Typology
operator:
and / or
Language
operator:
and / or
Date
operator:
and / or
Rights
operator:
and / or
2026
Other
Open Access
Robust computer vision for POCUS: achieving reliability in ultrasound imaging
Ignesti Giacomo
The rapid proliferation of Deep Learning (DL) has fundamentally transformed the scientific landscape, shifting the paradigm of problem-solving from explicit, algorithmic logic to data-driven feature learning. However, as these technologies permeate high-stakes domains such as medicine and precision agriculture, a critical tension has emerged. The very non-linearity that grants Neural Networks their immense expressive power, as described by the Universal Approximation Theorem, simultaneously causes intrinsic opacity, gradient instability, and a susceptibility to spurious correlations. Consequently, the deployment of these black-box models faces significant hurdles in terms of reliability and trust, necessitating the development of Artificial Intelligence (AI) that is not merely accurate but also fundamentally reliable and interpretable. The Clinical Context: Telemedicine and the TiAssisto Project. This thesis grounds its research in the critical context of the SARS-CoV-2 pandemic, which exposed systemic fragilities in healthcare and highlighted the need for resilient, decentralised care models. A significant portion of the applied research was conducted within the TiAssisto project, a server-based web platform designed to support multi-pathological patients by bridging the gap between hospitalisation and home care. To operationalise the vision of Medicine 5.0—a human-centric approach to healthcare—TiAssisto integrates a hybrid AI framework. This comprises a Strong AI component utilising a rule-based algorithm and logic for monitoring vital signs, and a Weak AI component employing a DL Model for the automatic classification of medical imaging. The platform prioritises patient security through strict local encryption and anonymisation protocols, anticipating the growing necessity for Reliable AI architectures. The Challenge of Point-of-Care Ultrasound (POCUS). Within this framework, the thesis identifies Lung Ultrasound (LUS) as a pivotal technology for telemedicine, given its portability and safety. However, LUS presents unique challenges for Computer Vision; unlike MRI or CT scans, LUS images are characterised by a low signal-to-noise ratio and are artefact-rich. Diagnosis often relies on interpreting acoustic artefacts, such as "A-lines" (horizontal repetitions indicating a healthy lung) or "B-lines" (vertical comet-tail artefacts indicating pathology), rather than direct anatomical visualization. To address the computational constraints of Point-of-Care (POCUS) devices and the scarcity of labelled data, the research proposes a novel Efficient Adaptive Ensembling methodology. This approach utilises the EfficientNet-b0 DL architecture as a backbone due to its optimal balance of accuracy and complexity. By training weak learners on data subsets to induce diversity and subsequently synthesising their deep features via a trainable linear combination layer, the model achieves robustness without the computational cost of traditional ensembles. When validated on a large public LUS dataset, this ensemble model achieved 100% accuracy in distinguishing between COVID-19, Pneumonia, and Healthy classes, significantly outperforming previous state-of-the-art models. Crucially, it maintained high efficiency, requiring only 0.78 G-FLOPs, making it suitable for real-time deployment on edge devices. Foundations of Reliable AI: From Diagnosis to Constraint. While achieving high accuracy is essential, the thesis argues that clinical adoption hinges on trust. Initial efforts utilised post-hoc Explainable AI (XAI) methods, specifically Grad-CAM, to visually validate that the LUS model focused on clinically relevant areas, such as the pleural line. However, the research identifies a critical limitation in post-hoc analysis: it is purely diagnostic and cannot correct a model’s flawed reasoning. To transcend this limitation, the thesis introduces a paradigm shift towards Intrinsically Guided Training. The research proposes moving the generation of Class Activation Maps (CAM) into the training loop itself. By incorporating a custom Contrastive Loss term, the training process actively penalises the model if its internal attention does not align with the relevant input features. This concept evolved into the Batch-CAM paradigm. This advanced methodology replaces instance-specific reconstruction with a Class Prototype approach, where the model compares batch-averaged CAMs against a learned prototype of the target class. This enforces semantic consistency, compelling the model to learn a generalised representation of a class (e.g., the concept of a "T-shirt" or a specific digit) rather than memorising individual examples. Experimental validation on MNIST and Fashion-MNIST demonstrated that this approach yields models with competitive accuracy and significantly superior, human-intelligible internal representations. Conclusion and Future Outlook. The thesis concludes that the path to Reliable AI lies in shifting from passive analysis to active constraint during the learning process. By embedding interpretability into the loss function, models can be guided to learn for the right reasons, mitigating the risk of spurious correlations. The first application of the method to LUS interpretation yields high accuracy, and the systems, now partially guided, seem to learn meaningful anatomical details to inform their decisions. Looking forward, the research outlines a trajectory towards Continual Learning to prevent catastrophic forgetting as medical protocols evolve, and Federated Learning to facilitate privacy-preserving training on decentralised data. Ultimately, integrating these Plausible AI models into hybrid Decision Support Systems offers a structure for the future of healthcare systems that are efficient, transparent, and trustworthy partners in human decision-making.
Ignesti Giacomo
The rapid proliferation of Deep Learning (DL) has fundamentally transformed the scientific landscape, shifting the paradigm of problem-solving from explicit, algorithmic logic to data-driven feature learning. However, as these technologies permeate high-stakes domains such as medicine and precision agriculture, a critical tension has emerged. The very non-linearity that grants Neural Networks their immense expressive power, as described by the Universal Approximation Theorem, simultaneously causes intrinsic opacity, gradient instability, and a susceptibility to spurious correlations. Consequently, the deployment of these black-box models faces significant hurdles in terms of reliability and trust, necessitating the development of Artificial Intelligence (AI) that is not merely accurate but also fundamentally reliable and interpretable. The Clinical Context: Telemedicine and the TiAssisto Project. This thesis grounds its research in the critical context of the SARS-CoV-2 pandemic, which exposed systemic fragilities in healthcare and highlighted the need for resilient, decentralised care models. A significant portion of the applied research was conducted within the TiAssisto project, a server-based web platform designed to support multi-pathological patients by bridging the gap between hospitalisation and home care. To operationalise the vision of Medicine 5.0—a human-centric approach to healthcare—TiAssisto integrates a hybrid AI framework. This comprises a Strong AI component utilising a rule-based algorithm and logic for monitoring vital signs, and a Weak AI component employing a DL Model for the automatic classification of medical imaging. The platform prioritises patient security through strict local encryption and anonymisation protocols, anticipating the growing necessity for Reliable AI architectures. The Challenge of Point-of-Care Ultrasound (POCUS). Within this framework, the thesis identifies Lung Ultrasound (LUS) as a pivotal technology for telemedicine, given its portability and safety. However, LUS presents unique challenges for Computer Vision; unlike MRI or CT scans, LUS images are characterised by a low signal-to-noise ratio and are artefact-rich. Diagnosis often relies on interpreting acoustic artefacts, such as "A-lines" (horizontal repetitions indicating a healthy lung) or "B-lines" (vertical comet-tail artefacts indicating pathology), rather than direct anatomical visualization. To address the computational constraints of Point-of-Care (POCUS) devices and the scarcity of labelled data, the research proposes a novel Efficient Adaptive Ensembling methodology. This approach utilises the EfficientNet-b0 DL architecture as a backbone due to its optimal balance of accuracy and complexity. By training weak learners on data subsets to induce diversity and subsequently synthesising their deep features via a trainable linear combination layer, the model achieves robustness without the computational cost of traditional ensembles. When validated on a large public LUS dataset, this ensemble model achieved 100% accuracy in distinguishing between COVID-19, Pneumonia, and Healthy classes, significantly outperforming previous state-of-the-art models. Crucially, it maintained high efficiency, requiring only 0.78 G-FLOPs, making it suitable for real-time deployment on edge devices. Foundations of Reliable AI: From Diagnosis to Constraint. While achieving high accuracy is essential, the thesis argues that clinical adoption hinges on trust. Initial efforts utilised post-hoc Explainable AI (XAI) methods, specifically Grad-CAM, to visually validate that the LUS model focused on clinically relevant areas, such as the pleural line. However, the research identifies a critical limitation in post-hoc analysis: it is purely diagnostic and cannot correct a model’s flawed reasoning. To transcend this limitation, the thesis introduces a paradigm shift towards Intrinsically Guided Training. The research proposes moving the generation of Class Activation Maps (CAM) into the training loop itself. By incorporating a custom Contrastive Loss term, the training process actively penalises the model if its internal attention does not align with the relevant input features. This concept evolved into the Batch-CAM paradigm. This advanced methodology replaces instance-specific reconstruction with a Class Prototype approach, where the model compares batch-averaged CAMs against a learned prototype of the target class. This enforces semantic consistency, compelling the model to learn a generalised representation of a class (e.g., the concept of a "T-shirt" or a specific digit) rather than memorising individual examples. Experimental validation on MNIST and Fashion-MNIST demonstrated that this approach yields models with competitive accuracy and significantly superior, human-intelligible internal representations. Conclusion and Future Outlook. The thesis concludes that the path to Reliable AI lies in shifting from passive analysis to active constraint during the learning process. By embedding interpretability into the loss function, models can be guided to learn for the right reasons, mitigating the risk of spurious correlations. The first application of the method to LUS interpretation yields high accuracy, and the systems, now partially guided, seem to learn meaningful anatomical details to inform their decisions. Looking forward, the research outlines a trajectory towards Continual Learning to prevent catastrophic forgetting as medical protocols evolve, and Federated Learning to facilitate privacy-preserving training on decentralised data. Ultimately, integrating these Plausible AI models into hybrid Decision Support Systems offers a structure for the future of healthcare systems that are efficient, transparent, and trustworthy partners in human decision-making.
2026
Conference article
Open Access
Advancements in Artificial Intelligence and Computer Vision for Biomedical Applications @SI-Lab
Marco Cafiso, Andrea Carboni, Claudia Caudai, Sara Colantonio, Francesco Conti, Mario D’acunto, Said Daoudagh, Giulio Del Corso, Riccardo Del Gratta, Danila Germanese, Giacomo Ignesti, Gianmarco Lazzini, Giuseppe Riccardo Leone, Barbara Leporini, Massimo Magrini, Massimo Martinelli, Davide Moroni, Ali Reza Omrani, Francesca Pardini, Maria Antonietta Pascali, Paolo Paradisi, Laura Sebastiaini, Marco Tampucci, Valeria Tateo, Federico Volpini
This paper summarizes recent research at the Signals and Images Lab (ISTI-CNR) leveraging artificial intelligence, machine learning, and computer vision to address complex biomedical challenges. We highlight advancements in trustworthy, explainable algorithmic foundations alongside their practical clinical applications across diagnostic imaging, physiological signal analysis and neuromotor rehabilitation.
Marco Cafiso, Andrea Carboni, Claudia Caudai, Sara Colantonio, Francesco Conti, Mario D’acunto, Said Daoudagh, Giulio Del Corso, Riccardo Del Gratta, Danila Germanese, Giacomo Ignesti, Gianmarco Lazzini, Giuseppe Riccardo Leone, Barbara Leporini, Massimo Magrini, Massimo Martinelli, Davide Moroni, Ali Reza Omrani, Francesca Pardini, Maria Antonietta Pascali, Paolo Paradisi, Laura Sebastiaini, Marco Tampucci, Valeria Tateo, Federico Volpini
This paper summarizes recent research at the Signals and Images Lab (ISTI-CNR) leveraging artificial intelligence, machine learning, and computer vision to address complex biomedical challenges. We highlight advancements in trustworthy, explainable algorithmic foundations alongside their practical clinical applications across diagnostic imaging, physiological signal analysis and neuromotor rehabilitation.
2026
Conference article
Restricted
Reliable and trustworthy learning prototype: insight from POCUS
Ignesti G., D’angelo G., Pratali L., Moroni D., Martinelli M.
Deep learning models often lack the interpretability and trustworthiness required for clinical use. This paper proposes a prototype-regularised training method to analyse 1,208 lung ultrasound images, focusing on B-line artefacts. A ConvNeXt- Tiny architecture is used, adding a novel reconstruction loss to the standard classification loss. The model is guided to extract meaningful prototypes and uses them to classify the ultrasound images. To prevent these constraints from hindering generalisation, it is used in pairs with the proposed reconstruction loss, a set of plausible data augmentation of the ideal researched prototypes, and a geometry-aware network, a spatial transformer network, to measure which solutions help the network towards outputting the most reliable outcomes. The resulting models are precise, lightweight and interpretable, indicating that the proposed solution can be embedded in an ultrasound device to assist healthcare specialists in point-of-care applications.Project(s): TiAssisto
Ignesti G., D’angelo G., Pratali L., Moroni D., Martinelli M.
Deep learning models often lack the interpretability and trustworthiness required for clinical use. This paper proposes a prototype-regularised training method to analyse 1,208 lung ultrasound images, focusing on B-line artefacts. A ConvNeXt- Tiny architecture is used, adding a novel reconstruction loss to the standard classification loss. The model is guided to extract meaningful prototypes and uses them to classify the ultrasound images. To prevent these constraints from hindering generalisation, it is used in pairs with the proposed reconstruction loss, a set of plausible data augmentation of the ideal researched prototypes, and a geometry-aware network, a spatial transformer network, to measure which solutions help the network towards outputting the most reliable outcomes. The resulting models are precise, lightweight and interpretable, indicating that the proposed solution can be embedded in an ultrasound device to assist healthcare specialists in point-of-care applications.Project(s): TiAssisto
2026
Conference article
Open Access
Fusing parametric and spatial constraints in continual learning
Ignesti Giacomo, D'Angelo Gennaro, Pratali Lorenza, Martinelli Massimo
The deployment of powerful foundation models in medical imaging is severely hampered by the fact that neural networks cannot learn sequentially from decentralised data with- out catastrophic forgetting. Early continual learning frameworks, such as Elastic Weight Consolidation (EWC), offer a parametric defence against forgetting but treat the network as a black box and neglect to preserve learned spatial representations. To solve this problem, we propose a novel continual learning framework that anchors EWC within a stable reconstruction-classification training paradigm using batch-trained prototypes. Our approach fundamentally constrains representation drift by enforcing spa- tial alignment between the network’s dynamically generated saliency maps and static visual class prototypes. Our approach combines weight consolidation with spatial reconstruction penal- ties to explicitly prevent the prior classification state-space from being distorted during novel task updates. Empirical evaluations show that the fusion not only retains high classification efficacy on historical tasks but also successfully maintains the network’s visual interpretability. Ultimately, this framework establishes a new direction for continual learning in medical imaging, shifting the focus from purely weight-based regularisation to the holistic preservation of both classifier stability and interpretable feature representations.
Ignesti Giacomo, D'Angelo Gennaro, Pratali Lorenza, Martinelli Massimo
The deployment of powerful foundation models in medical imaging is severely hampered by the fact that neural networks cannot learn sequentially from decentralised data with- out catastrophic forgetting. Early continual learning frameworks, such as Elastic Weight Consolidation (EWC), offer a parametric defence against forgetting but treat the network as a black box and neglect to preserve learned spatial representations. To solve this problem, we propose a novel continual learning framework that anchors EWC within a stable reconstruction-classification training paradigm using batch-trained prototypes. Our approach fundamentally constrains representation drift by enforcing spa- tial alignment between the network’s dynamically generated saliency maps and static visual class prototypes. Our approach combines weight consolidation with spatial reconstruction penal- ties to explicitly prevent the prior classification state-space from being distorted during novel task updates. Empirical evaluations show that the fusion not only retains high classification efficacy on historical tasks but also successfully maintains the network’s visual interpretability. Ultimately, this framework establishes a new direction for continual learning in medical imaging, shifting the focus from purely weight-based regularisation to the holistic preservation of both classifier stability and interpretable feature representations.
See at:
attend.ieee.org 
| CNR IRIS | CNR IRIS 
2025
Other
Open Access
SI-Lab Annual Research Report 2024
Awais Ch Muhammad, Baiamonte A., Benassi A., Berti A., Bertini G., Buongiorno R, Bulotta D., Cafiso M., Carboni A., Carloni G., Caudai C., Colantonio S., Conti F., Daoudagh S., Del Corso G., Fusco G., Galesi G., Germanese D., Gravili S., Ignesti G., Kuruoglu E. E., Lazzini G., Leone G. R., Leporini B., Magrini M., Martinelli M., Omrani Ali Reza, Pachetti E., Papini O., Paradisi P., Pardini F., Pascali M. A., Pieri G., Reggiannini M., Righi M., Salerno E., Salvetti O., Scozzari A., Sebastiani L., Straface S., Tampucci M., Tarabella L., Tonazzini A., Moroni D.
The Signal & Images Laboratory (SI-Lab) is an interdisciplinary research group in computer vision, signal analysis, intelligent vision systems and multimedia data understanding. It is part of the Institute of Information Science and Technologies (ISTI) of the National Research Council of Italy (CNR). This report accounts for the research activities of the Signal and Images Laboratory of the Institute of Information Science and Technologies during the year 2024.DOI: 10.32079/isti-ar-2025/002
Metrics:
Awais Ch Muhammad, Baiamonte A., Benassi A., Berti A., Bertini G., Buongiorno R, Bulotta D., Cafiso M., Carboni A., Carloni G., Caudai C., Colantonio S., Conti F., Daoudagh S., Del Corso G., Fusco G., Galesi G., Germanese D., Gravili S., Ignesti G., Kuruoglu E. E., Lazzini G., Leone G. R., Leporini B., Magrini M., Martinelli M., Omrani Ali Reza, Pachetti E., Papini O., Paradisi P., Pardini F., Pascali M. A., Pieri G., Reggiannini M., Righi M., Salerno E., Salvetti O., Scozzari A., Sebastiani L., Straface S., Tampucci M., Tarabella L., Tonazzini A., Moroni D.
The Signal & Images Laboratory (SI-Lab) is an interdisciplinary research group in computer vision, signal analysis, intelligent vision systems and multimedia data understanding. It is part of the Institute of Information Science and Technologies (ISTI) of the National Research Council of Italy (CNR). This report accounts for the research activities of the Signal and Images Laboratory of the Institute of Information Science and Technologies during the year 2024.DOI: 10.32079/isti-ar-2025/002
Metrics:
2025
Other
Open Access
Batch-CAM: introduction to better reasoning in convolutional deep learning models
Ignesti G., Moroni D., Martinelli M.
Understanding the inner workings of deep learning models is crucial for advancing artificial intelligence, particularly in high-stakes fields such as healthcare, where accurate explanations are as vital as precision. This paper introduces Batch- CAM, a novel training paradigm that fuses a batch implementation of the Grad- CAM algorithm with a prototypical reconstruction loss. This combination guides the model to focus on salient image features, thereby enhancing its performance across classification tasks. Our results demonstrate that Batch-CAM achieves a simultaneous improvement in accuracy and image reconstruction quality while reducing training and inference times. By ensuring models learn from evidence- relevant information, this approach makes a relevant contribution to building more transparent, explainable, and trustworthy AI systems.DOI: 10.48550/arxiv.2510.00664
Metrics:
Ignesti G., Moroni D., Martinelli M.
Understanding the inner workings of deep learning models is crucial for advancing artificial intelligence, particularly in high-stakes fields such as healthcare, where accurate explanations are as vital as precision. This paper introduces Batch- CAM, a novel training paradigm that fuses a batch implementation of the Grad- CAM algorithm with a prototypical reconstruction loss. This combination guides the model to focus on salient image features, thereby enhancing its performance across classification tasks. Our results demonstrate that Batch-CAM achieves a simultaneous improvement in accuracy and image reconstruction quality while reducing training and inference times. By ensuring models learn from evidence- relevant information, this approach makes a relevant contribution to building more transparent, explainable, and trustworthy AI systems.DOI: 10.48550/arxiv.2510.00664
Metrics:
2025
Conference article
Open Access
Leveraging AI for Signal and Image Analysis in Medicine and Health
Marco Cafiso, Andrea Carboni, Claudia Caudai, Sara Colantonio, Francesco Conti, Mario D’acunto, Said Daoudagh, Giulio Del Corso, Danila Germanese, Giacomo Ignesti, Gianmarco Lazzini, Giuseppe Riccardo Leone, Massimo Magrini, Davide Moroni, Francesca Pardini, Maria Antonietta Pascali, Paolo Paradisi, Federico Volpini
The integration of artificial intelligence (AI) into the medical domain is driving innovation and progress in healthcare. This paper summarizes the research activities that a multidisciplinary research group within the Signals and Images Lab of the Institute of Information Science and Technologies of the National Research Council of Italy is carrying out to explore the great potential of AI in several applications, e.g., in the analysis of biomedical data, and in the development of tools for enhancing trustworthiness and reliability of AI based systems. From cancer diagnosis and grading, to the analysis of body physiological signals to improve the understanding of dance movement therapy as an approach to healthy aging, this work highlights the paradigm shift that AI has brought into medicine and healthcare.Source: CEUR WORKSHOP PROCEEDINGS, vol. 4121. Trieste, June 23-24, 2025
Marco Cafiso, Andrea Carboni, Claudia Caudai, Sara Colantonio, Francesco Conti, Mario D’acunto, Said Daoudagh, Giulio Del Corso, Danila Germanese, Giacomo Ignesti, Gianmarco Lazzini, Giuseppe Riccardo Leone, Massimo Magrini, Davide Moroni, Francesca Pardini, Maria Antonietta Pascali, Paolo Paradisi, Federico Volpini
The integration of artificial intelligence (AI) into the medical domain is driving innovation and progress in healthcare. This paper summarizes the research activities that a multidisciplinary research group within the Signals and Images Lab of the Institute of Information Science and Technologies of the National Research Council of Italy is carrying out to explore the great potential of AI in several applications, e.g., in the analysis of biomedical data, and in the development of tools for enhancing trustworthiness and reliability of AI based systems. From cancer diagnosis and grading, to the analysis of body physiological signals to improve the understanding of dance movement therapy as an approach to healthy aging, this work highlights the paradigm shift that AI has brought into medicine and healthcare.Source: CEUR WORKSHOP PROCEEDINGS, vol. 4121. Trieste, June 23-24, 2025
See at:
ceur-ws.org | CNR IRIS | CNR IRIS
2025
Conference article
Open Access
You’ve got the wrong outfit: evaluating deep learning paradigms on digit and fashion recognition
Ignesti G., Martinelli M., Moroni D.
To build more accurate and trustworthy artificial intelligence algorithms in deep learning, it is essential to understand the mechanisms driving classification systems to identify their targets. Typically, post hoc methods are used to provide insights into this process. By contrast, in this work, we investigate the possibility of using class activation maps in combination with contrastive loss to enhance the reliability of train- ing of a deep learning model. MNIST and Fashion MNIST datasets are considered in our investigation since they have already proven a prac- tical starting point for assessing an almost tautologic training strategy for deep learning algorithms, given that classification targets are the pri- mary significant content of the images in these datasets. Starting from the raw comparison of accuracy and system complexity of the proposed approach, a further investigation of the technique’s feasibility in a deep learning study is conducted over six random seed splits of the training data and model performance. A modern deep learning network, such as ConvNeXT, determines whether a more robust architecture trained with the proposed mechanics provides better insights than a simple convolu- tional neural network. This investigation also addresses the importance of skip connections, structured learning layers, and feature map dimen- sions in the learning process.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 15616, pp. 202-215. Kolkata, India, 01/12/2024
DOI: 10.1007/978-3-031-87663-9
Metrics:
Ignesti G., Martinelli M., Moroni D.
To build more accurate and trustworthy artificial intelligence algorithms in deep learning, it is essential to understand the mechanisms driving classification systems to identify their targets. Typically, post hoc methods are used to provide insights into this process. By contrast, in this work, we investigate the possibility of using class activation maps in combination with contrastive loss to enhance the reliability of train- ing of a deep learning model. MNIST and Fashion MNIST datasets are considered in our investigation since they have already proven a prac- tical starting point for assessing an almost tautologic training strategy for deep learning algorithms, given that classification targets are the pri- mary significant content of the images in these datasets. Starting from the raw comparison of accuracy and system complexity of the proposed approach, a further investigation of the technique’s feasibility in a deep learning study is conducted over six random seed splits of the training data and model performance. A modern deep learning network, such as ConvNeXT, determines whether a more robust architecture trained with the proposed mechanics provides better insights than a simple convolu- tional neural network. This investigation also addresses the importance of skip connections, structured learning layers, and feature map dimen- sions in the learning process.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 15616, pp. 202-215. Kolkata, India, 01/12/2024
DOI: 10.1007/978-3-031-87663-9
Metrics:
See at:
CNR IRIS | link.springer.com | CNR IRIS | CNR IRIS
2025
Conference article
Open Access
Assessment of dance movement therapy outcomes: a preliminary proposal
Daoudagh S., Ignesti G., Moroni D., Sebastiani L., Paradisi P.
Context: Dance Movement Therapy (DMT) is a therapeutic modality that utilizes movement to promote holistic well-being. Current DMT assessment methods rely heavily on the subjective judgment of the DMT professional. Objective: Our research aims to develop a framework composed of different components with specific functionalities that can be integrated with the DMT modality to improve the objectivity and efficiency of DMT evaluations. Method: The DMT framework consists of an experimental protocol for data collection and a reference-supporting architecture that includes components for video analysis, physiological signal management, and evaluation tools. Artificial Intelligence (AI) based human pose estimation techniques are also employed to derive the DMT participants’ poses during the DMT sessions for more reliable movement analysis. Results: Our preliminary results consist of demonstrating the effectiveness of the AI-based pose estimation tool, YOLO-NAS-Pose, in accurately detecting participants’ poses. Conclusion: The proposed framework offers a promising approach to improving DMT practices by integrating and leveraging AI-based human pose estimation to evaluate participants’ movement in the DMT setting objectively. Future research will focus on refining the framework and developing user-friendly tools for widespread adoption in real DMT contexts.Source: COMMUNICATIONS IN COMPUTER AND INFORMATION SCIENCE, vol. 2371, pp. 382-395. Porto, Portogallo, 21-22/11/2024
DOI: 10.1007/978-3-031-83845-3_23
Project(s): Tuscany Health Ecosystem
Metrics:
Daoudagh S., Ignesti G., Moroni D., Sebastiani L., Paradisi P.
Context: Dance Movement Therapy (DMT) is a therapeutic modality that utilizes movement to promote holistic well-being. Current DMT assessment methods rely heavily on the subjective judgment of the DMT professional. Objective: Our research aims to develop a framework composed of different components with specific functionalities that can be integrated with the DMT modality to improve the objectivity and efficiency of DMT evaluations. Method: The DMT framework consists of an experimental protocol for data collection and a reference-supporting architecture that includes components for video analysis, physiological signal management, and evaluation tools. Artificial Intelligence (AI) based human pose estimation techniques are also employed to derive the DMT participants’ poses during the DMT sessions for more reliable movement analysis. Results: Our preliminary results consist of demonstrating the effectiveness of the AI-based pose estimation tool, YOLO-NAS-Pose, in accurately detecting participants’ poses. Conclusion: The proposed framework offers a promising approach to improving DMT practices by integrating and leveraging AI-based human pose estimation to evaluate participants’ movement in the DMT setting objectively. Future research will focus on refining the framework and developing user-friendly tools for widespread adoption in real DMT contexts.Source: COMMUNICATIONS IN COMPUTER AND INFORMATION SCIENCE, vol. 2371, pp. 382-395. Porto, Portogallo, 21-22/11/2024
DOI: 10.1007/978-3-031-83845-3_23
Project(s): Tuscany Health Ecosystem
Metrics:
See at:
CNR IRIS | link.springer.com | CNR IRIS | CNR IRIS
2025
Journal article
Open Access
You've got the wrong number: evaluating deep learning training paradigms using handwritten digit recognition data
Ignesti G., Martinelli M., Moroni D.
To build more accurate and trustworthy artificial intelligence algorithms in deep learning, it is essential to understand the mechanisms driving classification systems to identify their targets. Typically, post hoc methods provide insights into this process. In this preliminary work, we shift the reconstruction of the class activation map to the training phase to evaluate how the model's performance changes compared to standard classification approaches. The MNIST dataset and its variants, such as Fashion MNIST, consist of well-defined images that facilitate testing this type of training process. Specifically, the classification targets are the only significant content in the images, excluding the background, allowing for a direct comparison of the reconstruction against the input images. To enhance the guidance of the network, we introduce a contrastive loss term to complement the standard classification function, which often uses categorical cross-entropy. By comparing the accuracy and the extracted pattern of the standard approach with the proposed method, we can gain valuable insights into the network's learning process. This approach aims toimprove the interpretability and effectiveness of the model during training, ultimately leading to higher classification accuracy and reliability.Source: PATTERN RECOGNITION AND IMAGE ANALYSIS, vol. 34 (issue 4), pp. 978-983
DOI: 10.1134/s1054661824700974
Metrics:
Ignesti G., Martinelli M., Moroni D.
To build more accurate and trustworthy artificial intelligence algorithms in deep learning, it is essential to understand the mechanisms driving classification systems to identify their targets. Typically, post hoc methods provide insights into this process. In this preliminary work, we shift the reconstruction of the class activation map to the training phase to evaluate how the model's performance changes compared to standard classification approaches. The MNIST dataset and its variants, such as Fashion MNIST, consist of well-defined images that facilitate testing this type of training process. Specifically, the classification targets are the only significant content in the images, excluding the background, allowing for a direct comparison of the reconstruction against the input images. To enhance the guidance of the network, we introduce a contrastive loss term to complement the standard classification function, which often uses categorical cross-entropy. By comparing the accuracy and the extracted pattern of the standard approach with the proposed method, we can gain valuable insights into the network's learning process. This approach aims toimprove the interpretability and effectiveness of the model during training, ultimately leading to higher classification accuracy and reliability.Source: PATTERN RECOGNITION AND IMAGE ANALYSIS, vol. 34 (issue 4), pp. 978-983
DOI: 10.1134/s1054661824700974
Metrics:
See at:
CNR IRIS | link.springer.com | CNR IRIS | CNR IRIS
2024
Other
Restricted
BANDO RICERCA COVID 19 TOSCANA. Relazione tecnica finale
Lorenza Pratali, Michela Rial, Serena Cardinali, Jessica De Giovanni, Emma Buzzigoli, Luca Bastiani, Chiara Deri, Gennaro D'Angelo, Laura Roth, Ida Rebecca Borth, Antonella Tomei, Silvia Memmini, Alessandro Iala, Alessandro Dini, Franca Marzoli, Cesare Rivieri, Mirko Passera, Luca Serasini, Davide Moroni, Antonio Bruno, Ignesti Giacomo, Giulio Galesi, Francesca Pardini, Francesca Borri, Massimo Martinelli
Project(s): TiAssisto
Lorenza Pratali, Michela Rial, Serena Cardinali, Jessica De Giovanni, Emma Buzzigoli, Luca Bastiani, Chiara Deri, Gennaro D'Angelo, Laura Roth, Ida Rebecca Borth, Antonella Tomei, Silvia Memmini, Alessandro Iala, Alessandro Dini, Franca Marzoli, Cesare Rivieri, Mirko Passera, Luca Serasini, Davide Moroni, Antonio Bruno, Ignesti Giacomo, Giulio Galesi, Francesca Pardini, Francesca Borri, Massimo Martinelli
Project(s): TiAssisto
2024
Conference article
Open Access
Plant-traits: how citizen science and artificial intelligence can impact natural science
Ignesti G., Moroni D., Martinelli M.
Citizen science has emerged as a valuable resource for scientific research, providing large volumes of data for training deep learning models. However, the quality and accuracy of crowd-sourced data pose significant challenges for supervised learning tasks such as plant trait detection. This study investigates the application of AI techniques to address these issues within natural science. We explore the potential of multi-modal data analysis and ensemble methods to improve the accuracy of plant trait classification using citizen science data. Additionally, we examine the effectiveness of transfer learning from authoritative datasets like PlantVillage to enhance model performance on open- access platforms such as iNaturalist. By analysing the strengths and limitations of AI-driven approaches in this context, we aim to contribute to developing robust and reliable methods for utilising citizen science data in natural science.Source: ANNALS OF COMPUTER SCIENCE AND INFORMATION SYSTEMS, vol. 39, pp. 625-630. Belgrade, Serbia, 9-11/09/2024
DOI: 10.15439/2024f8703
Metrics:
Ignesti G., Moroni D., Martinelli M.
Citizen science has emerged as a valuable resource for scientific research, providing large volumes of data for training deep learning models. However, the quality and accuracy of crowd-sourced data pose significant challenges for supervised learning tasks such as plant trait detection. This study investigates the application of AI techniques to address these issues within natural science. We explore the potential of multi-modal data analysis and ensemble methods to improve the accuracy of plant trait classification using citizen science data. Additionally, we examine the effectiveness of transfer learning from authoritative datasets like PlantVillage to enhance model performance on open- access platforms such as iNaturalist. By analysing the strengths and limitations of AI-driven approaches in this context, we aim to contribute to developing robust and reliable methods for utilising citizen science data in natural science.Source: ANNALS OF COMPUTER SCIENCE AND INFORMATION SYSTEMS, vol. 39, pp. 625-630. Belgrade, Serbia, 9-11/09/2024
DOI: 10.15439/2024f8703
Metrics:
See at:
IRIS Cnr | annals-csis.org | Annals of computer science and information systems | IRIS Cnr | Archivio della Ricerca - Università di Pisa | CNR IRIS | CNR IRIS
2024
Other
Restricted
D6.2 – TiAssisto: Manuale di uso
Giacomo Ignesti, Chiara Deri, Gennaro D’angelo, Davide Cicalini, Davide Moroni, Lorenza Pratali, Massimo Martinelli
Project(s): TiAssisto
Giacomo Ignesti, Chiara Deri, Gennaro D’angelo, Davide Cicalini, Davide Moroni, Lorenza Pratali, Massimo Martinelli
Project(s): TiAssisto
2024
Other
Open Access
A comprehensive system supporting sustainable agricultural production from farm to fork
Carboni A., Galesi G., Ignesti G., Leone G. R., Magrini M., Martinelli M., Martino G., Moroni D., Pardini F., Scozzari A.
Poster presented at ISTI Day 2023-2024 edition on June 14 2024.DOI: 10.5281/zenodo.12168200
Metrics:
Carboni A., Galesi G., Ignesti G., Leone G. R., Magrini M., Martinelli M., Martino G., Moroni D., Pardini F., Scozzari A.
Poster presented at ISTI Day 2023-2024 edition on June 14 2024.DOI: 10.5281/zenodo.12168200
Metrics:
2024
Conference article
Open Access
Towards the actual deployment of robust, adaptable, and maintainable AI models for sustainable agriculture
Ignesti G., Moroni D., Martinelli M.
In the past two decades, computer vision and arti- ficial intelligence (AI) have made significant strides in delivering practical solutions to aid farmers directly in the fields, thereby contributing to the integration of advanced technology in pre- cision agriculture. However, extending these methods to diverse crops and broader applications, including low-resource situations, raises several concerns. Indeed, the adaptability of AI methods to new cases and domains is not always straightforward. Moreover, the dynamic global panorama requires a continuous adaptation and refinement of artificial intelligence models. In this position paper, we examine the current opportunities and challenges, and propose a new approach to address these issues, currently in the implementation phase at CNR-ISTI.Source: ACSIS PUBLICATIONS, vol. 40, pp. 33-39. Belgrade, Serbia, 9-11/09/2024
DOI: 10.15439/2024f2991
Metrics:
Ignesti G., Moroni D., Martinelli M.
In the past two decades, computer vision and arti- ficial intelligence (AI) have made significant strides in delivering practical solutions to aid farmers directly in the fields, thereby contributing to the integration of advanced technology in pre- cision agriculture. However, extending these methods to diverse crops and broader applications, including low-resource situations, raises several concerns. Indeed, the adaptability of AI methods to new cases and domains is not always straightforward. Moreover, the dynamic global panorama requires a continuous adaptation and refinement of artificial intelligence models. In this position paper, we examine the current opportunities and challenges, and propose a new approach to address these issues, currently in the implementation phase at CNR-ISTI.Source: ACSIS PUBLICATIONS, vol. 40, pp. 33-39. Belgrade, Serbia, 9-11/09/2024
DOI: 10.15439/2024f2991
Metrics:
See at:
annals-csis.org | Annals of computer science and information systems | IRIS Cnr | Archivio della Ricerca - Università di Pisa | CNR IRIS | CNR IRIS
2024
Other
Restricted
D4.2 (2) – TiAssisto: Implementazione dei servizi intelligenti di supporto alla decisione
Ignesti G., Deri C., D’angelo G., Moroni D., Pratali L., Martinelli M.
Il presente documento è il primo deliverable dell’Obiettivo Operativo 4 “Implementazione dei servizi intelligenti di supporto alla decisione (IDSS)” L'obiettivo è trasferire le conoscenze acquisite mediante i sistemi di intelligenza artificiale (IA) grazie alle fasi di apprendimento eseguite sui dati acquisiti dai dispositivi in un insieme di protocolli basati sui principi della evidence-based medicine in modo da fornire e un supporto decisionale fattibile e affidabile. È evidente che tutte queste funzionalità necessitano di servizi specifici per essere adeguatamente sfruttate, per la cui definizione sarà necessaria una forte collaborazione con la controparte clinica. In particolare si eseguiranno le seguenti sotto attività: - Trasformazione del flusso del nuovo protocollo di cura in un insieme organico e formalizzato di suggerimenti e di regole definite dai medici in una forma adeguata per l'implementazione nell’IDSS. Ciò sarà derivato dall'attività 2.1 e dalla conoscenza estratta mediante paradigmi di intelligenza artificiale dai dati raccolti. - Definizione e implementazione degli score e dei suggerimenti sul rischio.Project(s): TiAssisto
Ignesti G., Deri C., D’angelo G., Moroni D., Pratali L., Martinelli M.
Il presente documento è il primo deliverable dell’Obiettivo Operativo 4 “Implementazione dei servizi intelligenti di supporto alla decisione (IDSS)” L'obiettivo è trasferire le conoscenze acquisite mediante i sistemi di intelligenza artificiale (IA) grazie alle fasi di apprendimento eseguite sui dati acquisiti dai dispositivi in un insieme di protocolli basati sui principi della evidence-based medicine in modo da fornire e un supporto decisionale fattibile e affidabile. È evidente che tutte queste funzionalità necessitano di servizi specifici per essere adeguatamente sfruttate, per la cui definizione sarà necessaria una forte collaborazione con la controparte clinica. In particolare si eseguiranno le seguenti sotto attività: - Trasformazione del flusso del nuovo protocollo di cura in un insieme organico e formalizzato di suggerimenti e di regole definite dai medici in una forma adeguata per l'implementazione nell’IDSS. Ciò sarà derivato dall'attività 2.1 e dalla conoscenza estratta mediante paradigmi di intelligenza artificiale dai dati raccolti. - Definizione e implementazione degli score e dei suggerimenti sul rischio.Project(s): TiAssisto
2023
Journal article
Open Access
Efficient lung ultrasound classification
Bruno A, Ignesti G, Salvetti O, Moroni D, Martinelli M
A machine learning method for classifying Lung UltraSound is here proposed to provide a point of care tool for supporting a safe, fast and accurate diagnosis, that can also be useful during a pandemic like as SARS-CoV-2. Given the advantages (e.g. safety, rapidity, portability, cost-effectiveness) provided by the ultrasound technology over other methods (e.g. X-ray, computer tomography, magnetic resonance imaging), our method was validated on the largest LUS public dataset. Focusing on both accuracy and efficiency, our solution is based on an efficient adaptive ensembling of two EfficientNet-b0 models reaching 100% of accuracy, which, to our knowledge, outperforms the previous state-of-the-art. The complexity of this solution keeps the number of parameters in the same order as an EfficientNet-b0 by adopting specific design choices that are adaptive ensembling with a combination layer, ensembling performed on the deep features, minimal ensemble only two weak models. Moreover, a visual analysis of the saliency maps on sample images of all the classes of the dataset reveals where the focus is on an inaccurate weak model versus an accurate model.Source: BIOENGINEERING, vol. 10 (issue 5)
DOI: 10.3390/bioengineering10050555
Metrics:
Bruno A, Ignesti G, Salvetti O, Moroni D, Martinelli M
A machine learning method for classifying Lung UltraSound is here proposed to provide a point of care tool for supporting a safe, fast and accurate diagnosis, that can also be useful during a pandemic like as SARS-CoV-2. Given the advantages (e.g. safety, rapidity, portability, cost-effectiveness) provided by the ultrasound technology over other methods (e.g. X-ray, computer tomography, magnetic resonance imaging), our method was validated on the largest LUS public dataset. Focusing on both accuracy and efficiency, our solution is based on an efficient adaptive ensembling of two EfficientNet-b0 models reaching 100% of accuracy, which, to our knowledge, outperforms the previous state-of-the-art. The complexity of this solution keeps the number of parameters in the same order as an EfficientNet-b0 by adopting specific design choices that are adaptive ensembling with a combination layer, ensembling performed on the deep features, minimal ensemble only two weak models. Moreover, a visual analysis of the saliency maps on sample images of all the classes of the dataset reveals where the focus is on an inaccurate weak model versus an accurate model.Source: BIOENGINEERING, vol. 10 (issue 5)
DOI: 10.3390/bioengineering10050555
Metrics:
See at:
CNR IRIS | ISTI Repository | www.mdpi.com | CNR IRIS | CNR IRIS
2023
Other
Restricted
Artificial Intelligence in TiAssisto: first results
Ignesti G, Bruno A, Deri C, D'Angelo G, Salvetti O, Moroni D, Pratali L, Martinelli M
Artificial Intelligence (AI) is integrated into medical applications since its beginning. The advent of deep learning algorithms, powerful computation power and large datasets has made possible the development of numerous new medical applications. A significant part of these applications is focused on the classification or on the segmentation of medical images. In this paper we present an innovative solution for clinical images classification.
Ignesti G, Bruno A, Deri C, D'Angelo G, Salvetti O, Moroni D, Pratali L, Martinelli M
Artificial Intelligence (AI) is integrated into medical applications since its beginning. The advent of deep learning algorithms, powerful computation power and large datasets has made possible the development of numerous new medical applications. A significant part of these applications is focused on the classification or on the segmentation of medical images. In this paper we present an innovative solution for clinical images classification.
2023
Journal article
Open Access
Explaining ensemble models for lung ultrasound classification
Bruno A, Ignesti G, Martinelli M
Correct classification is the main aspect in evaluating the quality of an artificial intelligence system, but what happens when you reach top accuracy and no method explains how it works? In our study, we aim at addressing the black-box problem using an ad-hoc built classifier for lung ultrasound images.Source: ERCIM NEWS, vol. 134
Bruno A, Ignesti G, Martinelli M
Correct classification is the main aspect in evaluating the quality of an artificial intelligence system, but what happens when you reach top accuracy and no method explains how it works? In our study, we aim at addressing the black-box problem using an ad-hoc built classifier for lung ultrasound images.Source: ERCIM NEWS, vol. 134
See at:
ercim-news.ercim.eu | CNR IRIS | ISTI Repository | CNR IRIS
2023
Conference article
Open Access
Deep learning methods for point-of-care ultrasound examination
Ignesti G, Deri C, D'Angelo G, Pratali L, Bruno A, Benassi A, Salvetti O, Moroni D, Martinelli M
Point-of-care Test (POCT) is the delivery of medical care at or near the patient's bedside. Primarily employed in emergencies, where rapid diagnosis and treatment are critical, POCT is now being used in domestic telehealth solutions, as in the TiAssisto project, thanks to technological advances such as the development of portable and affordable devices, high-speed Internet connections, video conferencing, and Artificial Intelligence (AI). Ultrasound (US) images of internal organs and structures are valuable tools in POCT medicine since this examination is portable, quick, and cost-effective. USs can help diagnose different conditions, including heart problems, abdominal pain, and pneumonia. Deep learning algorithms have proven to be highly effective in image recognition, enabling physicians to make informed decisions on-site. This article presents and investigates a decision support system based on deep learning algorithms. The primary aim of this research is to detect various signs in US images using cutting-edge classification methods. The proposed pipeline initially adopts an optical character recognition (OCR) algorithm: this technique inspects and cleans the US image, ensuring privacy and better classification potential. The collected images are forwarded to a state-of-the-art (SOTA) deep learning network, a fine-tuned EfficientNET-b0, able to detect any signs potentially related to pathology in a rapid way.The network classification is then assessed in the pipeline using a visual explanation method, i.e. Grad-CAM, to evaluate if the proper medical signs were identified, offering a quick and effective second opinion.The involved physician's feedback remarks that this system can detect important signs in pulmonary US imaging, although the dataset is not yet the final one since the TiAssisto project is still ongoing, with a planned conclusion in February 2024. Our ultimate goal is not merely to develop a classification system but to create an effective healthcare support system that can be used beyond primary healthcare facilities.DOI: 10.1109/sitis61268.2023.00078
Metrics:
Ignesti G, Deri C, D'Angelo G, Pratali L, Bruno A, Benassi A, Salvetti O, Moroni D, Martinelli M
Point-of-care Test (POCT) is the delivery of medical care at or near the patient's bedside. Primarily employed in emergencies, where rapid diagnosis and treatment are critical, POCT is now being used in domestic telehealth solutions, as in the TiAssisto project, thanks to technological advances such as the development of portable and affordable devices, high-speed Internet connections, video conferencing, and Artificial Intelligence (AI). Ultrasound (US) images of internal organs and structures are valuable tools in POCT medicine since this examination is portable, quick, and cost-effective. USs can help diagnose different conditions, including heart problems, abdominal pain, and pneumonia. Deep learning algorithms have proven to be highly effective in image recognition, enabling physicians to make informed decisions on-site. This article presents and investigates a decision support system based on deep learning algorithms. The primary aim of this research is to detect various signs in US images using cutting-edge classification methods. The proposed pipeline initially adopts an optical character recognition (OCR) algorithm: this technique inspects and cleans the US image, ensuring privacy and better classification potential. The collected images are forwarded to a state-of-the-art (SOTA) deep learning network, a fine-tuned EfficientNET-b0, able to detect any signs potentially related to pathology in a rapid way.The network classification is then assessed in the pipeline using a visual explanation method, i.e. Grad-CAM, to evaluate if the proper medical signs were identified, offering a quick and effective second opinion.The involved physician's feedback remarks that this system can detect important signs in pulmonary US imaging, although the dataset is not yet the final one since the TiAssisto project is still ongoing, with a planned conclusion in February 2024. Our ultimate goal is not merely to develop a classification system but to create an effective healthcare support system that can be used beyond primary healthcare facilities.DOI: 10.1109/sitis61268.2023.00078
Metrics:
See at:
CNR IRIS | ieeexplore.ieee.org | ISTI Repository | CNR IRIS | CNR IRIS