André Santanchè
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
André Santanchè has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Marco Antonio de Carvalho Filho
Wenckebach Institute for Education and Training (WIOO), Lifelong Learning, Education and Assessment Research Network (LEARN), University of Groningen University Medical Center Groningen Groningen the Netherlands
4 shared publications - Claudia Bauzer Medeiros
Universidade Estadual de Campinas
3 shared publications - Ligia Maria Cayres Ribeiro
Cross‐Connection Education and Training (DVOO), Lifelong Learning, Education, Assessment, and Research Network (LEARN) University of Groningen, University Medical Center Groningen Groningen The Netherlands
2 shared publications - Silvia Mamede
University Medical Centre Groningen University of Groningen Groningen The Netherlands
2 shared publications - Ana Claudia Tonelli de Oliveira
Department of Internal Medicine, Hospital Clínicas de Porto Alegre, Porto Alegre, Brazil
1 shared publication Bruno S. C. M. Vilar
1 shared publication
Latest Publications
- Open Access
- 28 October 2026
- Published by Pensoft Publishers in JUCS - Journal of Universal Computer Science
- Vol. 32 (11) , 1560-1589
Semantic Representations based on Neural Topic Models
Topic Models is a tool to extract semantics from textual collections. However, in spite of extensive work on topic modeling, including its evaluation using Language Models (LM), there is need for more research to measure, analyze and validate the topics learned. To this end, we present a Neural Topic Model called ABT which leverages semantic vector spaces (derived from language models) to uncover latent semantic representations in text collections. ABT enables one to explore the themes covered by a given textual collection and analyze them as a hierarchical structure of topics. We demonstrate the effectiveness of ABT by analyzing two text corpora: (1) CliCR, a collection of 10,539 real clinical cases; and (2) MovieLens, a dataset consisting of 9,742 movie titles. We have experimented with different language models to investigate the effects of using specific tasks to tune the neural network. The topics that resulted from our study highlighted concepts that are shown to be effectively used as prominent topics to represent the datasets. The results show that each experimented LM focuses on distinct features to produce the topics and may generate topics with different grades of generality and specificity. Our study sheds light on recent findings of the related literature, which state that the topic coherence metric does not align with assessments performed by human evaluators.
- 28 October 2026
- Published by Pensoft Publishers in JUCS - Journal of Universal Computer Science
- Vol. 32 (11) , 1560-1589
Open AccessSemantic Representations based on Neural Topic Models
Topic Models is a tool to extract semantics from textual collections. However, in spite of extensive work on topic modeling, including its evaluation using Language Models (LM), there is need for more research to measure, analyze and validate the topics learned. To this end, we present a Neural Topic Model called ABT which leverages semantic vector spaces (derived from language models) to uncover latent semantic representations in text collections. ABT enables one to explore the themes covered by a given textual collection and analyze them as a hierarchical structure of topics. We demonstrate the effectiveness of ABT by analyzing two text corpora: (1) CliCR, a collection of 10,539 real clinical cases; and (2) MovieLens, a dataset consisting of 9,742 movie titles. We have experimented with different language models to investigate the effects of using specific tasks to tune the neural network. The topics that resulted from our study highlighted concepts that are shown to be effectively used as prominent topics to represent the datasets. The results show that each experimented LM focuses on distinct features to produce the topics and may generate topics with different grades of generality and specificity. Our study sheds light on recent findings of the related literature, which state that the topic coherence metric does not align with assessments performed by human evaluators.
- Open Access
- 17 September 2026
- Published by Wiley in Medical Education
- Vol. 60 (12) , 1392-1402
‘Tell me what you know’: Written recall as a window into illness script quality
Introduction: Illness scripts (mental representations of disease knowledge) are central to diagnostic expertise, yet structured methods to examine their quality across medical training are lacking. This exploratory study investigated whether textual indicators derived from free‐recall protocols could reveal aspects of illness script quality in medical students. Methods: In this cross‐sectional study, students from years one through six at seven Brazilian medical schools completed a free‐recall task for chronic obstructive pulmonary disease (COPD) and myocardial infarction (MI) after a progress test. Written responses were segmented into idea units (IUs) and analysed for accuracy (correct idea units), scope (number of disease dimensions), organisation (balanced adjusted ratio of clustering) and encapsulation (higher‐order summary terms). Differences across training years were examined using Kruskal–Wallis and Dunn tests. Results: Accuracy and scope increased significantly across training years for both diseases (median correct ideas for COPD: 3 in year 1 versus 14 in year 5, p < 0.01; for MI: 8 versus 18, p < 0.01). Most gains occurred between years 1–2 and 3, with limited differentiation thereafter. Organisation increased from 0.31 in year 1 to 0.61 in year 5 for COPD and from 0.28 to 0.60 for MI (range −1 to +1; both p < 0.01). Encapsulated and incorrect ideas were infrequent at all training levels. Progress test scores correlated positively with the number of correct ideas (COPD rho = 0.56; MI rho = 0.46; p < 0.01). Conclusions: Free‐recall protocols analysed along theoretically grounded dimensions detected differences in accuracy, scope and organisation across training years. Constraints, such as a time‐imposed ceiling effect and limited sensitivity to encapsulated knowledge, should guide future methodological refinement. Despite its preliminary nature, this methodology may contribute to research on clinical expertise by offering a novel approach to studying illness scripts across training levels.
- 17 September 2026
- Published by Wiley in Medical Education
- Vol. 60 (12) , 1392-1402
Open Access‘Tell me what you know’: Written recall as a window into illness script quality
Introduction: Illness scripts (mental representations of disease knowledge) are central to diagnostic expertise, yet structured methods to examine their quality across medical training are lacking. This exploratory study investigated whether textual indicators derived from free‐recall protocols could reveal aspects of illness script quality in medical students. Methods: In this cross‐sectional study, students from years one through six at seven Brazilian medical schools completed a free‐recall task for chronic obstructive pulmonary disease (COPD) and myocardial infarction (MI) after a progress test. Written responses were segmented into idea units (IUs) and analysed for accuracy (correct idea units), scope (number of disease dimensions), organisation (balanced adjusted ratio of clustering) and encapsulation (higher‐order summary terms). Differences across training years were examined using Kruskal–Wallis and Dunn tests. Results: Accuracy and scope increased significantly across training years for both diseases (median correct ideas for COPD: 3 in year 1 versus 14 in year 5, p < 0.01; for MI: 8 versus 18, p < 0.01). Most gains occurred between years 1–2 and 3, with limited differentiation thereafter. Organisation increased from 0.31 in year 1 to 0.61 in year 5 for COPD and from 0.28 to 0.60 for MI (range −1 to +1; both p < 0.01). Encapsulated and incorrect ideas were infrequent at all training levels. Progress test scores correlated positively with the number of correct ideas (COPD rho = 0.56; MI rho = 0.46; p < 0.01). Conclusions: Free‐recall protocols analysed along theoretically grounded dimensions detected differences in accuracy, scope and organisation across training years. Constraints, such as a time‐imposed ceiling effect and limited sensitivity to encapsulated knowledge, should guide future methodological refinement. Despite its preliminary nature, this methodology may contribute to research on clinical expertise by offering a novel approach to studying illness scripts across training levels.
- Open Access
- 1 January 2026
- Published by EJournal Publishing in International Journal of Information and Education Technology
- Vol. 16 (3) , 695-707
Clinical Reasoning-Driven Progress Evaluation of Medical Students Using Large Language Models
Evaluating medical students’ written answers to questions on a given topic can provide information about their mental representations of a disease―i.e., illness script. However, limitations in methods for assessing how medical students develop clinical expertise hinder the advancement of educational practices. This study, therefore, proposes a technique that utilizes semantic annotations of the students’ answers to trace a map of their knowledge concerning the topic of a question. Since manual text annotation is time- and effort-intensive, this study developed an innovative, illness-script-driven strategy using large language models. It identifies relevant medical information in the texts, creates a profile for each answer, clusters them, and categorizes the clusters. Practical experiments with Brazilian students demonstrate that the technique automatically traces consistent profiles from the responses, quantifying how knowledge of disease diagnosis evolves throughout the medical course.
- 1 January 2026
- Published by EJournal Publishing in International Journal of Information and Education Technology
- Vol. 16 (3) , 695-707
Open AccessClinical Reasoning-Driven Progress Evaluation of Medical Students Using Large Language Models
Evaluating medical students’ written answers to questions on a given topic can provide information about their mental representations of a disease―i.e., illness script. However, limitations in methods for assessing how medical students develop clinical expertise hinder the advancement of educational practices. This study, therefore, proposes a technique that utilizes semantic annotations of the students’ answers to trace a map of their knowledge concerning the topic of a question. Since manual text annotation is time- and effort-intensive, this study developed an innovative, illness-script-driven strategy using large language models. It identifies relevant medical information in the texts, creates a profile for each answer, clusters them, and categorizes the clusters. Practical experiments with Brazilian students demonstrate that the technique automatically traces consistent profiles from the responses, quantifying how knowledge of disease diagnosis evolves throughout the medical course.
- 1 December 2025
- Published by Springer Nature
- p. 371-381
Interactive Digital Narrative and Structured Debate for Learning Ethics
- 1 December 2025
- Published by Springer Nature
- p. 371-381
Interactive Digital Narrative and Structured Debate for Learning Ethics
- 30 January 2025
- Published by Springer Nature
- p. 199-211
LLM-Driven Chest X-Ray Report Generation With a Modular, Reduced-Size Architecture
- 30 January 2025
- Published by Springer Nature
- p. 199-211
LLM-Driven Chest X-Ray Report Generation With a Modular, Reduced-Size Architecture
- Open Access
- 13 December 2024
- Published by Springer Nature in Journal of Imaging Informatics in Medicine
- Vol. 38 (5) , 2664-2676
Semi-supervised Ensemble Learning for Automatic Interpretation of Lung Ultrasound Videos
Point-of-care ultrasound (POCUS) stands as a safe, portable, and cost-effective imaging modality for swift bedside patient examinations. Specifically, lung ultrasonography (LUS) has proven useful in evaluating both acute and chronic pulmonary conditions. Despite its clinical value, automatic LUS interpretation remains relatively unexplored, particularly in multi-label contexts. This work proposes a novel deep learning (DL) framework tailored for interpreting lung POCUS videos, whose outputs are the finding(s) present in these videos (such as A-lines, B-lines, or consolidations). The pipeline, based on a residual (2+1)D architecture, initiates with a pre-processing routine for video masking and standardisation, and employs a semi-supervised approach to harness available unlabeled data. Additionally, we introduce an ensemble modeling strategy that aggregates outputs from models trained to predict distinct label sets, thereby leveraging the hierarchical nature of LUS findings. The proposed framework and its building blocks were evaluated through extensive experiments with both multi-class and multi-label models, highlighting its versatility. In a held-out test set, the categorical proposal, suited for expedite triage, achieved an average F1-score of 92.4%, while the multi-label proposal, helpful for patient management and referral, achieved an average F1-score of 70.5% across five relevant LUS findings. Overall, the semi-supervised methodology contributed significantly to improved performance, while the proposed hierarchy-aware ensemble provided moderate additional gains.
- 13 December 2024
- Published by Springer Nature in Journal of Imaging Informatics in Medicine
- Vol. 38 (5) , 2664-2676
Open AccessSemi-supervised Ensemble Learning for Automatic Interpretation of Lung Ultrasound Videos
Point-of-care ultrasound (POCUS) stands as a safe, portable, and cost-effective imaging modality for swift bedside patient examinations. Specifically, lung ultrasonography (LUS) has proven useful in evaluating both acute and chronic pulmonary conditions. Despite its clinical value, automatic LUS interpretation remains relatively unexplored, particularly in multi-label contexts. This work proposes a novel deep learning (DL) framework tailored for interpreting lung POCUS videos, whose outputs are the finding(s) present in these videos (such as A-lines, B-lines, or consolidations). The pipeline, based on a residual (2+1)D architecture, initiates with a pre-processing routine for video masking and standardisation, and employs a semi-supervised approach to harness available unlabeled data. Additionally, we introduce an ensemble modeling strategy that aggregates outputs from models trained to predict distinct label sets, thereby leveraging the hierarchical nature of LUS findings. The proposed framework and its building blocks were evaluated through extensive experiments with both multi-class and multi-label models, highlighting its versatility. In a held-out test set, the categorical proposal, suited for expedite triage, achieved an average F1-score of 92.4%, while the multi-label proposal, helpful for patient management and referral, achieved an average F1-score of 70.5% across five relevant LUS findings. Overall, the semi-supervised methodology contributed significantly to improved performance, while the proposed hierarchy-aware ensemble provided moderate additional gains.
- Open Access
- 1 August 2022
- Published by Journal of Graduate Medical Education in Journal of Graduate Medical Education
- Vol. 14 (4) , 475-481
The Effect of Information Presentation Order on Residents' Diagnostic Accuracy of Online Simulated Patients With Chest Pain
Background: Physicians may receive diagnostic information in different orders, and there is a lack of empirical evidence that the order of presentation may influence clinical reasoning. Objective: We investigated whether diagnostic accuracy of chest pain cases is influenced by the order of presentation of the history and electrocardiogram (EKG) to cardiology residents. Methods: We conducted an experimental study during a resident training in 2019. Twelve clinical cases were presented in 2 diagnostic rounds. Residents were randomly allocated to seeing the EKG first (EKGF) or the history first (HF). The mean diagnostic accuracy scores (range 0-1) and confidence level (0-100) in each diagnostic round and time needed to make the diagnosis were evaluated. Results: The final diagnostic accuracy was higher than the initial in both groups. After the first round, diagnostic accuracy was higher in HF (n=24) than in EKGF (n=28). Time taken to judge the history was comparable in both groups. Time taken to judge the EKG was shorter in HF (40±11 vs 64±13 seconds; P<.01). Time invested in the second round was significantly correlated with changing the initial diagnosis. A significant difference was observed in confidence ratings after the initial diagnosis, with EKGF reporting less confidence relative to HF. Conclusions: The order in which history and EKG are presented influences the clinical reasoning process.
- 1 August 2022
- Published by Journal of Graduate Medical Education in Journal of Graduate Medical Education
- Vol. 14 (4) , 475-481
Open AccessThe Effect of Information Presentation Order on Residents' Diagnostic Accuracy of Online Simulated Patients With Chest Pain
Background: Physicians may receive diagnostic information in different orders, and there is a lack of empirical evidence that the order of presentation may influence clinical reasoning. Objective: We investigated whether diagnostic accuracy of chest pain cases is influenced by the order of presentation of the history and electrocardiogram (EKG) to cardiology residents. Methods: We conducted an experimental study during a resident training in 2019. Twelve clinical cases were presented in 2 diagnostic rounds. Residents were randomly allocated to seeing the EKG first (EKGF) or the history first (HF). The mean diagnostic accuracy scores (range 0-1) and confidence level (0-100) in each diagnostic round and time needed to make the diagnosis were evaluated. Results: The final diagnostic accuracy was higher than the initial in both groups. After the first round, diagnostic accuracy was higher in HF (n=24) than in EKGF (n=28). Time taken to judge the history was comparable in both groups. Time taken to judge the EKG was shorter in HF (40±11 vs 64±13 seconds; P<.01). Time invested in the second round was significantly correlated with changing the initial diagnosis. A significant difference was observed in confidence ratings after the initial diagnosis, with EKGF reporting less confidence relative to HF. Conclusions: The order in which history and EKG are presented influences the clinical reasoning process.
- 31 July 2015
- Published by Springer Nature
- p. 397-414
Interpretation of Construction Patterns for Biodiversity Spreadsheets
- 31 July 2015
- Published by Springer Nature
- p. 397-414
Interpretation of Construction Patterns for Biodiversity Spreadsheets
- 1 January 2015
- Published by Springer Nature
- p. 317-321
Querying and Managing Complex Data
- 1 January 2013
- Published by Elsevier in Journal of Web Semantics
- Vol. 18 (1) , 19-30
Folksonomized ontology and the 3E steps technique to support ontology evolvement
- 1 January 2013
- Published by Elsevier in Journal of Web Semantics
- Vol. 18 (1) , 19-30
Folksonomized ontology and the 3E steps technique to support ontology evolvement