Laura Bergomi

Dr. Laura Bergomi

orcid

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
16 July 2025 in BMC Medical Informatics and Decision Making
Verified
Journal Article
Open access
PDF file
LBLaura Bergomi
Laura Bergomi
GNGiovanna Nicora
Giovanna Nicora
MOMarta Anna Orlowska
Marta Anna Orlowska
Chiara PodreccaChiara Podrecca
Chiara Podrecca
RBRiccardo Bellazzi
Riccardo Bellazzi
Enea ParimbelliEnea Parimbelli
Enea Parimbelli
Background: This study aims to address the gap in understanding clinicians’ attitudes toward explainable AI (XAI) methods applied to machine learning models using tabular data, commonly found in clinical settings. It specifically explores clinicians’ perceptions of different XAI methods from the ALFABETO project, which predicts COVID-19 patient hospitalization based on clinical, laboratory, and chest X-ray at time of presentation to the Emergency Department. The focus is on two cognitive dimensions: understandability and actionability of the explanations provided by explainable-by-design and post-hoc methods. Methods: A questionnaire-based experiment was conducted with 10 clinicians from the IRCCS Policlinico San Matteo Foundation in Pavia, Italy. Each clinician evaluated 10 real-world cases, rating predictions and explanations from three XAI tools: Bayesian networks, SHapley Additive exPlanations (SHAP), and AraucanaXAI. Two cognitive statements for each method were rated on a Likert scale, as well as the agreement with the prediction. Two clinicians answered the survey during think-aloud interviews. Results: Clinicians demonstrated generally positive attitudes toward AI, but high compliance rates (86% on average) indicate a risk of automation bias. Understandability and actionability are positively correlated, with SHAP being the preferred method due to its simplicity. However, the perception of methods varies according to specialty and expertise. Conclusions: The findings suggest that SHAP and AraucanaXAI are promising candidates for improving the use of XAI in clinical decision support systems (DSSs), highlighting the importance of clinicians’ expertise, specialty, and setting on the selection and development of supportive XAI advice. Finally, the study provides valuable insights into the design of future XAI DSSs.

26 June 2024 in Artificial Intelligence in Medicine
Verified
Journal Article
Open access
LBLaura Bergomi
Laura Bergomi
TBTommaso M. Buonocore
Tommaso M. Buonocore
PAPaolo Antonazzo
Paolo Antonazzo
LALorenzo Alberghi
Lorenzo Alberghi
RBRiccardo Bellazzi
Riccardo Bellazzi
Enea ParimbelliEnea Parimbelli
Enea Parimbelli
Radiology reports are typically written in a free-text format, making clinical information difficult to extract and use. Recently, the adoption of structured reporting (SR) has been recommended by various medical societies thanks to the advantages it offers, e.g. standardization, completeness, and information retrieval. We propose a pipeline to extract information from Italian free-text radiology reports that fits with the items of the reference SR registry proposed by a national society of interventional and medical radiology, focusing on CT staging of patients with lymphoma. Our work aims to leverage the potential of Natural Language Processing and Transformer-based models to deal with automatic SR registry filling. With the availability of 174 Italian radiology reports, we investigate a rule-free generative Question Answering approach based on the Italian-specific version of T5: IT5. To address information content discrepancies, we focus on the six most frequently filled items in the annotations made on the reports: three categorical (multichoice), one free-text (free-text), and two continuous numerical (factual). In the preprocessing phase, we encode also information that is not supposed to be entered. Two strategies (batch-truncation and ex-post combination) are implemented to comply with the IT5 context length limitations. Performance is evaluated in terms of strict accuracy, f1, and format accuracy, and compared with the widely used GPT-3.5 Large Language Model. Unlike multichoice and factual, free-text answers do not have 1-to-1 correspondence with their reference annotations. For this reason, we collect human-expert feedback on the similarity between medical annotations and generated free-text answers, using a 5-point Likert scale questionnaire (evaluating the criteria of correctness and completeness). The combination of fine-tuning and batch splitting allows IT5 ex-post combination to achieve notable results in terms of information extraction of different types of structured data, performing on par with GPT-3.5. Human-based assessment scores of free-text answers show a high correlation with the AI performance metrics f1 (Spearman's correlation coefficients>0.5, p-values<0.001) for both IT5 ex-post combination and GPT-3.5. The latter is better at generating plausible human-like statements, even if it systematically provides answers even when they are not supposed to be given. In our experimental setting, a fine-tuned Transformer-based model with a modest number of parameters (i.e., IT5, 220 M) performs well as a clinical information extraction system for automatic SR registry filling task. It can extract information from more than one place in the report, elaborating it in a manner that complies with the response specifications provided by the SR registry (for multichoice and factual items), or that closely approximates the work of a human-expert (free-text items); with the ability to discern when an answer is supposed to be given or not to a user query.

10 December 2024 in bioRxiv
Verified
Preprint
Open access
PDF file
SRSimone Rancati
Simone Rancati
GNGiovanna Nicora
Giovanna Nicora
LBLaura Bergomi
Laura Bergomi
TBTommaso Mario Buonocore
Tommaso Mario Buonocore
Daniel CzyzDaniel Czyz
Daniel Czyz
SMSimone Marini
Simone Marini
The COVID-19 pandemic has profoundly impacted global health, economics, and daily life, with over 776 million cases and 7 million deaths from December 2019 to November 2024. Since the original SARS-CoV-2 Wuhan strain emerged, the virus has evolved into variants such as Alpha, Beta, Gamma, Delta, and Omicron, all characterized by mutations in the Spike glycoprotein, critical for viral entry into human cells via its S1 and S2 subunits. The S1 subunit, binding to the ACE2 receptor and mutating frequently, affects infectivity and immune evasion; the more conserved S2, on the other hand, facilitates membrane fusion. Predicting future mutations is crucial for developing vaccines and treatments adaptable to emerging strains, enhancing preparedness and intervention design. Generative Large Language Models (LLMs) are becoming increasingly common in the field of genomics, given their ability to generate realistic synthetic biological sequences, including applications in protein design and engineering. Here we present SARITA, an LLM with up to 1.2 billion parameters, based on GPT-3 architecture, designed to generate high-quality synthetic SARS-CoV-2 Spike S1 sequences. SARITA is trained via continuous learning on the pre-existing protein model RITA. When trained on Alpha, Beta, and Gamma variants (data up to February 2021 included), SARITA correctly predicts the evolution of future S1 mutations, including characterized mutations of Delta, Omicron and Iota variants. Furthermore, we show how SARITA outperforms alternative approaches, including other LLMs, in terms of sequence quality, realism, and similarity with real-world S1 sequences. These results indicate the potential of SARITA to predict future SARS-CoV-2 S1 evolution, potentially aiding in the development of adaptable vaccines and treatments.

22 June 2025
Verified
Book Chapter
Closed access
PDF file
LBLaura Bergomi
Laura Bergomi
Enea ParimbelliEnea Parimbelli
Enea Parimbelli
DPDaniele Pala
Daniele Pala
TBTommaso M. Buonocore
Tommaso M. Buonocore

29 May 2026 in JMIR Preprints
Verified
Preprint
Open access
BCBruno Curia
Bruno Curia
PAPaolo Antonazzo
Paolo Antonazzo
LBLaura Bergomi
Laura Bergomi
Enea ParimbelliEnea Parimbelli
Enea Parimbelli
RBRiccardo Bellazzi
Riccardo Bellazzi
LPLorenzo Preda
Lorenzo Preda
BACKGROUND Radiology reports are commonly written as free text, which limits data standardization, automated analysis, and secondary use for research and clinical decision support. Structured reporting templates may improve consistency and interoperability, but their adoption can be limited by increased reporting burden and reduced narrative flexibility. Large language models may support the automatic extraction of structured information from existing free-text radiology reports. OBJECTIVE To compare different large language models (LLMs) for natural language processing of free-text CT reports in patients with lymphoma, assessing their ability to automatically populate a standardized structured reporting (SR) template. METHODS In this single-center retrospective study, 174 Italian CT reports for lymphoma staging were analyzed. The Report section of the SIRM lymphoma CT SR template (65 fields: 19 free-text, 29 numerical, 17 multiple-choice) was used as reference structure. Two radiologists generated reference annotations and inter-annotator agreement was assessed with Krippendorff’s alpha. Claude 3.5 Sonnet, Gemini 1.5 Flash, and Mistral Large 2407 were tested in two settings: full schema and lesion-filtered schema. Performance was evaluated using accuracy, specificity, sensitivity, and F1-score. RESULTS Mean global inter-annotator agreement was 0.75±0.13. All models performed better with lesionfiltered schemas than with full schemas: mean accuracy increased from 0.63–0.64 to 0.72–0.74 and F1-score from 0.52–0.54 to 0.76–0.78, with specificity remaining >0.82. Multiple-choice fields showed the best performance (filtered-schema F1 >0.90; sensitivity ≥0.95), followed by numerical fields (F1 ~0.80), while free-text fields were the most challenging (F1 ~0.57). Claude 3.5 Sonnet achieved the best overall results (accuracy 0.74; F1-score 0.78), closely followed by Gemini 1.5 Flash. CONCLUSIONS General-purpose LLMs can effectively extract structured information from Italian lymphoma CT free-text reports without domain-specific fine-tuning. Schema filtering substantially improves performance and may facilitate integration of LLM-based tools into radiology workflows for SR completion, research data reuse, and clinical decision support.

21 May 2026
Verified
Book Chapter
Open access
LBLaura Bergomi
Laura Bergomi
TBTommaso Mario Buonocore
Tommaso Mario Buonocore
EPEnea Parimbelli
Enea Parimbelli
MLMarco Vincenzo Lenti
Marco Vincenzo Lenti
GSGiovanni Santacroce
Giovanni Santacroce
DPDaniele Pala
Daniele Pala
While Large Language Models (LLMs) hold great potential for clinical applications, their use is limited by concerns regarding data privacy, high computational demand, and the risk of hallucinations. Small Language Models (SLMs) are a promising solution, enabling efficient and secure on-device processing. This study presents the application of a local IT5 model finetuned to extract endoscopic markers from Italian annotated clinical notes of patients with Autoimmune Atrophic Gastritis (AAG). The results show that this model performs competitively with both GPT-4o mini—a general-purpose model—and MedGemma—a medical-oriented model—in this specific task, achieving high sensitivity, which is crucial for rare disease detection. These findings highlight the advantages of local, task-specific SLMs for privacy-preserving applications within healthcare settings.