Annaclaudia Montanino
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Annaclaudia Montanino has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Svein Kleiven
Division of Neuronic Engineering, KTH Royal Institute of Technology, Stockholm, Sweden
4 shared publications - Alessandra Villa
Philochem AG, Libernstrasse 3, 8112 Otelfingen, Switzerland
3 shared publications - Ioanna Miliou
Department of Computer and Systems Sciences, Stockholm University, Stockholm, Sweden
2 shared publications - Lena Mondrejevski
Stockholm University,Dept. of Computer and Systems Sciences,Stockholm,Sweden
2 shared publications - Panagiotis Papapetrou
Department of Computer and Systems Sciences, Stockholm University, Stockholm, Sweden
2 shared publications Svein Kleiven
2 shared publications
Latest Publications
- 12 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 819-826
X4FL: Explainability for Federated Learning in ICU Mortality Prediction
Federated Learning (FL) enables collaborative Machine Learning (ML) model training across institutions without sharing sensitive data, a critical advantage in healthcare. However, existing FL approaches for Intensive Care Unit (ICU) mortality prediction largely overlook explainability, which is essential for clinician trust. In this study, we present X4FL, a novel framework that integrates intrinsic and post-hoc interpretability into FL for ICU mortality prediction using the MIMIC-III dataset. X4FL employs an attention-based Long Short-Term Memory (LSTM) to capture temporal relevance at the feature level and combines it with SHapley Additive exPlanations (SHAP) to quantify feature importance. This dual explainability design yields clinically meaningful insights by highlighting both influential time segments and predictive features. Under a realistic non-Independent and Identically Distributed (non-IID) federated setup reflecting heterogeneous ICU populations, X4FL achieves predictive performance comparable to centralized models while consistently outperforming locally trained models, particularly in smaller or specialized ICUs. Attention heatmaps capture clinically relevant temporal patterns, and SHAP analyses reveal robust feature attributions across distributed settings. By addressing the trade-off between performance and interpretability, X4FL enhances transparency and trust in FL-based ICU mortality prediction, offering a step toward clinically viable federated solutions in critical care.
- 12 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 819-826
X4FL: Explainability for Federated Learning in ICU Mortality Prediction
Federated Learning (FL) enables collaborative Machine Learning (ML) model training across institutions without sharing sensitive data, a critical advantage in healthcare. However, existing FL approaches for Intensive Care Unit (ICU) mortality prediction largely overlook explainability, which is essential for clinician trust. In this study, we present X4FL, a novel framework that integrates intrinsic and post-hoc interpretability into FL for ICU mortality prediction using the MIMIC-III dataset. X4FL employs an attention-based Long Short-Term Memory (LSTM) to capture temporal relevance at the feature level and combines it with SHapley Additive exPlanations (SHAP) to quantify feature importance. This dual explainability design yields clinically meaningful insights by highlighting both influential time segments and predictive features. Under a realistic non-Independent and Identically Distributed (non-IID) federated setup reflecting heterogeneous ICU populations, X4FL achieves predictive performance comparable to centralized models while consistently outperforming locally trained models, particularly in smaller or specialized ICUs. Attention heatmaps capture clinically relevant temporal patterns, and SHAP analyses reveal robust feature attributions across distributed settings. By addressing the trade-off between performance and interpretability, X4FL enhances transparency and trust in FL-based ICU mortality prediction, offering a step toward clinically viable federated solutions in critical care.
- Open Access
FLICU: A Federated Learning Workflow for Intensive Care Unit Mortality Prediction
Although Machine Learning (ML) can be seen as a promising tool to improve clinical decision-making for supporting the improvement of medication plans, clinical procedures, diagnoses, or medication prescriptions, it remains limited by access to healthcare data. Healthcare data is sensitive, requiring strict privacy practices, and typically stored in data silos, making traditional machine learning challenging. Federated learning can counteract those limitations by training machine learning models over data silos while keeping the sensitive data localized. This study proposes a federated learning workflow for ICU mortality prediction. Hereby, the applicability of federated learning as an alternative to centralized machine learning and local machine learning is investigated by introducing federated learning to the binary classification problem of predicting ICU mortality. We extract multivariate time series data from the MIMIC-III database (lab values and vital signs), and benchmark the predictive performance of four deep sequential classifiers (FRNN, LSTM, GRU, and 1DCNN) varying the patient history window lengths (8h, 16h, 24h, 48h) and the number of FL clients (2, 4, 8). The experiments demonstrate that both centralized machine learning and federated learning are comparable in terms of AUPRC and F1-score. Furthermore, the federated approach shows superior performance over local machine learning. Thus, the federated approach can be seen as a valid and privacy-preserving alternative to centralized machine learning for classifying ICU mortality when sharing sensitive patient data between hospitals is not possible.
Open AccessFLICU: A Federated Learning Workflow for Intensive Care Unit Mortality Prediction
Although Machine Learning (ML) can be seen as a promising tool to improve clinical decision-making for supporting the improvement of medication plans, clinical procedures, diagnoses, or medication prescriptions, it remains limited by access to healthcare data. Healthcare data is sensitive, requiring strict privacy practices, and typically stored in data silos, making traditional machine learning challenging. Federated learning can counteract those limitations by training machine learning models over data silos while keeping the sensitive data localized. This study proposes a federated learning workflow for ICU mortality prediction. Hereby, the applicability of federated learning as an alternative to centralized machine learning and local machine learning is investigated by introducing federated learning to the binary classification problem of predicting ICU mortality. We extract multivariate time series data from the MIMIC-III database (lab values and vital signs), and benchmark the predictive performance of four deep sequential classifiers (FRNN, LSTM, GRU, and 1DCNN) varying the patient history window lengths (8h, 16h, 24h, 48h) and the number of FL clients (2, 4, 8). The experiments demonstrate that both centralized machine learning and federated learning are comparable in terms of AUPRC and F1-score. Furthermore, the federated approach shows superior performance over local machine learning. Thus, the federated approach can be seen as a valid and privacy-preserving alternative to centralized machine learning for classifying ICU mortality when sharing sensitive patient data between hospitals is not possible.
- Open Access
- 1 January 2021
- Published by Elsevier in Brain Multiphysics
- Vol. 2
Subject-specific multiscale analysis of concussion: from macroscopic loads to molecular-level damage
- 1 January 2021
- Published by Elsevier in Brain Multiphysics
- Vol. 2
Open AccessSubject-specific multiscale analysis of concussion: from macroscopic loads to molecular-level damage
- Open Access
- 30 January 2020
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 11, 25
Localized Axolemma Deformations Suggest Mechanoporation as Axonal Injury Trigger
Traumatic brain injuries are a leading cause of morbidity and mortality worldwide. With almost 50% of traumatic brain injuries being related to axonal damage, understanding the nature of cellular level impairment is crucial. Experimental observations have so far led to the formulation of conflicting theories regarding the cellular primary injury mechanism. Disruption of the axolemma, or alternatively cytoskeletal damage has been suggested mainly as injury trigger. However, mechanoporation thresholds of generic membranes seem not to overlap with the axonal injury deformation range and microtubules appear too stiff and too weakly connected to undergo mechanical breaking. Here, we aim to shed a light on the mechanism of primary axonal injury, bridging finite element and molecular dynamics simulations. Despite the necessary level of approximation, our models can accurately describe the mechanical behavior of the unmyelinated axon and its membrane. More importantly, they give access to quantities that would be inaccessible with an experimental approach. We show that in a typical injury scenario, the axonal cortex sustains deformations large enough to entail pore formation in the adjoining lipid bilayer. The observed axonal deformation of 10–12% agree well with the thresholds proposed in the literature for axonal injury and, above all, allow us to provide quantitative evidences that do not exclude pore formation in the membrane as a result of trauma. Our findings bring to an increased knowledge of axonal injury mechanism that will have positive implications for the prevention and treatment of brain injuries.
- 30 January 2020
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 11, 25
Open AccessLocalized Axolemma Deformations Suggest Mechanoporation as Axonal Injury Trigger
Traumatic brain injuries are a leading cause of morbidity and mortality worldwide. With almost 50% of traumatic brain injuries being related to axonal damage, understanding the nature of cellular level impairment is crucial. Experimental observations have so far led to the formulation of conflicting theories regarding the cellular primary injury mechanism. Disruption of the axolemma, or alternatively cytoskeletal damage has been suggested mainly as injury trigger. However, mechanoporation thresholds of generic membranes seem not to overlap with the axonal injury deformation range and microtubules appear too stiff and too weakly connected to undergo mechanical breaking. Here, we aim to shed a light on the mechanism of primary axonal injury, bridging finite element and molecular dynamics simulations. Despite the necessary level of approximation, our models can accurately describe the mechanical behavior of the unmyelinated axon and its membrane. More importantly, they give access to quantities that would be inaccessible with an experimental approach. We show that in a typical injury scenario, the axonal cortex sustains deformations large enough to entail pore formation in the adjoining lipid bilayer. The observed axonal deformation of 10–12% agree well with the thresholds proposed in the literature for axonal injury and, above all, allow us to provide quantitative evidences that do not exclude pore formation in the membrane as a result of trauma. Our findings bring to an increased knowledge of axonal injury mechanism that will have positive implications for the prevention and treatment of brain injuries.
- Open Access
Localized axolemma deformations suggest mechanoporation as axonal injury trigger
Traumatic brain injuries are a leading cause of morbidity and mortality worldwide. With almost 50% of traumatic brain injuries being related to axonal damage, understanding the nature of cellular level impairment is crucial. Experimental observations have so far led to the formulation of conflicting theories regarding the cellular primary injury mechanism. Disruption of the axolemma, or alternatively cytoskeletal damage has been suggested mainly as injury trigger. However, mechanoporation thresholds of generic membranes seem not to overlap with the axonal injury deformation range and microtubules appear too stiff and too weakly connected to undergo mechanical breaking. Here, we aim to shed a light on the mechanism of primary axonal injury, bridging finite element and molecular dynamics simulations. Despite the necessary level of approximation, our models can accurately describe the mechanical behavior of the unmyelinated axon and its membrane. More importantly, they give access to quantities that would be inaccessible with an experimental approach. We show that in a typical injury scenario, the axonal cortex sustains deformations large enough to entail pore formation in the adjoining lipid bilayer. The observed axonal deformation of 10-12% agree well with the thresholds proposed in the literature for axonal injury and, above all, allow us to provide quantitative evidences that do not exclude pore formation in the membrane as a result of trauma. Our findings bring to an increased knowledge of axonal injury mechanism that will have positive implications for the prevention and treatment of brain injuries.
Open AccessLocalized axolemma deformations suggest mechanoporation as axonal injury trigger
Traumatic brain injuries are a leading cause of morbidity and mortality worldwide. With almost 50% of traumatic brain injuries being related to axonal damage, understanding the nature of cellular level impairment is crucial. Experimental observations have so far led to the formulation of conflicting theories regarding the cellular primary injury mechanism. Disruption of the axolemma, or alternatively cytoskeletal damage has been suggested mainly as injury trigger. However, mechanoporation thresholds of generic membranes seem not to overlap with the axonal injury deformation range and microtubules appear too stiff and too weakly connected to undergo mechanical breaking. Here, we aim to shed a light on the mechanism of primary axonal injury, bridging finite element and molecular dynamics simulations. Despite the necessary level of approximation, our models can accurately describe the mechanical behavior of the unmyelinated axon and its membrane. More importantly, they give access to quantities that would be inaccessible with an experimental approach. We show that in a typical injury scenario, the axonal cortex sustains deformations large enough to entail pore formation in the adjoining lipid bilayer. The observed axonal deformation of 10-12% agree well with the thresholds proposed in the literature for axonal injury and, above all, allow us to provide quantitative evidences that do not exclude pore formation in the membrane as a result of trauma. Our findings bring to an increased knowledge of axonal injury mechanism that will have positive implications for the prevention and treatment of brain injuries.
- 22 October 2019
- Published by ASME International in Journal of Biomechanical Engineering
- Vol. 141 (12)
Axons Embedded in a Tissue May Withstand Larger Deformations Than Isolated Axons Before Mechanoporation Occurs
Diffuse axonal injury (DAI) is the pathological consequence of traumatic brain injury (TBI) that most of all requires a multiscale approach in order to be, first, understood and then possibly prevented. While in fact the mechanical insult usually happens at the head (or macro) level, the consequences affect structures at the cellular (or microlevel). The quest for axonal injury tolerances has so far been addressed both with experimental and computational approaches. On one hand, the experimental approach presents challenges connected to both temporal and spatial resolution in the identification of a clear axonal injury trigger after the application of a mechanical load. On the other hand, computational approaches usually consider axons as homogeneous entities and therefore are unable to make inferences about their viability, which is thought to depend on subcellular damages. Here, we propose a computational multiscale approach to investigate the onset of axonal injury in two typical experimental scenarios. We simulated single-cell and tissue stretch injury using a composite finite element axonal model in isolation and embedded in a matrix, respectively. Inferences on axonal damage are based on the comparison between axolemma strains and previously established mechanoporation thresholds. Our results show that, axons embedded in a tissue could withstand higher deformations than isolated axons before mechanoporation occurred and this is exacerbated by the increase in strain rate from 1/s to 10/s.
- 22 October 2019
- Published by ASME International in Journal of Biomechanical Engineering
- Vol. 141 (12)
Axons Embedded in a Tissue May Withstand Larger Deformations Than Isolated Axons Before Mechanoporation Occurs
Diffuse axonal injury (DAI) is the pathological consequence of traumatic brain injury (TBI) that most of all requires a multiscale approach in order to be, first, understood and then possibly prevented. While in fact the mechanical insult usually happens at the head (or macro) level, the consequences affect structures at the cellular (or microlevel). The quest for axonal injury tolerances has so far been addressed both with experimental and computational approaches. On one hand, the experimental approach presents challenges connected to both temporal and spatial resolution in the identification of a clear axonal injury trigger after the application of a mechanical load. On the other hand, computational approaches usually consider axons as homogeneous entities and therefore are unable to make inferences about their viability, which is thought to depend on subcellular damages. Here, we propose a computational multiscale approach to investigate the onset of axonal injury in two typical experimental scenarios. We simulated single-cell and tissue stretch injury using a composite finite element axonal model in isolation and embedded in a matrix, respectively. Inferences on axonal damage are based on the comparison between axolemma strains and previously established mechanoporation thresholds. Our results show that, axons embedded in a tissue could withstand higher deformations than isolated axons before mechanoporation occurred and this is exacerbated by the increase in strain rate from 1/s to 10/s.
- Open Access
- 20 June 2019
- Published by Springer Nature in Scientific Reports
- Vol. 9 (1) , 1-10
Mechanical characterization of squid giant axon membrane sheath and influence of the collagenous endoneurium on its properties
To understand traumas to the nervous system, the relation between mechanical load and functional impairment needs to be explained. Cellular-level computational models are being used to capture the mechanism behind mechanically-induced injuries and possibly predict these events. However, uncertainties in the material properties used in computational models undermine the validity of their predictions. For this reason, in this study the squid giant axon was used as a model to provide a description of the axonal mechanical behavior in a large strain and high strain rate regime $$({\boldsymbol{\varepsilon }}={\bf{10}}{\boldsymbol{ \% }},\,\mathop{{\boldsymbol{\varepsilon }}}\limits^{\cdot }={\bf{1}}{{\boldsymbol{s}}}^{-{\bf{1}}})$$ , which is relevant for injury investigations. More importantly, squid giant axon membrane sheaths were isolated and tested under dynamic uniaxial tension and relaxation. From the lumen outward, the membrane sheath presents: an axolemma, a layer of Schwann cells followed by the basement membrane and a prominent layer of loose connective tissue consisting of fibroblasts and collagen. Our results highlight the load-bearing role of this enwrapping structure and provide a constitutive description that could in turn be used in computational models. Furthermore, tests performed on collagen-depleted membrane sheaths reveal both the substantial contribution of the endoneurium to the total sheath’s response and an interesting increase in material nonlinearity when the collagen in this connective layer is digested. All in all, our results provide useful insights for modelling the axonal mechanical response and in turn will lead to a better understanding of the relationship between mechanical insult and electrophysiological outcome.- 20 June 2019
- Published by Springer Nature in Scientific Reports
- Vol. 9 (1) , 1-10
Open AccessMechanical characterization of squid giant axon membrane sheath and influence of the collagenous endoneurium on its properties
To understand traumas to the nervous system, the relation between mechanical load and functional impairment needs to be explained. Cellular-level computational models are being used to capture the mechanism behind mechanically-induced injuries and possibly predict these events. However, uncertainties in the material properties used in computational models undermine the validity of their predictions. For this reason, in this study the squid giant axon was used as a model to provide a description of the axonal mechanical behavior in a large strain and high strain rate regime $$({\boldsymbol{\varepsilon }}={\bf{10}}{\boldsymbol{ \% }},\,\mathop{{\boldsymbol{\varepsilon }}}\limits^{\cdot }={\bf{1}}{{\boldsymbol{s}}}^{-{\bf{1}}})$$ , which is relevant for injury investigations. More importantly, squid giant axon membrane sheaths were isolated and tested under dynamic uniaxial tension and relaxation. From the lumen outward, the membrane sheath presents: an axolemma, a layer of Schwann cells followed by the basement membrane and a prominent layer of loose connective tissue consisting of fibroblasts and collagen. Our results highlight the load-bearing role of this enwrapping structure and provide a constitutive description that could in turn be used in computational models. Furthermore, tests performed on collagen-depleted membrane sheaths reveal both the substantial contribution of the endoneurium to the total sheath’s response and an interesting increase in material nonlinearity when the collagen in this connective layer is digested. All in all, our results provide useful insights for modelling the axonal mechanical response and in turn will lead to a better understanding of the relationship between mechanical insult and electrophysiological outcome. - Open Access
- 6 August 2018
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 9, 643
Utilizing a Structural Mechanics Approach to Assess the Primary Effects of Injury Loads Onto the Axon and Its Components
Diffuse axonal injury (DAI) occurs as a result of the transmission of rapid dynamic loads from the head to the brain and specifically to its neurons. Despite being one of the most common and most deleterious types of traumatic brain injury (TBI), the inherent cell injury mechanism is yet to be understood. Experimental observations have led to the formulation of different hypotheses, such as mechanoporation of the axolemma and microtubules (MTs) breakage. With the goal of singling out the mechanical aspect of DAI and to resolve the ambiguity behind its injury mechanism, a composite finite element (FE) model of a representative volume of an axon was developed. Once calibrated and validated against published experimental data, the axonal model was used to simulate injury scenarios. The resulting strain distributions along its components were then studied in dependence of strain rate and of typical modeling choices such as the applied MT constraints and tau proteins failure. Results show that oversimplifying the MT bundle kinematic entails a systematic attenuation (cf = 2.33) of the computed maximum MT strain. Nevertheless, altogether, results support the hypothesis of axolemma mechanoporation as a cell-injury trigger. Moreover, for the first time the interconnection between the axolemma and the MT bundle is shown to play a role in damage localization. The proposed FE axonal model is a valuable tool to understand the axonal injury mechanism from a mechanical perspective and could be used in turn for the definition of well-informed injury criteria at the tissue and organ level.
- 6 August 2018
- Published by Frontiers Media SA in Frontiers in Neurology
- Vol. 9, 643
Open AccessUtilizing a Structural Mechanics Approach to Assess the Primary Effects of Injury Loads Onto the Axon and Its Components
Diffuse axonal injury (DAI) occurs as a result of the transmission of rapid dynamic loads from the head to the brain and specifically to its neurons. Despite being one of the most common and most deleterious types of traumatic brain injury (TBI), the inherent cell injury mechanism is yet to be understood. Experimental observations have led to the formulation of different hypotheses, such as mechanoporation of the axolemma and microtubules (MTs) breakage. With the goal of singling out the mechanical aspect of DAI and to resolve the ambiguity behind its injury mechanism, a composite finite element (FE) model of a representative volume of an axon was developed. Once calibrated and validated against published experimental data, the axonal model was used to simulate injury scenarios. The resulting strain distributions along its components were then studied in dependence of strain rate and of typical modeling choices such as the applied MT constraints and tau proteins failure. Results show that oversimplifying the MT bundle kinematic entails a systematic attenuation (cf = 2.33) of the computed maximum MT strain. Nevertheless, altogether, results support the hypothesis of axolemma mechanoporation as a cell-injury trigger. Moreover, for the first time the interconnection between the axolemma and the MT bundle is shown to play a role in damage localization. The proposed FE axonal model is a valuable tool to understand the axonal injury mechanism from a mechanical perspective and could be used in turn for the definition of well-informed injury criteria at the tissue and organ level.