Pasar al contenido principal

ARQUITECTURAS EDGE-COMPUTING Y ALGORITMOS DE PROCESAMIENTO DISTIBUIDO PARA IMAGEN ULTRASONICA INDELIGENTE

PID2022-143271OB-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 2022
Plan Estatal 2021-2023 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario AGENCIA ESTATAL CONSEJO SUPERIOR DE INVESTIGACIONES CIENTIFICAS (CSIC)

Resultados relacionados

Resultados totales (Incluyendo duplicados): 15
Encontrada(s) 1 página(s)

DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Artículo científico (JournalArticle)

Automatic estimation of surface and probe location for 3D imaging with bidimensional arrays

  • Cosarinsky, Guillermo
  • Cruza, Jorge F.
  • Muñoz, Mario
  • Camacho Sosa-Días, Jorge
17 páginas, 22 figuras, 5 tablas, Ultrasound imaging for Non Destructive Testing is frequently performed in an immersion setup, where water is used as coupling medium between the probe and the component under test. For the computation of the time delays needed for beam-forming, the shape of the component surface and probe location and orientation (PLO) must be known. In this work we develop methods for the automatic detection of the surface and the estimation of PLO for 2D array probes. In particular, the methods developed apply to three types of elementary surfaces which are usually found in industrial and structural components: planes, cylinders and spheres. The methods use the measured surface echoes Time of Flight (TOF) to fit parametric models based on ray propagation and reflection on the surface, giving the coordinates and Euler angles that define the PLO relative to the component under test. Validation experiments with four test specimens representing the three types of surfaces are presented. The accuracy and precision of estimated PLO coordinates and angles are analyzed, and a Total Focusing Method (TFM) imaging example is shown achieving a correct detection of artificial defects in the component for a different PLOs., This research was supported by the projects PDC2022-133504-I00 founded by MCIN/AEI/10.13039/501100011033 (Spain) and ‘‘Next Generation EU’’/PRTR (European Union), by the fellowship PRE2019- 088602 founded by MCIU (Spain) and by the project PID2022- 143271OB-I00, founded MCIN/AEI /10.13039/501100011033/FEDER, UE
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Artículo científico (JournalArticle)

Deep Learning-Based Algorithms for Real-Time Lung Ultrasound Assisted Diagnosis

  • Muñoz, Mario
  • Rubio, Adrián
  • Cosarinsky, Guillermo
  • Cruza, Jorge F.
  • Camacho Sosa-Días, Jorge
Lung ultrasound is an increasingly utilized non-invasive imaging modality for assessing lung condition but interpreting it can be challenging and depends on the operator’s experience. To address these challenges, this work proposes an approach that combines artificial intelligence (AI) with feature-based signal processing algorithms. We introduce a specialized deep learning model designed and trained to facilitate the analysis and interpretation of lung ultrasound images by automating the detection and location of pulmonary features, including the pleura, A-lines, B-lines, and consolidations. Employing Convolutional Neural Networks (CNNs) trained on a semi-automatically annotated dataset, the model delineates these pulmonary patterns with the objective of enhancing diagnostic precision. Real-time post-processing algorithms further refine prediction accuracy by reducing false-positives and false-negatives, augmenting interpretational clarity and obtaining a final processing rate of up to 20 frames per second with accuracy levels of 89% for consolidation, 92% for B-lines, 66% for A-lines, and 92% for detecting normal lungs compared with an expert opinion., This research was partially supported by the project PID2022-143271OB-I00, founded MCIN/AEI/10.13039/501100011033/FEDER, UE and by the fellowship PRE2019-088602 founded by MCIU (Spain)., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Deep learning for ultrasound surface echo detection

  • Cosarinsky, Guillermo
  • Muñoz, Mario
  • Fernández Cruza, Jorge
  • Rubio, Adrián
  • Camacho Sosa-Días, Jorge
20th World Conference for Non-Destructive Testing (WCNDT), 27-31 May 2024, Songdo Convensia, Incheon, Korea. 9 páginas, 6 figuras, 1 tabla., Ultrasound array imaging techniques are often applied using a coupling medium to transmit waves from the transducer to the scanned component. In this type of tests, the echoes generated by the wave reflection at the component surface contain information that can be used to estimate the surface shape and its position relative to the probe, which are needed to compute the imaging focal laws. The most interesting feature to extract from surfaces echoes is the Time of Flight (TOF) for each reception channel. TOF measurement in acoustic signals is a problem that frequently arises not only in Non Destructive Testing (NDT), but also in medical imaging, geophysics and other fields. The surface echo in a signal is usually identified as the first high amplitude peak, and is traditionally detected with standard methods as threshold crossing, peak search or matched filters. All these methods are easily affected by Signal to Noise Ratio, which might be high in some channels, for example if the ray arrives at the corresponding element at a low sensibility angle. Moreover, spurious echoes from bubbles, the emission signal tail or echoes from other parts of the component can be incorrectly labelled as surface echoes. This kind of outliers are difficult to filter by conventional means and, thus, there is a need for more robust surface estimation methods. In this work, we study the application of a deep 3D Convolutional Neural Network (CNN) to detect surface echoes in Full Matrix Capture (FMC) ultrasound data. The CNN is trained using signals acquired with a matrix array and a reference component, using a robotic arm set-up to measure the probe position and orientation relative to the component. Thus, with knowledge of the set-up geometry, a model is used to compute the theoretical surface echo TOFs, which are used to label the data for training. The model was tested with a validation set, presenting good agreement with the ground truth. The results were compared with the threshold crossing method observing a significant reduction of outliers in the TOF estimation., This research was supported by the projects PDC2022-133504-I00 founded by MCIN/AEI/10.13039/501100011033 (Spain) and “Next Generation EU”/PRTR (European Union), by the fellowship PRE2019-088602 founded by MCIU (Spain) and by the project PID2022-143271OB-I00, founded MCIN/AEI /10.13039/501100011033/FEDER, UE, Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Artículo científico (JournalArticle)

Three-Dimensional Convolutional Neural Network for Ultrasound Surface Echo Detection

  • Muñoz, Mario
  • Rubio, Adrián
  • Larrea, Marcelo
  • Cruza, Jorge F.
  • Camacho Sosa-Días, Jorge
  • Cosarinsky, Guillermo
This article belongs to the Special Issue Ultrasonic Imaging and Sensors II., Ultrasound array imaging frequently employs a coupling medium to facilitate wave transmission from the transducer to the target component. Surface echoes, identified by their high-amplitude peaks, are crucial for determining the Time of Flight (TOF) in each channel, which is essential for deriving imaging focal laws. Accurate TOF measurement is vital in numerous applications, such as Non-Destructive Testing (NDT) and medical imaging. Conventional methods, such as threshold crossing and peak search, are highly sensitive to noise and spurious signals, therefore, more robust estimation techniques are needed. This study explores the application of a deep 3D Convolutional Neural Network (CNN) to detect surface echoes in Full Matrix Capture (FMC) ultrasound data. The CNN was trained on signals obtained with a matrix array and a set of reference components, utilizing a robotic arm setup to ensure precise probe positioning. Theoretical TOFs were computed based on the setup geometry to generate labeled training data. Test results indicated that the CNN model, which we have called DeepEcho3D, closely aligned with the ground truth and significantly reduced TOF estimation outliers (up to 98%) compared to traditional methods, demonstrating its potential for improved accuracy in surface echo detection., This research was supported by the project PID2022-143271OB-I00, funded by MCIN/AEI/10.13039/501100011033/FEDER, UE, the project 202450E192, funded by CSIC, the project PLEC2024-011165, funded by MICIU/AEI/10.13039/501100011033/FEDER, UE, and by the European Commission—NextGenerationEU, through Momentum CSIC Programme: Develop Your Digital Talent. The funding for these actions/grants and contracts comes from the European Union’s Recovery and Resilience Facility—Next Generation, in the framework of the General Invitation of the Spanish Government’s public business entity Red.es to participate in talent attraction and retention programs within Investment 4 of Component 19 of the Recovery, Transformation and Resilience Plan. G. COSARINKSY STAFF hired under the Generation D initiative, promoted by Red.es, an organization attached to the Ministry for Digital Transformation and the Civil Service, for the attraction and retention of talent through grants and training contracts, financed by the Recovery, Transformation and Resilience Plan through the European Union’s Next Generation funds., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Artículo científico (JournalArticle)

Free-form surface reconstruction for adaptive imaging with 2D ultrasound arrays

  • Cosarinsky, Guillermo
  • Cruza, Jorge F.
  • Muñoz, Mario
  • Rubio, Adrián
  • Camacho Sosa-Días, Jorge
This research was supported by the project PID2022-143271OB- I00, funded by MCIN/AEI /10.13039/501100011033/FEDER, UE., the project PLEC2024-011165, funded by MICIU/AEI/10.13039/ 501100011033/ FEDER, UE , and by the European Commission – NextGenerationEU, through Momentum CSIC Programme: Develop Your Digital Talent. The funding for these actions/grants and contracts comes from the European Union’s Recovery and Resilience Facility- Next Generation, in the framework of the General Invitation of the Spanish Government’s public business entity Red.es to participate in talent attraction and retention programmes within Investment 4 of Component 19 of the Recovery, Transformation and Resilience Plan. G. Cosarinsky staff is hired under the Generation D initiative, pro- moted by Red.es, an organization attached to the Ministry for Digital Transformation and the Civil Service, for the attraction and retention of talent through grants and training contracts, financed by the Recovery, Transformation and Resilience Plan through the European Union’s Next Generation funds, This paper introduces the Aperture Interpolation Method (AIM) for free-form surface reconstruction using 2D ultrasound arrays. Surface reconstruction plays an essential role in adaptive imaging methods, allowing for effective operation even with imprecise or unknown system geometries. The AIM method leverages the time of flight (TOF) of echos from surface reflections to estimate the surface geometry of a component under test, without requiring a pre-existing component model. In this paper we present the theoretical basis of AIM for three acquisition modes: pulse-echo, pitch-catch, and plane wave. Experimental validation is performed using a complex-shaped aluminum test piece scanned with an 11x11, 3MHz matrix array mounted on a robotic arm. The ultrasonically reconstructed surface is compared with laser scanning data for accuracy analysis. Furthermore, the ultrasonically measured surface data is used to generate Total Focusing Method (TFM) images of the component, which are then compared against micro-CT images for validation. The experimental results validate the AIM method’s capability to accurately reconstruct surfaces, facilitating reliable imaging of the components defects, thereby demonstrating its potential applicability in industrial settings., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Artículo científico (JournalArticle)

Evaluating deep learning approaches for AI-assisted lung ultrasound diagnosis: an international multi-center and multi-scanner study

  • Muñoz, Mario
  • Han, Xi
  • Camacho Sosa-Días, Jorge
  • Perrone, Tiziano
  • Smargiassi, Andrea
  • Inchingolo, Riccardo
  • Tung-Chen, Yale
  • Demi, Libertario
18 páginas, 11 figuras, 7 tablas, Lung ultrasound (LUS) interpretation is often subjective and operator-dependent, motivating the development of automated, artificial intelligence (AI)-based methods. This international, multi-center study evaluated two distinct deep learning approaches for automated LUS severity scoring for pulmonary infections caused by COVID-19: a pre-trained classification model (CM) and a segmentation model based method (SM); assessing performance at video, exam, and prognostic levels. Two datasets were analyzed: one comprising data from multiple scanners and another using data from a single scanner. Results showed that the SM achieved prognostic-level agreement with expert clinicians comparable to that of the CM. Furthermore, at the exam level, over 84% of examinations were classified with acceptable error (≤ 10 score difference) across both models and datasets, reaching both methods an agreement higher than 95% on the dataset acquired by a single scanner. The obtained results demonstrate the potential of AI-assisted LUS for reliable prognostic assessment and highlight that image quality and acquisition technique are key factors in achieving consistent and generalizable model performance, as well as the potential for international clinical translations., This research was partially supported by the project PID2022-143271OB-I00, founded MCIN/AEI /h t t ps : // d o i. org /1 0 .1 3 0 39 /5 0 1 1 0 0 0 1 1 0 3 3 / FEDER, UE and by the fellowship IMOVE24246 founded by CSIC (Spain). As well as by the grant Hub Life Science Advanced Diagnosis (HLS-AD), PNRR PNC-E3-2022- 23683266 PNC HLS-DA, INNOVA–CUP: E63C22003780001, funded by the Italian Ministry of Health under the National Complementary Plan Innovative Health Ecosystem—Unique Investment Code: PNC-E.3, Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Artículo científico (JournalArticle)

Plane Wave Imaging with Large-Scale 2D Sparse Arrays: A Method for Near-Field Enhancement via Aperture Diversity

  • Martínez-Graullera, Óscar
  • Camacho Sosa-Días, Jorge
  • Huecas, Jorge
  • Cosarinsky, Guillermo
  • Elvira Segura, Luis
  • Parrilla Romero, Montserrat
31 páginas, 28 figuras, 2 tablas, In the context of a medical imaging application for preclinical research, specifically cere- brovascular imaging in small animals, this work addresses the challenges associated with using a large-scale 2D ultrasonic array comprising 32 × 32 elements (96λ × 96λ ). The application imposes stringent requirements: operation in the extreme near field, high spatial resolution, high frequency, high frame rate, and imaging within a highly attenu- ating medium. These demands, combined with current technological limitations, such as element size and constraints on the number of channels that can be driven in parallel, present significant challenges for system design and implementation. To assess system performance, plane wave imaging is employed as a reference modality due to its ability to meet high acquisition speed requirements. Our analysis reveals limitations in spatial coverage and image quality when operating the full aperture under plane wave trans- mission constraints. To address these limitations, we propose a sparse aperture strategy. When combined with advanced signal processing techniques, this approach enhances both contrast and resolution while preserving acquisition speed, making it a promising solution for high-performance ultrasonic imaging under the demanding conditions of preclinical research., Jorge Huecas and Guillermo Cosarinsky, Staff hired under the Generation D initiative, promoted by Red.es, an organisation attached to the Ministry for Digital Transformation and the Civil Service, for the attraction and retention of talent through grants and training contracts, financed by the Recovery, Transformation and Resilience Plan through the European Union’s Next Generation funds. This research work was funded by the European Commission–NextGenerationEU, through Momentum CSIC Programme: Develop Your Digital Talent. This research work was funded by: PID2022-138013OB-I00, MCIN/AEI /10.13039/501100011033/FEDER, UE. This research work was funded by: PID2022-143271OB-I00, MCIN/AEI /10.13039/501100011033/FEDER, UE. This work has been developed within the framework of the TEC-2024/TEC-43 LUNABRAIN-CM project, funded with 1.026 million by the Community of Madrid through the R&D activities program in technologies, granted by Order 5696/2024., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Benchmark of Hardware Acceleration of AI-based Lung Ultrasound Segmentation

  • Rubio, Adrián
  • Muñoz, Mario
  • Cosarinsky, Guillermo
  • Camacho Sosa-Días, Jorge
II Jornadas de Jóvenes Investigadores ITEFI, 11 diciembre 2025, Madrid, España., Lung Ultrasound (LUS) imaging is a valuable diagnostic technique for lung condition evaluation, due to its non-ionizing and portable nature. However, its complex interpretation can be enhanced by Machine Learning (ML) tools, yet traditional solutions often fail to meet the speed demands of real-time applications. This work presents a Field Programmable Gate Array (FPGA)-based hardware solution for real-time segmentation of lung ultrasound images using a Convolutional Neural Network (CNN), achieving a throughput of 80 inferences per second., Supported by the project PID2022-143271OB-I00, funded by MCIN/AEI/10.13039/501100011033/FEDER, UE., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Segmentation of Lung Ultrasound images using Cross Domain Transfer Learning

  • Muñoz, Mario
  • Rubio, Adrián
  • Cosarinsky, Guillermo
  • Camacho Sosa-Días, Jorge
II Jornadas de Jóvenes Investigadores ITEFI, 11 de diciembre 2025, This research was partially supported by the project PID2022 143271OB I00, funded by MCIN/AEI/10.13039/501100011033/FEDER, UE; and supported by the European Commission + NextGenerationEU , through Momentum CSIC Programme, Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Ultrasound Surface Echo Detection by Volumetric Convolutional Neural Network

  • Muñoz, Mario
  • Rubio, Adrián
  • Larrea, Marcelo
  • Cruza, Jorge F.
  • Camacho Sosa-Días, Jorge
  • Cosarinsky, Guillermo
II Jornadas Jóvenes Investigadores ITEFI, 11 diciembre 2025, This research was partially supported by the project PID2022-143271OB-I00, funded by MCIN/AEI/10.13039/501100011033/FEDER, UE; and supported by the European Commission + NextGenerationEU, through Momentum CSIC Programme., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Adaptive 3D Plane Wave Imaging through interfaces

  • Cosarinsky, Guillermo
  • Rubio, Adrián
  • Muñoz, Mario
  • Cruza, Jorge F.
  • Camacho Sosa-Días, Jorge
Forum Acusticum / Euronoise 2025, the 11th EAA Annual European Conference on Acoustics and Noise Control Engineering, 23rd to 26th june 2025, Málaga (Spain) 7 páginas, 3 figuras., Plane Wave Imaging (PWI) with ultrasonic arrays, ini- tially developed for fast medical imaging, is nowa- days being applied in Ultrasonic Testing (UT) for Non- Destructive Testing (NDT). In NDT, the Total Focusing Method (TFM) is regarded as the gold standard for im- age quality. However, TFM requires multiple transmis- sion events (one for each array element), which can be slow for applications demanding very fast image acquisi- tion. Since PWI can generate images of comparable qual- ity with fewer transmissions, it presents a promising solu- tion for such high-speed applications. This study demon- strates how volumetric PWI can be achieved using matrix arrays for testing arbitrarily shaped components, where a coupling medium is required and ultrasound waves are re- fracted on the component surface. This introduces com- plexity in calculating the time delays required to gener- ate a plane wave within the component, a process that is straightforward in a single propagation medium scenario. Additionally, since the computation of those delays re- quires knowledge of the surface shape, we also explore how plane waves can be utilized for fast estimation of the surface shape., This research was supported by the project PID2022- 143271OB-I00, funded by MCIN/AEI /10.13039/501100011033/FEDER, UE., the project PLEC2024-011165, funded by MI- CIU/AEI/10.13039/501100011033/ FEDER, UE, and by the European Commission – NextGenerationEU, through Momentum CSIC Programme: Develop Your Digital Talent. The funding for these actions/grants and contracts comes from the European Union’s Recovery and Resilience Facility-Next Generation, in the framework of the General Invitation of the Spanish Government’s public business entity Red.es to participate in talent attraction and retention programmes within Investment 4 of Component 19 of the Recovery, Transformation and Resilience Plan. G. Cosarinsky staff is hired under the Generation D initiative, promoted by Red.es, an organisation attached to the Ministry for Digital Transformation and the Civil Service, for the attraction and retention of talent through grants and training contracts, financed by the Recovery, Transformation and Resilience Plan through the European Union’s Next Generation funds., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Cross-Domain Transfer Learning for Segmentation of Lung Ultrasound images

  • Muñoz, Mario
  • Rubio, Adrián
  • Cosarinsky, Guillermo
  • Camacho Sosa-Días, Jorge
Forum Acusticum / Euronoise 2025, the 11th EAA Annual European Conference on Acoustics and Noise Control Engineering, joint with the XLVI Spanish Acoustic Conference TECNIACUSTICA 2025, from June 23rd to 26th 2025, Málaga (Spain) 5 páginas, 2 figuras, 1 tabla, Developing deep learning models for lung ultrasound (LUS) segmentation is hindered by limited annotated data. This study demonstrates the effectiveness of cross-domain transfer learning (TL) using ImageNet pre-trained encoders (VGG16, ResNet50, MobileNetV2) within a U-Net architecture. Compared to training from scratch, TL with ResNet50 and VGG16 significantly improved segmentation Dice score (up to 30%). Furthermore, TL allowed reducing the training dataset to 45% while achieving performance comparable to the baseline trained on full data. These findings validate cross-domain TL as a valuable and data- efficient approach for LUS analysis., This research was partially supported by the project PID2022-143271OB-I00, funded by MCIN/AEI /10.13039/501100011033/FEDER, UE; and supported by the European Commission + NextGenerationEU, through Momentum CSIC Programme: Develop Your Digital Talent. The funding for these actions/grants and contracts comes from the European Union's Recovery and Resilience Facility-Next Generation, in the framework of the General Invitation of the Spanish Government's public business entity Red.es to participate in talent attraction and retention programmes within Investment 4 of Component 19 of the Recovery, Transformation and Resilience Plan. G. Cosarinsky staff is hired under the Generation D initiative, promoted by Red.es, an organisation attached to the Ministry for Digital Transformation and the Civil Service, for the attraction and retention of talent through grants and training contracts, financed by the Recovery, Transformation and Resilience Plan through the European Union's Next Generation funds., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Modular and flexible hardware architecture for real-time 3D ultrasound imaging

  • Cruza, Jorge F.
  • Mateos, Raul
  • Bueno, Ricardo
  • Montaldo, Gabriel
  • Camacho Sosa-Días, Jorge
Forum Acusticum / Euronoise 2025, the 11th EAA Annual European Conference on Acoustics and Noise Control Engineering, joint with the XLVI Spanish Acoustic Conference TECNIACUSTICA 2025, June 23rd to 26th 2025, Málaga (Spain) 8 páginas, 6 figuras, 2 tablas, Ultrasound 3D imaging with arrays usually requires hardware systems with more active channels than for conventional 2D imaging, which increases systems complexity. Furthermore, it increases the beamforming difficulty, as more image lines have to be generated (3D volume) using more acquired raw signals (A-Scans). Typical approaches to this problem can be divided in hardware and software solutions. In the first, the beamforming process is carried out by a specialized hardware (FPGA, ASIC’s, etc.), with lower power consumption, but also with lower flexibility. On the other hand, software beamformers are very flexible in terms of programming, but they require a much larger bandwidth with the acquisition system. This work presents a modular and flexible hardware architecture for 2D and 3D ultrasound imaging, which combines hardware beamforming capabilities with high-speed links with a processing computer. This approach enables to work with both hardware and software beamformers, taking full advantage of each one of them. A description of the architecture is given, along with the expected performance in terms of hardware processing power and data transfer rate. Finally, application examples are given: Total Focusing Method imaging in non-destructive-testing, and functional ultrasound imaging in the brain for pre-clinical models., Supported by the projects PID2022-143271OB-I00, funded by MCIN/AEI /10.13039/501100011033/FEDER, UE, PLEC2024-011165, funded by MICIU/AEI/10.13039/501100011033/ FEDER, UE, and by TEC-2024/TEC-43, LUNABRAIN-CM, funded by Comunidad de Madrid., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Deep learning for ultrasound surface echo detection with matrix arrays

  • Larrea, Marcelo
  • Muñoz, Mario
  • Rubio, Adrián
  • Cruza, Jorge F.
  • Cosarinsky, Guillermo
Forum Acusticum / Euronoise 2025, the 11th EAA Annual European Conference on Acoustics and Noise Control Engineering, joint with the XLVI Spanish Acoustic Conference TECNIACUSTICA 2025, June 23rd to 26th 2025, Málaga (Spain) 8 páginas, 5 figuras, 1 tabla, In Non-Destructive Testing (NDT), imaging with ar- ray probes is frequently performed in a two propaga- tion medium scenario, where the first acts as a coupling medium, and the second is the object under test. There- fore, the position and shape of the refracting interface be- tween the two mediums must be known in order to com- pute the imaging focal laws. This geometrical information can be inferred from the arrival times of echoes reflected on the object’s surface, which must be identified within the signals. A common approach for detecting these sur- face echoes is the first-threshold crossing method; how- ever, it is susceptible to outliers, and more robust al- ternatives are needed. In a recent work, we developed and trained a 3D Convolutional Neural Network (Deep- Echo3D) using an 11x11 matrix array to accurately and reliably detect surface echoes. While effective, the model is tailored to the specific array used. In this study, we ex- plore extending DeepEcho3D to other matrix arrays with varying shapes and frequencies. Our findings demonstrate that it is unnecessary to start from scratch for each new ar- ray; instead, we can leverage transfer learning from the initial model., This research was supported by the project PID2022- 143271OB-I00, funded by MCIN/AEI /10.13039/501100011033/FEDER, UE., the project PLEC2024-011165, funded by MI- CIU/AEI/10.13039/501100011033/ FEDER, UE, and by the European Commission – NextGenerationEU, through Momentum CSIC Programme: Develop Your Digital Talent. The funding for these actions/grants and contracts comes from the European Union’s Recovery and Resilience Facility-Next Generation, in the framework of the General Invitation of the Spanish Government’s public business entity Red.es to participate in talent attraction and retention programmes within Investment 4 of Component 19 of the Recovery, Transformation and Resilience Plan. G. Cosarinsky staff is hired under the Generation D initiative, promoted by Red.es, an organisation attached to the Ministry for Digital Transformation and the Civil Service, for the attraction and retention of talent through grants and training contracts, financed by the Recovery, Transformation and Resilience Plan through the European Union’s Next Generation funds., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publicación Contribución de congreso (ConferenceOutput)

Hardware acceleration of Convolutional Neural Network for Lung Ultrasound segmentation

  • Rubio, Adrián
  • Muñoz, Mario
  • Cosarinsky, Guillermo
  • Cruza, Jorge F.
Forum Acusticum / Euronoise 2025, the 11th EAA Annual European Conference on Acoustics and Noise Control Engineering, joint with the XLVI Spanish Acoustic Conference TECNIACUSTICA 2025, June 23rd to 26th 2025, Málaga (Spain) 5 páginas, 4 figuras, 1 tabla, Lung Ultrasound (LUS) imaging is a valuable diagnostic technique for lung condition evaluation, due to its non-ionizing and portable nature. However, its complex interpretation can be enhanced by Machine Learning (ML) tools, yet traditional solutions often fail to meet the speed demands of real-time applications. This paper presents a Field Programmable Gate Array (FPGA)-based hardware solution for real-time segmentation of lung ultrasound images using a Convolutional Neural Network (CNN), achieving a throughput of 80 inferences per second., Supported by the European Commission + NextGen- erationEU, through Momentum CSIC Programme: Develop Your Digital Talent. The funding for these actions/grants and contracts comes from the European Union’s Recovery and Resilience Facility-Next Gener- ation, in the framework of the General Invitation of the Spanish Government’s public business entity Red.es to participate in talent attraction and retention programmes within Investment 4 of Component 19 of the Recovery, Transformation and Resilience Plan. G. Cosarinsky staff is hired under the Generation D initiative, promoted by Red.es, an organisation attached to the Ministry for Digital Transformation and the Civil Service, for the attraction and retention of talent through grants and train- ing contracts, financed by the Recovery, Transformation and Resilience Plan through the European Union’s Next Generation funds. Supported by the project PID2022-143271OB-I00, funded by MCIN/AEI /10.13039/501100011033/FEDER, UE., Peer reviewed