Pasar al contenido principal

APRENDIZAJE PROFUNDO EXPLICABLE EN EL ANALISIS DE IMAGENES MEDICAS: NUEVAS METODOLOGIAS, ESTRATEGIAS DE EVALUACION Y APLICACIONES CLINICAS

PID2023-146261OB-I00

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para Impulsar la Investigación Científico-Técnica y su Transferencia
Subprograma Subprograma Estatal de Generación de Conocimiento
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2023
Plan Estatal 2021-2023 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSITAT POLITECNICA DE CATALUNYA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 6
Encontrada(s) 1 página(s)

UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

Neural network pruning using nonlinear oblique subspace projections

  • Pena-Campos, Johan
  • Caicedo Dorado, Alexander
  • Patiño Guevara, Diego
  • Ocampo-Martínez, Carlos|||0000-0001-9251-6044
© 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., The increasing complexity of deep neural networks often leads to overparameterization, limiting their deployment in resource-constrained environments. This paper introduces a novel pruning methodology based on Nonlinear Oblique Subspace Projections (NObSP), an algorithm originally developed to enhance the interpretability of kernel regression models. In this work, NObSP is extended to neural networks to estimate the contribution of individual neurons and entire layers to the network's output. This is achieved by quantifying the nonlinear transformations performed by the network on its inputs and internal representations. By computing the magnitude of these variables, neurons contributing minimally to the network's output are identified. Specifically, neurons with a nonlinear effect magnitude falling below a defined threshold are pruned, effectively removing them from the network. Consequently, NObSP provides a data-driven, interpretability-based approach to neural network pruning, leading to simplified and efficient models with minor performance impact. This method offers a clear framework for understanding which components of a neural network are critical for its function, facilitating the development of more compact and interpretable deep learning models. Finally, the efficacy of NObSP for pruning is demonstrated in CIFAR-10 and MNIST classification problems, where it is compared to traditional pruning algorithms based on the Ln norm., The work of this paper is supported by the following funds: the Condonable Credit Doctoral Support Grant from the Academic Vice Rector’s Office, the Scholarship Loan Program Grant from Colfuturo, the Spanish project SEAMLESS: Sustainable learning-based Management of Multiresource Large-scale Systems (ref. PID2023-148840OBI00), funded by MCIN/AEI/10.13039/501100011033/FEDER, UE and the project PID2023-146261OB-I00, funded by Ministry of Science and Innovation of Spain., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

MDS-AN: An attention-based explainable model for detecting myelodysplastic syndrome using hematological variables

  • Barrera Llanga, Kevin Iván|||0000-0002-1072-6142
  • Merino González, Anna|||0000-0002-1889-8889
  • Díaz Beyá, Marina|||0000-0001-9624-2597
  • Molina Borrás, Ángel
  • Rodellar Benedé, José|||0000-0002-1514-7713
Background and objectives: This study aims to develop and evaluate a deep learning model, which is based on an attention architecture for automatic detection of Myelodysplastic Syndromes (MDS), using age, sex and numerical values of 19 variables obtained from peripheral blood samples on a hematologic analyzer used clinical laboratory workflow. Methods: A dataset was collected from a total of 223 patients, 121 with MDS and 102 without this condition. The dataset was divided into training, validation, and test sets to train and validate the system. The classification model was developed from scratch using a attention-based architecture specifically designed to handle 21 input variables. In addition, a separate final evaluation was performed with a new group of 150 patients (56 with MDS and 94 with non-MDS) to determine a range of uncertainty and performance metrics. Results: On the test set, the model reached an accuracy of 97.5%, with 95% sensitivity, 95% specificity, 100% precision, and an F1-score of 0.97 for patients with MDS. In the independent evaluation, the model obtained 94.34% sensitivity, 95.60% specificity, 92.59% precision, and an F1-score of 0.935 after applying the uncertainty interval. The explainability analysis, based on Integrated Gradients, SHAP, pairwise interaction analysis, and internal attention mechanisms, consistently highlighted hematological variables associated with MDS, supporting the clinical interpretability of the model’s decisions. Conclusions: The proposed approach enables the automated differentiation of patients with MDS from those with non-MDS. In addition, the integration of explainability techniques strengthens the potential use of the system as a support tool in the clinical recognition of MDS., This work is part of two research projects funded by the Ministry of Science and Innovation of Spain : Computational system for diagnosis of acute leukemia and lymphoma from peripheral blood images: proof of concept and roadmap for technological valorization, with reference PDC2022-133514-I00, and Explainable deep learning in medical image analysis: new methodologies, evaluation strategies, and clinical applications, with reference PID2023-146261OB-I00. K. Barrera appreciates the financial support of the Secretary of Higher Education, Science, Technology and Innovation of Ecuador as a personal sponsor entity., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Generative artificial intelligence in peripheral blood morphology: synthetic image creation and digital stain normalization

  • Merino González, Anna|||0000-0002-1889-8889
  • Barrera Llanga, Kevin Iván|||0000-0002-1072-6142
  • Alférez Baquero, Edwin Santiago|||0000-0001-8661-1096
  • Rodellar Benedé, José|||0000-0002-1514-7713
The morphological analysis of peripheral blood cells is increasingly adopting deep learning, mainly for the automatic recognition of the variety of normal and abnormal cell classes. However, its clinical application is limited by the scarcity of annotated datasets for rare diseases and the high variability of staining protocols between laboratories. This paper provides an overview of how Generative Artificial Intelligence methods can help overcome these obstacles. Firsts, some basic concepts are presented, distinguishing between discriminative AI and generative AI, specifically generative adversarial networks (GANs) and diffusion models. The core of the paper addresses two problems: (1) the automatic generation of artificial blood cell images and (2) the digital artificial staining to reduce inter-laboratory differences. In these two cases, three sections are included: underlying concepts, a literature review and practical examples. These show how high-quality images of blood cells with realistic morphological characteristics are created, which strengthen the classifier's training. In multicenter tests, the examples illustrate that normalizing the staining allows a classifier trained on abnormal blood cell images from a single hospital to accurately recognize abnormal cells obtained in other laboratories, significantly improving performance without distorting cell morphology. A final section concludes the paper with observations and future perspectives on how generative tools can assist clinical pathologists as decision support systems that can operate consistently across diverse clinical settings., This work is part of the research project funded by the Ministry of Science and Innovation of Spain: Explainable deep learning in medical image analysis: new methodologies, evaluation strategies, and clinical applications, with reference PID2023-146261OB-I00., Peer Reviewed, 9 - Indústria, Innovació i Infraestructura, 3 - Salut i Benestar
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

MDS-AN: An attention-based explainable model for detecting myelodysplastic syndrome using hematological variables

  • Barrera Llanga, Kevin Iván
  • Merino González, Anna
  • Díaz Beyá, Marina
  • Molina Borrás, Ángel
  • Rodellar Benedé, José
Background and objectives: This study aims to develop and evaluate a deep learning model, which is based on an attention architecture for automatic detection of Myelodysplastic Syndromes (MDS), using age, sex and numerical values of 19 variables obtained from peripheral blood samples on a hematologic analyzer used clinical laboratory workflow. Methods: A dataset was collected from a total of 223 patients, 121 with MDS and 102 without this condition. The dataset was divided into training, validation, and test sets to train and validate the system. The classification model was developed from scratch using a attention-based architecture specifically designed to handle 21 input variables. In addition, a separate final evaluation was performed with a new group of 150 patients (56 with MDS and 94 with non-MDS) to determine a range of uncertainty and performance metrics. Results: On the test set, the model reached an accuracy of 97.5%, with 95% sensitivity, 95% specificity, 100% precision, and an F1-score of 0.97 for patients with MDS. In the independent evaluation, the model obtained 94.34% sensitivity, 95.60% specificity, 92.59% precision, and an F1-score of 0.935 after applying the uncertainty interval. The explainability analysis, based on Integrated Gradients, SHAP, pairwise interaction analysis, and internal attention mechanisms, consistently highlighted hematological variables associated with MDS, supporting the clinical interpretability of the model’s decisions. Conclusions: The proposed approach enables the automated differentiation of patients with MDS from those with non-MDS. In addition, the integration of explainability techniques strengthens the potential use of the system as a support tool in the clinical recognition of MDS., This work is part of two research projects funded by the Ministry of Science and Innovation of Spain : Computational system for diagnosis of acute leukemia and lymphoma from peripheral blood images: proof of concept and roadmap for technological valorization, with reference PDC2022-133514-I00, and Explainable deep learning in medical image analysis: new methodologies, evaluation strategies, and clinical applications, with reference PID2023-146261OB-I00. K. Barrera appreciates the financial support of the Secretary of Higher Education, Science, Technology and Innovation of Ecuador as a personal sponsor entity., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Generative artificial intelligence in peripheral blood morphology: synthetic image creation and digital stain normalization

  • Merino González, Anna
  • Barrera Llanga, Kevin Iván
  • Alférez Baquero, Edwin Santiago
  • Rodellar Benedé, José
The morphological analysis of peripheral blood cells is increasingly adopting deep learning, mainly for the automatic recognition of the variety of normal and abnormal cell classes. However, its clinical application is limited by the scarcity of annotated datasets for rare diseases and the high variability of staining protocols between laboratories. This paper provides an overview of how Generative Artificial Intelligence methods can help overcome these obstacles. Firsts, some basic concepts are presented, distinguishing between discriminative AI and generative AI, specifically generative adversarial networks (GANs) and diffusion models. The core of the paper addresses two problems: (1) the automatic generation of artificial blood cell images and (2) the digital artificial staining to reduce inter-laboratory differences. In these two cases, three sections are included: underlying concepts, a literature review and practical examples. These show how high-quality images of blood cells with realistic morphological characteristics are created, which strengthen the classifier's training. In multicenter tests, the examples illustrate that normalizing the staining allows a classifier trained on abnormal blood cell images from a single hospital to accurately recognize abnormal cells obtained in other laboratories, significantly improving performance without distorting cell morphology. A final section concludes the paper with observations and future perspectives on how generative tools can assist clinical pathologists as decision support systems that can operate consistently across diverse clinical settings., This work is part of the research project funded by the Ministry of Science and Innovation of Spain: Explainable deep learning in medical image analysis: new methodologies, evaluation strategies, and clinical applications, with reference PID2023-146261OB-I00., Peer Reviewed, 9 - Indústria, Innovació i Infraestructura, 3 - Salut i Benestar, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Contribución de congreso (ConferenceOutput)

Neural network pruning using nonlinear oblique subspace projections

  • Pena-Campos, Johan
  • Caicedo Dorado, Alexander
  • Patiño Guevara, Diego
  • Ocampo-Martínez, Carlos
© 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., The increasing complexity of deep neural networks often leads to overparameterization, limiting their deployment in resource-constrained environments. This paper introduces a novel pruning methodology based on Nonlinear Oblique Subspace Projections (NObSP), an algorithm originally developed to enhance the interpretability of kernel regression models. In this work, NObSP is extended to neural networks to estimate the contribution of individual neurons and entire layers to the network's output. This is achieved by quantifying the nonlinear transformations performed by the network on its inputs and internal representations. By computing the magnitude of these variables, neurons contributing minimally to the network's output are identified. Specifically, neurons with a nonlinear effect magnitude falling below a defined threshold are pruned, effectively removing them from the network. Consequently, NObSP provides a data-driven, interpretability-based approach to neural network pruning, leading to simplified and efficient models with minor performance impact. This method offers a clear framework for understanding which components of a neural network are critical for its function, facilitating the development of more compact and interpretable deep learning models. Finally, the efficacy of NObSP for pruning is demonstrated in CIFAR-10 and MNIST classification problems, where it is compared to traditional pruning algorithms based on the Ln norm., The work of this paper is supported by the following funds: the Condonable Credit Doctoral Support Grant from the Academic Vice Rector’s Office, the Scholarship Loan Program Grant from Colfuturo, the Spanish project SEAMLESS: Sustainable learning-based Management of Multiresource Large-scale Systems (ref. PID2023-148840OBI00), funded by MCIN/AEI/10.13039/501100011033/FEDER, UE and the project PID2023-146261OB-I00, funded by Ministry of Science and Innovation of Spain., Peer Reviewed, Postprint (author's final draft)