Naples, Italy
2 shared publications - Francesco Izzo
Istituto Nazionale Tumori di Napoli, IRCCS “G. Pascale”, Via Mariano Semmola, 80131 Naples, Italy
2 shared publications
Latest Publications
- Open Access
- 17 April 2026
- Published by MDPI AG in Bioengineering
- Vol. 13 (4) , 475
Clinically Robust Deep Learning for Contrast-Enhanced Mammography: Multicenter Evaluation Across Convolutional Neural Network Architectures
Background: This study investigates the impact of anatomically constrained preprocessing and deep learning architecture selection on benign versus malignant breast lesion classification in contrast-enhanced mammography (CEM), with the goal of improving robustness and clinical reliability across heterogeneous data sources. Methods: In this retrospective multicenter study, CEM images from 300 patients (314 lesions) were combined with 1003 publicly available CEM images, yielding a total of 1120 breast cases. Automatic breast segmentation was performed using the LIBRA framework to generate breast-mask images. Eleven deep learning models, including classical convolutional neural networks, attention-based networks, hybrid convolutional neural networks (CNNs), Transformer architectures, and mammography-specific models, were trained and evaluated using both original DICOM images and breast-mask inputs. Performance was assessed using accuracy, balanced accuracy, sensitivity, specificity, AUROC, and AUPRC on cross-validation and independent test sets. Hyperparameter optimization was conducted for the best-performing architecture. Results: Models trained on breast-mask images consistently outperformed those trained on original DICOM images across all architectures and metrics, with AUROC improvements ranging from +0.06 to +0.21. Among all models, ResNet50 trained on breast-mask images achieved the best performance (AUROC = 0.931; AUPRC = 0.933; balanced accuracy = 0.834), further improved after optimization (balanced accuracy = 0.886; sensitivity = 0.842; specificity = 0.930). Classical CNN architectures demonstrated performance comparable to or exceeding that of more complex hybrid CNN–Transformer models when anatomically focused preprocessing and rigorous optimization were applied. Conclusions: Anatomically constrained preprocessing through breast-mask segmentation substantially enhances deep learning performance and stability in CEM-based breast lesion classification. These findings indicate that input representation quality and training optimization are critical determinants of clinically relevant performance, often outweighing architectural complexity, and may support more reliable AI-assisted decision support in CEM workflows.
- 17 April 2026
- Published by MDPI AG in Bioengineering
- Vol. 13 (4) , 475
Open AccessClinically Robust Deep Learning for Contrast-Enhanced Mammography: Multicenter Evaluation Across Convolutional Neural Network Architectures
Background: This study investigates the impact of anatomically constrained preprocessing and deep learning architecture selection on benign versus malignant breast lesion classification in contrast-enhanced mammography (CEM), with the goal of improving robustness and clinical reliability across heterogeneous data sources. Methods: In this retrospective multicenter study, CEM images from 300 patients (314 lesions) were combined with 1003 publicly available CEM images, yielding a total of 1120 breast cases. Automatic breast segmentation was performed using the LIBRA framework to generate breast-mask images. Eleven deep learning models, including classical convolutional neural networks, attention-based networks, hybrid convolutional neural networks (CNNs), Transformer architectures, and mammography-specific models, were trained and evaluated using both original DICOM images and breast-mask inputs. Performance was assessed using accuracy, balanced accuracy, sensitivity, specificity, AUROC, and AUPRC on cross-validation and independent test sets. Hyperparameter optimization was conducted for the best-performing architecture. Results: Models trained on breast-mask images consistently outperformed those trained on original DICOM images across all architectures and metrics, with AUROC improvements ranging from +0.06 to +0.21. Among all models, ResNet50 trained on breast-mask images achieved the best performance (AUROC = 0.931; AUPRC = 0.933; balanced accuracy = 0.834), further improved after optimization (balanced accuracy = 0.886; sensitivity = 0.842; specificity = 0.930). Classical CNN architectures demonstrated performance comparable to or exceeding that of more complex hybrid CNN–Transformer models when anatomically focused preprocessing and rigorous optimization were applied. Conclusions: Anatomically constrained preprocessing through breast-mask segmentation substantially enhances deep learning performance and stability in CEM-based breast lesion classification. These findings indicate that input representation quality and training optimization are critical determinants of clinically relevant performance, often outweighing architectural complexity, and may support more reliable AI-assisted decision support in CEM workflows.
- Open Access
- 7 January 2023
- Published by MDPI AG in Current Oncology
- Vol. 30 (1) , 839-853
Machine Learning Approaches with Textural Features to Calculate Breast Density on Mammography
Background: breast cancer (BC) is the world’s most prevalent cancer in the female population, with 2.3 million new cases diagnosed worldwide in 2020. The great efforts made to set screening campaigns, early detection programs, and increasingly targeted treatments led to significant improvement in patients’ survival. The Full-Field Digital Mammograph (FFDM) is considered the gold standard method for the early diagnosis of BC. From several previous studies, it has emerged that breast density (BD) is a risk factor in the development of BC, affecting the periodicity of screening plans present today at an international level. Objective: in this study, the focus is the development of mammographic image processing techniques that allow the extraction of indicators derived from textural patterns of the mammary parenchyma indicative of BD risk factors. Methods: a total of 168 patients were enrolled in the internal training and test set while a total of 51 patients were enrolled to compose the external validation cohort. Different Machine Learning (ML) techniques have been employed to classify breasts based on the values of the tissue density. Textural features were extracted only from breast parenchyma with which to train classifiers, thanks to the aid of ML algorithms. Results: the accuracy of different tested classifiers varied between 74.15% and 93.55%. The best results were reached by a Support Vector Machine (accuracy of 93.55% and a percentage of true positives and negatives equal to TPP = 94.44% and TNP = 92.31%). The best accuracy was not influenced by the choice of the features selection approach. Considering the external validation cohort, the SVM, as the best classifier with the 7 features selected by a wrapper method, showed an accuracy of 0.95, a sensitivity of 0.96, and a specificity of 0.90. Conclusions: our preliminary results showed that the Radiomics analysis and ML approach allow us to objectively identify BD.
- 7 January 2023
- Published by MDPI AG in Current Oncology
- Vol. 30 (1) , 839-853
Open AccessMachine Learning Approaches with Textural Features to Calculate Breast Density on Mammography
Background: breast cancer (BC) is the world’s most prevalent cancer in the female population, with 2.3 million new cases diagnosed worldwide in 2020. The great efforts made to set screening campaigns, early detection programs, and increasingly targeted treatments led to significant improvement in patients’ survival. The Full-Field Digital Mammograph (FFDM) is considered the gold standard method for the early diagnosis of BC. From several previous studies, it has emerged that breast density (BD) is a risk factor in the development of BC, affecting the periodicity of screening plans present today at an international level. Objective: in this study, the focus is the development of mammographic image processing techniques that allow the extraction of indicators derived from textural patterns of the mammary parenchyma indicative of BD risk factors. Methods: a total of 168 patients were enrolled in the internal training and test set while a total of 51 patients were enrolled to compose the external validation cohort. Different Machine Learning (ML) techniques have been employed to classify breasts based on the values of the tissue density. Textural features were extracted only from breast parenchyma with which to train classifiers, thanks to the aid of ML algorithms. Results: the accuracy of different tested classifiers varied between 74.15% and 93.55%. The best results were reached by a Support Vector Machine (accuracy of 93.55% and a percentage of true positives and negatives equal to TPP = 94.44% and TNP = 92.31%). The best accuracy was not influenced by the choice of the features selection approach. Considering the external validation cohort, the SVM, as the best classifier with the 7 features selected by a wrapper method, showed an accuracy of 0.95, a sensitivity of 0.96, and a specificity of 0.90. Conclusions: our preliminary results showed that the Radiomics analysis and ML approach allow us to objectively identify BD.
- Open Access
Risk Assessment and Pancreatic Cancer: Diagnostic Management and Artificial Intelligence
Pancreatic cancer (PC) is one of the deadliest cancers, and it is responsible for a number of deaths almost equal to its incidence. The high mortality rate is correlated with several explanations; the main one is the late disease stage at which the majority of patients are diagnosed. Since surgical resection has been recognised as the only curative treatment, a PC diagnosis at the initial stage is believed the main tool to improve survival. Therefore, patient stratification according to familial and genetic risk and the creation of screening protocol by using minimally invasive diagnostic tools would be appropriate. Pancreatic cystic neoplasms (PCNs) are subsets of lesions which deserve special management to avoid overtreatment. The current PC screening programs are based on the annual employment of magnetic resonance imaging with cholangiopancreatography sequences (MR/MRCP) and/or endoscopic ultrasonography (EUS). For patients unfit for MRI, computed tomography (CT) could be proposed, although CT results in lower detection rates, compared to MRI, for small lesions. The actual major limit is the incapacity to detect and characterize the pancreatic intraepithelial neoplasia (PanIN) by EUS and MR/MRCP. The possibility of utilizing artificial intelligence models to evaluate higher-risk patients could favour the diagnosis of these entities, although more data are needed to support the real utility of these applications in the field of screening. For these motives, it would be appropriate to realize screening programs in research settings.
Open AccessRisk Assessment and Pancreatic Cancer: Diagnostic Management and Artificial Intelligence
Pancreatic cancer (PC) is one of the deadliest cancers, and it is responsible for a number of deaths almost equal to its incidence. The high mortality rate is correlated with several explanations; the main one is the late disease stage at which the majority of patients are diagnosed. Since surgical resection has been recognised as the only curative treatment, a PC diagnosis at the initial stage is believed the main tool to improve survival. Therefore, patient stratification according to familial and genetic risk and the creation of screening protocol by using minimally invasive diagnostic tools would be appropriate. Pancreatic cystic neoplasms (PCNs) are subsets of lesions which deserve special management to avoid overtreatment. The current PC screening programs are based on the annual employment of magnetic resonance imaging with cholangiopancreatography sequences (MR/MRCP) and/or endoscopic ultrasonography (EUS). For patients unfit for MRI, computed tomography (CT) could be proposed, although CT results in lower detection rates, compared to MRI, for small lesions. The actual major limit is the incapacity to detect and characterize the pancreatic intraepithelial neoplasia (PanIN) by EUS and MR/MRCP. The possibility of utilizing artificial intelligence models to evaluate higher-risk patients could favour the diagnosis of these entities, although more data are needed to support the real utility of these applications in the field of screening. For these motives, it would be appropriate to realize screening programs in research settings.
- Open Access
- 6 July 2021
- Published by MDPI AG in Journal of Personalized Medicine
- Vol. 11 (7) , 641
Evolution of CT Findings and Lung Residue in Patients with COVID-19 Pneumonia: Quantitative Analysis of the Disease with a Computer Automatic Tool
Purpose: the purpose of this study was to assess the evolution of computed tomography (CT) findings and lung residue in patients with COVID-19 pneumonia, via quantified evaluation of the disease, using a computer aided tool. Materials and methods: we retrospectively evaluated 341 CT examinations of 140 patients (68 years of median age) infected with COVID-19 (confirmed by real-time reverse transcriptase polymerase chain reaction (RT-PCR)), who were hospitalized, and who received clinical and CT examinations. All CTs were evaluated by two expert radiologists, in consensus, at the same reading session, using a computer-aided tool for quantification of the pulmonary disease. The parameters obtained using the computer tool included the healthy residual parenchyma, ground glass opacity, consolidation, and total lung volume. Results: statistically significant differences (p value ≤ 0.05) were found among quantified volumes of healthy residual parenchyma, ground glass opacity (GGO), consolidation, and total lung volume, considering different clinical conditions (stable, improved, and worsened). Statistically significant differences were found among quantified volumes for healthy residual parenchyma, GGO, and consolidation (p value ≤ 0.05) between dead patients and discharged patients. CT was not performed on cadavers; the death was an outcome, which was retrospectively included to differentiate findings of patients who survived vs. patients who died during hospitalization. Among discharged patients, complete disease resolutions on CT scans were observed in 62/129 patients with lung disease involvement ≤5%; lung disease involvement from 5% to 15% was found in 40/129 patients, while 27/129 patients had lung disease involvement between 16 and 30%. Moreover, 8–21 days (after hospital admission) was an “advanced period” with the most severe lung disease involvement. After the extent of involvement started to decrease—particularly after 21 days—the absorption was more obvious. Conclusions: a complete disease resolution on chest CT scans was observed in 48.1% of discharged patients using a computer-aided tool to quantify the GGO and consolidation volumes; after 16 days of hospital admission, the abnormalities identified by chest CT began to improve; in particular, the absorption was more obvious after 21 days.
- 6 July 2021
- Published by MDPI AG in Journal of Personalized Medicine
- Vol. 11 (7) , 641
Open AccessEvolution of CT Findings and Lung Residue in Patients with COVID-19 Pneumonia: Quantitative Analysis of the Disease with a Computer Automatic Tool
Purpose: the purpose of this study was to assess the evolution of computed tomography (CT) findings and lung residue in patients with COVID-19 pneumonia, via quantified evaluation of the disease, using a computer aided tool. Materials and methods: we retrospectively evaluated 341 CT examinations of 140 patients (68 years of median age) infected with COVID-19 (confirmed by real-time reverse transcriptase polymerase chain reaction (RT-PCR)), who were hospitalized, and who received clinical and CT examinations. All CTs were evaluated by two expert radiologists, in consensus, at the same reading session, using a computer-aided tool for quantification of the pulmonary disease. The parameters obtained using the computer tool included the healthy residual parenchyma, ground glass opacity, consolidation, and total lung volume. Results: statistically significant differences (p value ≤ 0.05) were found among quantified volumes of healthy residual parenchyma, ground glass opacity (GGO), consolidation, and total lung volume, considering different clinical conditions (stable, improved, and worsened). Statistically significant differences were found among quantified volumes for healthy residual parenchyma, GGO, and consolidation (p value ≤ 0.05) between dead patients and discharged patients. CT was not performed on cadavers; the death was an outcome, which was retrospectively included to differentiate findings of patients who survived vs. patients who died during hospitalization. Among discharged patients, complete disease resolutions on CT scans were observed in 62/129 patients with lung disease involvement ≤5%; lung disease involvement from 5% to 15% was found in 40/129 patients, while 27/129 patients had lung disease involvement between 16 and 30%. Moreover, 8–21 days (after hospital admission) was an “advanced period” with the most severe lung disease involvement. After the extent of involvement started to decrease—particularly after 21 days—the absorption was more obvious. Conclusions: a complete disease resolution on chest CT scans was observed in 48.1% of discharged patients using a computer-aided tool to quantify the GGO and consolidation volumes; after 16 days of hospital admission, the abnormalities identified by chest CT began to improve; in particular, the absorption was more obvious after 21 days.
- Open Access
Lymphadenopathy after BNT162b2 Covid-19 Vaccine: Preliminary Ultrasound Findings
During a spontaneous and autonomous study, we assessed the ultrasound finding of lymphadenopathy after BNT162b2 Pfizer vaccine. We enrolled 18 patients with 58 lymphadenopathies: in 10 patients, they were in the laterocervical side, while in 8 patients in the axillar site. The largest diameter was 16 mm with a range from 7 to 16 mm (median value = 10 mm). In the same patient, we found different ultrasound nodal findings. A total of 25 nodes showed eccentric cortical thickening with wide echogenic hilum and oval shape. In total, 19 nodes showed asymmetric eccentric cortical thickening with wide echogenic hilum and oval shape. Overall, 10 nodes showed concentric cortical thickening with reduction in the width of the echogenic hilum and oval shape. A total of four nodes showed huge reduction and displacement of the echogenic hilum and round or oval shape. No anomaly was found at the Doppler echocolor study. In conclusion, eccentric cortical thickening with wide echogenic hilum and oval shape, asymmetric eccentric cortical thickening with wide echogenic hilum and oval shape, concentric cortical thickening with reduction in the width of the echogenic hilum and oval shape, and a huge reduction and displacement of the echogenic hilum and round shape are the features that we found in post BNT162b2 Covid-19 Vaccine lymphadenopathies.
Open AccessLymphadenopathy after BNT162b2 Covid-19 Vaccine: Preliminary Ultrasound Findings
During a spontaneous and autonomous study, we assessed the ultrasound finding of lymphadenopathy after BNT162b2 Pfizer vaccine. We enrolled 18 patients with 58 lymphadenopathies: in 10 patients, they were in the laterocervical side, while in 8 patients in the axillar site. The largest diameter was 16 mm with a range from 7 to 16 mm (median value = 10 mm). In the same patient, we found different ultrasound nodal findings. A total of 25 nodes showed eccentric cortical thickening with wide echogenic hilum and oval shape. In total, 19 nodes showed asymmetric eccentric cortical thickening with wide echogenic hilum and oval shape. Overall, 10 nodes showed concentric cortical thickening with reduction in the width of the echogenic hilum and oval shape. A total of four nodes showed huge reduction and displacement of the echogenic hilum and round or oval shape. No anomaly was found at the Doppler echocolor study. In conclusion, eccentric cortical thickening with wide echogenic hilum and oval shape, asymmetric eccentric cortical thickening with wide echogenic hilum and oval shape, concentric cortical thickening with reduction in the width of the echogenic hilum and oval shape, and a huge reduction and displacement of the echogenic hilum and round shape are the features that we found in post BNT162b2 Covid-19 Vaccine lymphadenopathies.