Felipe Antunes-Santos
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Felipe Antunes-Santos has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Carlos Lopez-Molina
Department of Statistics, Computer Science and Mathematics, Public University of Navarre (UPNA), 31006 Pamplona, Spain
4 shared publications - Bernard De Baets
KERMIT, Department of Data Analysis and Mathematical Modelling, Ghent University, Gent, Belgium
3 shared publications - Alberto Gonçalves
Federal University of Amazonas (UFAM)
2 shared publications - Alexandre Reis Pires Ferreira
College of Physical Education and Exercise Science, University of Brasília (UnB), Brasília 70910-900, Brazil
2 shared publications - Allan Inoue
Exercise and Sport Sciences Postgraduate Program, Rio de Janeiro State University, Rio de Janeiro, Brazil
2 shared publications - Arnau Mir-Fuentes
Universitat de les Illes Balears
2 shared publications
Latest Publications
- Open Access
- 16 April 2026
- Published by JMIR Publications Inc. in JMIR Cardio
- Vol. 10, e81917
Heart Rate Estimation Using the Galaxy Watch During Maximal Cardiopulmonary Exercise Testing: Cross-Sectional Validation Study
Background: Photoplethysmography-based smartwatches are increasingly used for continuous heart rate (HR) monitoring. Their accuracy has been demonstrated at rest or during low-intensity activity, but data are scarce for maximal-intensity exercise, when motion artifacts and rapid hemodynamic changes can degrade the photoplethysmography signal. Validating these devices under such demanding conditions is essential before they are applied to clinical exercise testing, athletic training, or remote health monitoring. Objective: This study aimed to evaluate the validity of the Samsung Galaxy Watch 6 (GW6) in estimating HR throughout a graded, maximal ramp cardiopulmonary exercise test performed on a treadmill. A secondary aim was to explore whether measurement error varies across 5 predefined intensity zones (50%‐60%, 60%‐70%, 70%‐80%, 80%‐90%, and 90%‐100% of the maximum HR determined individually for each participant). Methods: Overall, 55 healthy adults (30 men, 25 women; mean age 30.3, SD 8.2 years) completed a symptom-limited incremental treadmill protocol to volitional exhaustion. Simultaneous HR recordings were obtained from the GW6 (left arm) and a Polar H10 chest strap monitor, which served as the reference standards. For each intensity zone, the following agreement indices were computed: intraclass correlation coefficient (ICC), median absolute error, median absolute percentage error, and root mean squared error. Bland-Altman analysis was performed to quantify the mean bias and 95% limits of agreement between the GW6 and the Polar H10. Statistical significance was set at P <.05. Results: Agreement between the GW6 and Polar H10 varied across exercise intensities. ICC indicated moderate to good agreement at low to moderate intensities (ICC=0.71 at 50%‐60%; ICC=0.89 at 60%‐70%; ICC=0.54 at 70%‐80%; and ICC=0.64 at 80%‐90% HRmax), and at 90%‐100% of HRmax the agreement was good-to-excellent (ICC=0.90). Absolute error metrics showed stable or reduced errors with increasing intensity, with median absolute error consistently around 1‐3 bpm and median absolute percentage error declining from 2.90% at 50%‐60% HRmax to 0.60%‐0.75% at ≥70% HRmax. Root mean squared error ranged from 4.62 to 4.88 bpm across intensity zones. Bland-Altman analysis showed that the GW6 consistently underestimated HR compared with the Polar H10, with an overall mean bias of −2.67 bpm and wide limits of agreement (−16.90 to 11.57 bpm). This negative bias was present across all HR zones. The agreement was adequate for group-level comparisons but displayed substantial individual variability. Conclusions: The GW6 provides a good degree of validity for HR monitoring during a maximal treadmill cardiopulmonary exercise test in healthy young adults. Although measurement error increases modestly at near-maximal workloads, absolute errors remain well within clinically acceptable thresholds. These findings support the potential use of GW6 as a convenient, noninvasive alternative for HR tracking in laboratory-based exercise testing.
- 16 April 2026
- Published by JMIR Publications Inc. in JMIR Cardio
- Vol. 10, e81917
Open AccessHeart Rate Estimation Using the Galaxy Watch During Maximal Cardiopulmonary Exercise Testing: Cross-Sectional Validation Study
Background: Photoplethysmography-based smartwatches are increasingly used for continuous heart rate (HR) monitoring. Their accuracy has been demonstrated at rest or during low-intensity activity, but data are scarce for maximal-intensity exercise, when motion artifacts and rapid hemodynamic changes can degrade the photoplethysmography signal. Validating these devices under such demanding conditions is essential before they are applied to clinical exercise testing, athletic training, or remote health monitoring. Objective: This study aimed to evaluate the validity of the Samsung Galaxy Watch 6 (GW6) in estimating HR throughout a graded, maximal ramp cardiopulmonary exercise test performed on a treadmill. A secondary aim was to explore whether measurement error varies across 5 predefined intensity zones (50%‐60%, 60%‐70%, 70%‐80%, 80%‐90%, and 90%‐100% of the maximum HR determined individually for each participant). Methods: Overall, 55 healthy adults (30 men, 25 women; mean age 30.3, SD 8.2 years) completed a symptom-limited incremental treadmill protocol to volitional exhaustion. Simultaneous HR recordings were obtained from the GW6 (left arm) and a Polar H10 chest strap monitor, which served as the reference standards. For each intensity zone, the following agreement indices were computed: intraclass correlation coefficient (ICC), median absolute error, median absolute percentage error, and root mean squared error. Bland-Altman analysis was performed to quantify the mean bias and 95% limits of agreement between the GW6 and the Polar H10. Statistical significance was set at P <.05. Results: Agreement between the GW6 and Polar H10 varied across exercise intensities. ICC indicated moderate to good agreement at low to moderate intensities (ICC=0.71 at 50%‐60%; ICC=0.89 at 60%‐70%; ICC=0.54 at 70%‐80%; and ICC=0.64 at 80%‐90% HRmax), and at 90%‐100% of HRmax the agreement was good-to-excellent (ICC=0.90). Absolute error metrics showed stable or reduced errors with increasing intensity, with median absolute error consistently around 1‐3 bpm and median absolute percentage error declining from 2.90% at 50%‐60% HRmax to 0.60%‐0.75% at ≥70% HRmax. Root mean squared error ranged from 4.62 to 4.88 bpm across intensity zones. Bland-Altman analysis showed that the GW6 consistently underestimated HR compared with the Polar H10, with an overall mean bias of −2.67 bpm and wide limits of agreement (−16.90 to 11.57 bpm). This negative bias was present across all HR zones. The agreement was adequate for group-level comparisons but displayed substantial individual variability. Conclusions: The GW6 provides a good degree of validity for HR monitoring during a maximal treadmill cardiopulmonary exercise test in healthy young adults. Although measurement error increases modestly at near-maximal workloads, absolute errors remain well within clinically acceptable thresholds. These findings support the potential use of GW6 as a convenient, noninvasive alternative for HR tracking in laboratory-based exercise testing.
- Open Access
- 7 August 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Validity of Galaxy Watch for estimating heart rate during maximal cardiopulmonary exercise testing (Preprint)
BACKGROUND Photoplethysmography (PPG)–based smartwatches are increasingly used for continuous heart‑rate (HR) monitoring. Their accuracy has been demonstrated at rest or during low‑intensity activity, but data are scarce for maximal‑intensity exercise, when motion artifacts and rapid hemodynamic changes can degrade the PPG signal. Validating these devices under such demanding conditions is essential before they are applied to clinical exercise testing, athletic training, or remote health monitoring. OBJECTIVE The primary aim of this study was to evaluate the validity of the Samsung Galaxy Watch 6 (GW6) in estimating HR throughout a graded, maximal ramp cardiopulmonary exercise test (CPET) performed on a treadmill. A secondary aim was to explore whether measurement error varies across five predefined intensity zones (50–60 %, 60–70 %, 70–80 %, 80–90 %, and 90–100 % of the maximum HR determined individually for each participant). METHODS Fifty‑five healthy adults (30 men, 25 women; mean ± SD age = 30.3 ± 8.2 years) completed a symptom‑limited incremental treadmill protocol to volitional exhaustion. Simultaneous HR recordings were obtained from the GW6 (left arm) and a Polar H10 chest‑strap monitor, which served as the reference standards. For each intensity zone, the following agreement indices were computed: intraclass correlation coefficient (ICC), median absolute error (MAE), and median absolute percentage error (MAPE), and root mean squared error (RMSE). Bland–Altman analysis was performed to quantify the mean bias and 95% limits of agreement (LoA) between the GW6 and the Polar H10. Statistical significance was set at p < 0.05. RESULTS Agreement between the GW6 and Polar H10 varied across exercise intensities. ICC indicated moderate to good agreement at low to moderate intensities (ICC = 0.71 at 50–60%; ICC = 0.89 at 60–70%; ICC = 0.54 at 70–80%; and ICC = 0.64 at 80–90% HRmax), and at 90–100% of HRmax the agreement was good-to-excellent (ICC = 0.90). Absolute error metrics showed stable or reduced errors with increasing intensity, with MAE consistently around 1–3 bpm and MAPE declining from 2.90% at 50–60% HRmax to 0.60–0.75% at ≥70% HRmax. RMSE ranged from 4.62 to 4.88 bpm across intensities zones. Bland–Altman analysis showed that the GW6 consistently underestimated HR compared with the Polar H10, with an overall mean bias of −2.67 bpm and wide limits of agreement (−16.90 to 11.57 bpm). This negative bias was present across all heart rate zones. The agreement was adequate for group-level comparisons but displayed substantial individual variability. CONCLUSIONS The Samsung GW6 provides a good degree of validity for HR monitoring during maximal treadmill CPET in healthy young adults. Although measurement error increases modestly at near‑maximal workloads, absolute errors remain well within clinically acceptable thresholds. These findings support the potential use of GW6 as a convenient, non‑invasive alternative for HR tracking in laboratory‑based exercise testing. CLINICALTRIAL
- 7 August 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessValidity of Galaxy Watch for estimating heart rate during maximal cardiopulmonary exercise testing (Preprint)
BACKGROUND Photoplethysmography (PPG)–based smartwatches are increasingly used for continuous heart‑rate (HR) monitoring. Their accuracy has been demonstrated at rest or during low‑intensity activity, but data are scarce for maximal‑intensity exercise, when motion artifacts and rapid hemodynamic changes can degrade the PPG signal. Validating these devices under such demanding conditions is essential before they are applied to clinical exercise testing, athletic training, or remote health monitoring. OBJECTIVE The primary aim of this study was to evaluate the validity of the Samsung Galaxy Watch 6 (GW6) in estimating HR throughout a graded, maximal ramp cardiopulmonary exercise test (CPET) performed on a treadmill. A secondary aim was to explore whether measurement error varies across five predefined intensity zones (50–60 %, 60–70 %, 70–80 %, 80–90 %, and 90–100 % of the maximum HR determined individually for each participant). METHODS Fifty‑five healthy adults (30 men, 25 women; mean ± SD age = 30.3 ± 8.2 years) completed a symptom‑limited incremental treadmill protocol to volitional exhaustion. Simultaneous HR recordings were obtained from the GW6 (left arm) and a Polar H10 chest‑strap monitor, which served as the reference standards. For each intensity zone, the following agreement indices were computed: intraclass correlation coefficient (ICC), median absolute error (MAE), and median absolute percentage error (MAPE), and root mean squared error (RMSE). Bland–Altman analysis was performed to quantify the mean bias and 95% limits of agreement (LoA) between the GW6 and the Polar H10. Statistical significance was set at p < 0.05. RESULTS Agreement between the GW6 and Polar H10 varied across exercise intensities. ICC indicated moderate to good agreement at low to moderate intensities (ICC = 0.71 at 50–60%; ICC = 0.89 at 60–70%; ICC = 0.54 at 70–80%; and ICC = 0.64 at 80–90% HRmax), and at 90–100% of HRmax the agreement was good-to-excellent (ICC = 0.90). Absolute error metrics showed stable or reduced errors with increasing intensity, with MAE consistently around 1–3 bpm and MAPE declining from 2.90% at 50–60% HRmax to 0.60–0.75% at ≥70% HRmax. RMSE ranged from 4.62 to 4.88 bpm across intensities zones. Bland–Altman analysis showed that the GW6 consistently underestimated HR compared with the Polar H10, with an overall mean bias of −2.67 bpm and wide limits of agreement (−16.90 to 11.57 bpm). This negative bias was present across all heart rate zones. The agreement was adequate for group-level comparisons but displayed substantial individual variability. CONCLUSIONS The Samsung GW6 provides a good degree of validity for HR monitoring during maximal treadmill CPET in healthy young adults. Although measurement error increases modestly at near‑maximal workloads, absolute errors remain well within clinically acceptable thresholds. These findings support the potential use of GW6 as a convenient, non‑invasive alternative for HR tracking in laboratory‑based exercise testing. CLINICALTRIAL
- Open Access
- 22 May 2025
- Published by MDPI AG in Journal of Imaging
- Vol. 11 (6)
CAS-SFCM: Content-Aware Image Smoothing Based on Fuzzy Clustering with Spatial Information
Image smoothing is a low-level image processing task mainly aimed at homogenizing an image, mitigating noise, or improving the visibility of certain image areas. There exist two main strategies for image smoothing. The first strategy is content-unaware image smoothing. This strategy replicates identical smoothing behavior at every region in the image, hence ignoring any local or semi-local properties of the image. The second strategy is content-aware image smoothing, which takes into account the local properties of the image in order to adapt the smoothing behavior. Such adaptation to local image conditions is intended to avoid the blurring of relevant structures (such as ridges, edges, and blobs) in the image. While the former strategy was ubiquitous in the early years of image processing, the last 20 years have seen an ever-increasing use of the latter, fueled by a combination of greater computational capability and more refined mathematical models. In this work, we propose a novel content-aware image smoothing method based on soft (fuzzy) clustering. Our proposal capitalizes on the strengths of soft clustering to produce content-aware smoothing and allows for the direct configuration of the most relevant parameters for the task: the number of distinctive regions in the image and the relative relevance of spatial and tonal information in the smoothing. The proposed method is put to the test on both artificial and real-world images, combining both qualitative and quantitative analyses. We also propose the use of a local homogeneity measure for the quantitative analysis of image smoothing results. We show that the proposed method is not sensitive to centroid initialization and can be used for both artificial and real-world images.
- 22 May 2025
- Published by MDPI AG in Journal of Imaging
- Vol. 11 (6)
Open AccessCAS-SFCM: Content-Aware Image Smoothing Based on Fuzzy Clustering with Spatial Information
Image smoothing is a low-level image processing task mainly aimed at homogenizing an image, mitigating noise, or improving the visibility of certain image areas. There exist two main strategies for image smoothing. The first strategy is content-unaware image smoothing. This strategy replicates identical smoothing behavior at every region in the image, hence ignoring any local or semi-local properties of the image. The second strategy is content-aware image smoothing, which takes into account the local properties of the image in order to adapt the smoothing behavior. Such adaptation to local image conditions is intended to avoid the blurring of relevant structures (such as ridges, edges, and blobs) in the image. While the former strategy was ubiquitous in the early years of image processing, the last 20 years have seen an ever-increasing use of the latter, fueled by a combination of greater computational capability and more refined mathematical models. In this work, we propose a novel content-aware image smoothing method based on soft (fuzzy) clustering. Our proposal capitalizes on the strengths of soft clustering to produce content-aware smoothing and allows for the direct configuration of the most relevant parameters for the task: the number of distinctive regions in the image and the relative relevance of spatial and tonal information in the smoothing. The proposed method is put to the test on both artificial and real-world images, combining both qualitative and quantitative analyses. We also propose the use of a local homogeneity measure for the quantitative analysis of image smoothing results. We show that the proposed method is not sensitive to centroid initialization and can be used for both artificial and real-world images.
- Open Access
- 25 September 2023
- Published by Sociedade Brasileira de Computação
- p. 272-284
Application of Deep Learning Techniques to Depth Images for Person Tracking and Detection
Nowadays, using neural networks for image processing and tracking of individuals/objects is a highly popular subject that can be applied to various real-world issues. However, for such cases, the image often needs to possess good quality and exhibit distinct features to aid in object detection, posing challenges in environments with low or no illumination. In our work, we present a comparative study on the performance of leading convolutional neural networks for the detection and tracking of individuals through depth color images generated by infrared sensors. Additionally, we aim to demonstrate the usability of the YOLO (You Only Look Once) architecture as an alternative for identifying objects in images generated by sensors that do not rely on illumination. Experimental results showcase that the approach using YOLO Tiny improves accuracy by approximately 9% and processes around 8 times more frames per second (FPS).
- 25 September 2023
- Published by Sociedade Brasileira de Computação
- p. 272-284
Open AccessApplication of Deep Learning Techniques to Depth Images for Person Tracking and Detection
Nowadays, using neural networks for image processing and tracking of individuals/objects is a highly popular subject that can be applied to various real-world issues. However, for such cases, the image often needs to possess good quality and exhibit distinct features to aid in object detection, posing challenges in environments with low or no illumination. In our work, we present a comparative study on the performance of leading convolutional neural networks for the detection and tracking of individuals through depth color images generated by infrared sensors. Additionally, we aim to demonstrate the usability of the YOLO (You Only Look Once) architecture as an alternative for identifying objects in images generated by sensors that do not rely on illumination. Experimental results showcase that the approach using YOLO Tiny improves accuracy by approximately 9% and processes around 8 times more frames per second (FPS).
- 4 July 2022
- Published by Springer Nature
- p. 259-271
A Framework for Active Contour Initialization with Application to Liver Segmentation in MRI
- 4 July 2022
- Published by Springer Nature
- p. 259-271
A Framework for Active Contour Initialization with Application to Liver Segmentation in MRI
- 4 July 2022
- Published by Springer Nature
- p. 443-454
Content-Aware Image Smoothing Based on Fuzzy Clustering
- 4 July 2022
- Published by Springer Nature
- p. 443-454
Content-Aware Image Smoothing Based on Fuzzy Clustering
- Open Access
- 21 April 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (9) , 3733
Unsupervised Cell Segmentation and Labelling in Neural Tissue Images
Neurodegenerative diseases are a group of largely incurable disorders characterised by the progressive loss of neurons and for which often the molecular mechanisms are poorly understood. To bridge this gap, researchers employ a range of techniques. A very prominent and useful technique adopted across many different fields is imaging and the analysis of histopathological and fluorescent label tissue samples. Although image acquisition has been efficiently automated recently, automated analysis still presents a bottleneck. Although various methods have been developed to automate this task, they tend to make use of single-purpose machine learning models that require extensive training, imposing a significant workload on the experts and introducing variability in the analysis. Moreover, these methods are impractical to audit and adapt, as their internal parameters are difficult to interpret and change. Here, we present a novel unsupervised automated schema for object segmentation of images, exemplified on a dataset of tissue images. Our schema does not require training data, can be fully audited and is based on a series of understandable biological decisions. In order to evaluate and validate our schema, we compared it with a state-of-the-art automated segmentation method for post-mortem tissues of ALS patients.
- 21 April 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (9) , 3733
Open AccessUnsupervised Cell Segmentation and Labelling in Neural Tissue Images
Neurodegenerative diseases are a group of largely incurable disorders characterised by the progressive loss of neurons and for which often the molecular mechanisms are poorly understood. To bridge this gap, researchers employ a range of techniques. A very prominent and useful technique adopted across many different fields is imaging and the analysis of histopathological and fluorescent label tissue samples. Although image acquisition has been efficiently automated recently, automated analysis still presents a bottleneck. Although various methods have been developed to automate this task, they tend to make use of single-purpose machine learning models that require extensive training, imposing a significant workload on the experts and introducing variability in the analysis. Moreover, these methods are impractical to audit and adapt, as their internal parameters are difficult to interpret and change. Here, we present a novel unsupervised automated schema for object segmentation of images, exemplified on a dataset of tissue images. Our schema does not require training data, can be fully audited and is based on a series of understandable biological decisions. In order to evaluate and validate our schema, we compared it with a state-of-the-art automated segmentation method for post-mortem tissues of ALS patients.
- Open Access
- 1 December 2019
- Published by Elsevier in Data in Brief
- Vol. 27, 104655
Image datasets of cocoa beans for taxonomy nuances evaluation
- 1 December 2019
- Published by Elsevier in Data in Brief
- Vol. 27, 104655
Open AccessImage datasets of cocoa beans for taxonomy nuances evaluation
- Open Access
- 1 June 2019
- Published by Elsevier in Data in Brief
- Vol. 24, 103916
An image dataset of cut-test-classified cocoa beans
- 1 June 2019
- Published by Elsevier in Data in Brief
- Vol. 24, 103916
Open AccessAn image dataset of cut-test-classified cocoa beans