David Chushig-Muzo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
David Chushig-Muzo has a total of 36 co-authors. Here are the top co-authors by number of shared publications:
- Cristina Soguero-Ruiz
Department of Signal Theory and Communications, Telematics and Computing Systems, Rey Juan Carlos University, Madrid, Spain
18 shared publications - Conceição Granja
University Hospital of North Norway
10 shared publications - Francisco Javier Lara-Abelenda
Department of Signal Theory and Communications, Telematics and Computing Systems, Rey Juan Carlos University, Madrid, Spain
7 shared publications - Vanesa Gómez-Martínez
King Juan Carlos University
7 shared publications - Ana Maria Wägner
Instituto Universitario de Investigaciones Biomédicas y Sanitarias, Universidad de Las Palmas de Gran Canaria, 35016 Las Palmas de Gran Canaria, Spain
6 shared publications - Inmaculada Mora-Jiménez
Universidad Rey Juan Carlos
5 shared publications
Latest Publications
- Open Access
- 5 November 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (21) , 10984
Enhancing Nocturnal Hypoglycemia Prediction Through Ensemble Feature Selection and Synthetic Data Augmentation
Nocturnal hypoglycemia (NH), which occurs when blood glucose levels fall below 70 mg/dL during sleep, remains a common complication in people with type 1 diabetes. Despite advances in continuous glucose monitoring (CGM), controlling both glucose and glycemic events remains challenging. Artificial Intelligence (AI) models are promising for identifying CGM patterns associated with NH and support its prediction, thereby reducing associated complications. However, two main challenges limit the performance of AI-based models for NH: (i) the class imbalance caused by the low number of NH events in the datasets; and (ii) the high-dimensionality caused by the extraction of CGM-derived features. To address this, we propose a two-stage methodology that combines ensemble feature selection (FS) methods and synthetic data augmentation techniques. We employed CGM features extracted using two window-based approaches (non-overlapping and overlapping) with three, four, and six windows. Model interpretability is provided through feature analysis using Shapley additive explanations (SHAP), enabling the identification of associated NH factors. We employed CGM data from 52 individuals with T1D, collected at Complejo Hospitalario Insular-Materno Infantil de Las Palmas de Gran Canaria, Spain. Among the evaluated configurations, the six-window configuration yielded the highest reported predictive performance. The two LR models examined in the SHAP analysis, both trained with RUS and overlapping windows, achieved test-set AUCROC values of 0.7403 using all features and 0.7400 using the selected feature subset, calculated from continuous prediction scores. Our findings extend prior evidence on temporal CGM representations to an adult NH setting not restricted to exercise-associated nights, using bootstrap-based Relief FS and alternative windowing configurations. Our interpretable methodology provides a clinically relevant approach for the early prediction of NH, supporting decision-making and enhancing the effectiveness of CGM-based monitoring systems. The proposed methodology may support timely preventive interventions and contribute to improving patient safety and confidence in diabetes management.
- 5 November 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (21) , 10984
Open AccessEnhancing Nocturnal Hypoglycemia Prediction Through Ensemble Feature Selection and Synthetic Data Augmentation
Nocturnal hypoglycemia (NH), which occurs when blood glucose levels fall below 70 mg/dL during sleep, remains a common complication in people with type 1 diabetes. Despite advances in continuous glucose monitoring (CGM), controlling both glucose and glycemic events remains challenging. Artificial Intelligence (AI) models are promising for identifying CGM patterns associated with NH and support its prediction, thereby reducing associated complications. However, two main challenges limit the performance of AI-based models for NH: (i) the class imbalance caused by the low number of NH events in the datasets; and (ii) the high-dimensionality caused by the extraction of CGM-derived features. To address this, we propose a two-stage methodology that combines ensemble feature selection (FS) methods and synthetic data augmentation techniques. We employed CGM features extracted using two window-based approaches (non-overlapping and overlapping) with three, four, and six windows. Model interpretability is provided through feature analysis using Shapley additive explanations (SHAP), enabling the identification of associated NH factors. We employed CGM data from 52 individuals with T1D, collected at Complejo Hospitalario Insular-Materno Infantil de Las Palmas de Gran Canaria, Spain. Among the evaluated configurations, the six-window configuration yielded the highest reported predictive performance. The two LR models examined in the SHAP analysis, both trained with RUS and overlapping windows, achieved test-set AUCROC values of 0.7403 using all features and 0.7400 using the selected feature subset, calculated from continuous prediction scores. Our findings extend prior evidence on temporal CGM representations to an adult NH setting not restricted to exercise-associated nights, using bootstrap-based Relief FS and alternative windowing configurations. Our interpretable methodology provides a clinically relevant approach for the early prediction of NH, supporting decision-making and enhancing the effectiveness of CGM-based monitoring systems. The proposed methodology may support timely preventive interventions and contribute to improving patient safety and confidence in diabetes management.
- Open Access
- 25 August 2026
- Published by MDPI AG in Preprints.org
Enhancing Nocturnal Hypoglycemia Prediction Through Ensemble Feature Selection and Synthetic Data Augmentation
Nocturnal Hypoglycemia (NH), which occurs when blood glucose levels fall below 70 mg/dL during sleep, remains a common complication in people with Type 1 Diabetes. Despite advances in Continuous Glucose Monitoring (CGM), controlling both glucose and glycemic events remains challenging. Artificial Intelligence (AI) models are promising for identifying CGM patterns associated with NH and supporting its prediction, thus reducing its occurrence and complications. However, two main challenges limit the performance of AI models for NH: (i) the class imbalance due to the low number of NH events in data; and (ii) the high-dimensionality caused by the extraction of CGM-derived features. To address this, we propose a two-stage methodology that combines ensemble Feature Selection (FS) methods and synthetic data augmentation. We performed CGM-derived feature extraction by using a time-window approach with overlapping and non-overlapping approaches. Model interpretability is provided through feature analysis using Shapley additive explanations, enabling the identification of associated NH factors. We employed CGM data from 52 individuals with T1D, collected at Complejo Hospitalario Insular-Materno Infantil de Las Palmas de Gran Canaria, Spain. Among the temporal configurations evaluated, the 6-hour window yielded the highest predictive performance. With ensemble FS retaining the top 40% of features, the best-performing model achieved an Area Under the Receiver Operating Characteristic Curve of 0.68 using a logistic regression and overlapping windows. Our findings highlighted the importance of window-based extraction in improving predictive performance. Our interpretable methodology provides a clinically relevant approach for the early prediction of NH, supporting decision-making and enhancing the effectiveness of CGM-based monitoring systems. The proposed methodology may support timely preventive interventions, reducing the occurrence of NH while improving patient safety and confidence in diabetes management.
- 25 August 2026
- Published by MDPI AG in Preprints.org
Open AccessEnhancing Nocturnal Hypoglycemia Prediction Through Ensemble Feature Selection and Synthetic Data Augmentation
Nocturnal Hypoglycemia (NH), which occurs when blood glucose levels fall below 70 mg/dL during sleep, remains a common complication in people with Type 1 Diabetes. Despite advances in Continuous Glucose Monitoring (CGM), controlling both glucose and glycemic events remains challenging. Artificial Intelligence (AI) models are promising for identifying CGM patterns associated with NH and supporting its prediction, thus reducing its occurrence and complications. However, two main challenges limit the performance of AI models for NH: (i) the class imbalance due to the low number of NH events in data; and (ii) the high-dimensionality caused by the extraction of CGM-derived features. To address this, we propose a two-stage methodology that combines ensemble Feature Selection (FS) methods and synthetic data augmentation. We performed CGM-derived feature extraction by using a time-window approach with overlapping and non-overlapping approaches. Model interpretability is provided through feature analysis using Shapley additive explanations, enabling the identification of associated NH factors. We employed CGM data from 52 individuals with T1D, collected at Complejo Hospitalario Insular-Materno Infantil de Las Palmas de Gran Canaria, Spain. Among the temporal configurations evaluated, the 6-hour window yielded the highest predictive performance. With ensemble FS retaining the top 40% of features, the best-performing model achieved an Area Under the Receiver Operating Characteristic Curve of 0.68 using a logistic regression and overlapping windows. Our findings highlighted the importance of window-based extraction in improving predictive performance. Our interpretable methodology provides a clinically relevant approach for the early prediction of NH, supporting decision-making and enhancing the effectiveness of CGM-based monitoring systems. The proposed methodology may support timely preventive interventions, reducing the occurrence of NH while improving patient safety and confidence in diabetes management.
- Open Access
- 3 March 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (5) , 2459
Tabular-to-Image Encoding Methods for Melanoma Detection: A Proof-of-Concept
Deep learning (DL) models have demonstrated strong performance in dermatological applications, particularly when trained on dermoscopic images. In contrast, tabular clinical data—such as patient metadata and lesion-level descriptors—are difficult to integrate into DL-based pipelines due to their heterogeneous, non-spatial, and often low-dimensional nature. As a result, these data are commonly handled using separate classical machine learning (ML) models. In this work, we present a proof-of-concept study that investigates whether dermatological tabular data can be transformed into two-dimensional image representations to enable convolutional neural network (CNN)-based learning. To this end, we employ the Low Mixed-Image Generator for Tabular Data (LM-IGTD), a framework designed to transform low-dimensional and heterogeneous tabular data into two-dimensional image representations, through type-aware encoding and controlled feature augmentation. Using this approach, we encode low-dimensional clinical metadata, high-dimensional lesion-level statistical features extracted from dermoscopic images, as well as their feature-level fusion, into grayscale image representations. The resulting image representations serve as input to CNNs, and the performance is compared with ML models trained on tabular data. Experiments conducted on the Derm7pt and PH2 datasets show that traditional ML models generally achieve the highest Area Under the Curve values, while LM-IGTD-based representations provide comparable performance and enable the use of CNNs on tabular clinical data used in dermatology.
- 3 March 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (5) , 2459
Open AccessTabular-to-Image Encoding Methods for Melanoma Detection: A Proof-of-Concept
Deep learning (DL) models have demonstrated strong performance in dermatological applications, particularly when trained on dermoscopic images. In contrast, tabular clinical data—such as patient metadata and lesion-level descriptors—are difficult to integrate into DL-based pipelines due to their heterogeneous, non-spatial, and often low-dimensional nature. As a result, these data are commonly handled using separate classical machine learning (ML) models. In this work, we present a proof-of-concept study that investigates whether dermatological tabular data can be transformed into two-dimensional image representations to enable convolutional neural network (CNN)-based learning. To this end, we employ the Low Mixed-Image Generator for Tabular Data (LM-IGTD), a framework designed to transform low-dimensional and heterogeneous tabular data into two-dimensional image representations, through type-aware encoding and controlled feature augmentation. Using this approach, we encode low-dimensional clinical metadata, high-dimensional lesion-level statistical features extracted from dermoscopic images, as well as their feature-level fusion, into grayscale image representations. The resulting image representations serve as input to CNNs, and the performance is compared with ML models trained on tabular data. Experiments conducted on the Derm7pt and PH2 datasets show that traditional ML models generally achieve the highest Area Under the Curve values, while LM-IGTD-based representations provide comparable performance and enable the use of CNNs on tabular clinical data used in dermatology.
- Open Access
- 30 January 2026
- Published by MDPI AG in Preprints.org
Tabular-to-Image Encoding Methods for Melanoma Detection: A Proof-of-Concept
Deep Learning (DL) models have demonstrated strong performance in dermatological applications, particularly when trained on dermoscopic images. In contrast, tabular clinical data—such as patient metadata and lesion-level descriptors—are difficult to integrate into DL-based pipelines due to their heterogeneous, non-spatial, and often low-dimensional nature. As a result, these data are commonly handled using separate classical machine learning (ML) models. In this work, we present a proof-of-concept study that investigates whether dermatological tabular data can be transformed into two-dimensional image representations to enable convolutional neural network (CNN)-based learning. To this end, we employ the Low Mixed-Image Generator for Tabular Data (LM-IGTD), a framework designed to transform low-dimensional and heterogeneous tabular data into two-dimensional image representations, through type-aware encoding and controlled feature augmentation. Using this approach, we encode low-dimensional clinical metadata, high-dimensional lesion-level statistical features extracted from dermoscopic images, as well as their feature-level fusion, into grayscale image representations. The resulting image representations serve as input to CNNs, and the performance is compared with ML models trained on tabular data. Experiments conducted on the Derm7pt and PH2 datasets show that traditional ML models generally achieve the highest Area Under the Curve values, while LM-IGTD-based representations provide comparable performance and enable the use of CNNs on structured clinical data used in dermatology.
- 30 January 2026
- Published by MDPI AG in Preprints.org
Open AccessTabular-to-Image Encoding Methods for Melanoma Detection: A Proof-of-Concept
Deep Learning (DL) models have demonstrated strong performance in dermatological applications, particularly when trained on dermoscopic images. In contrast, tabular clinical data—such as patient metadata and lesion-level descriptors—are difficult to integrate into DL-based pipelines due to their heterogeneous, non-spatial, and often low-dimensional nature. As a result, these data are commonly handled using separate classical machine learning (ML) models. In this work, we present a proof-of-concept study that investigates whether dermatological tabular data can be transformed into two-dimensional image representations to enable convolutional neural network (CNN)-based learning. To this end, we employ the Low Mixed-Image Generator for Tabular Data (LM-IGTD), a framework designed to transform low-dimensional and heterogeneous tabular data into two-dimensional image representations, through type-aware encoding and controlled feature augmentation. Using this approach, we encode low-dimensional clinical metadata, high-dimensional lesion-level statistical features extracted from dermoscopic images, as well as their feature-level fusion, into grayscale image representations. The resulting image representations serve as input to CNNs, and the performance is compared with ML models trained on tabular data. Experiments conducted on the Derm7pt and PH2 datasets show that traditional ML models generally achieve the highest Area Under the Curve values, while LM-IGTD-based representations provide comparable performance and enable the use of CNNs on structured clinical data used in dermatology.
- Open Access
- 12 November 2025
- Published by Springer Nature in Research Square
Multi-label transformer for dermoscopic artifact detection with an imbalance-aware and confidence-penalizing loss
Artifacts in dermoscopic images---such as rulers, gel bubbles, or ink stains---can compromise automated skin lesion analysis and reduce diagnostic reliability. Despite their clinical relevance, prior research has primarily targeted the removal or impact assessment of single artifact types, without addressing their automatic detection or co-occurrence modeling.In this work, we introduce the first deep learning framework for multi-label artifact detection in dermoscopy, leveraging the Classification Transformer (C-Tran) to model dependencies between co-occurring labels through attention mechanisms. To address class imbalance and overconfident predictions, we propose the Class-Adaptive Focal loss with Confidence penalty (CAFC), which adaptively weights underrepresented classes and penalizes miscalibrated outputs. On the ISIC-2019 dataset, the model trained with CAFC surpasses both Binary Cross-Entropy and Focal losses, yielding the highest overall F1-score (0.87) and recall (0.91). The proposed framework also generalizes effectively to the external PH2 dataset, achieving an overall F1-score of 0.86 and a recall of 0.91, confirming robustness across domains and imaging distribution shifts. Additionally, the approach provides attention-based interpretability, revealing the spatial regions associated with each artifact type and enhancing transparency. Beyond its strong performance, the proposed approach establishes a foundation for future artifact-aware systems that could assist in image quality control, preprocessing, and bias auditing within clinical diagnostic pipelines.
- 12 November 2025
- Published by Springer Nature in Research Square
Open AccessMulti-label transformer for dermoscopic artifact detection with an imbalance-aware and confidence-penalizing loss
Artifacts in dermoscopic images---such as rulers, gel bubbles, or ink stains---can compromise automated skin lesion analysis and reduce diagnostic reliability. Despite their clinical relevance, prior research has primarily targeted the removal or impact assessment of single artifact types, without addressing their automatic detection or co-occurrence modeling.In this work, we introduce the first deep learning framework for multi-label artifact detection in dermoscopy, leveraging the Classification Transformer (C-Tran) to model dependencies between co-occurring labels through attention mechanisms. To address class imbalance and overconfident predictions, we propose the Class-Adaptive Focal loss with Confidence penalty (CAFC), which adaptively weights underrepresented classes and penalizes miscalibrated outputs. On the ISIC-2019 dataset, the model trained with CAFC surpasses both Binary Cross-Entropy and Focal losses, yielding the highest overall F1-score (0.87) and recall (0.91). The proposed framework also generalizes effectively to the external PH2 dataset, achieving an overall F1-score of 0.86 and a recall of 0.91, confirming robustness across domains and imaging distribution shifts. Additionally, the approach provides attention-based interpretability, revealing the spatial regions associated with each artifact type and enhancing transparency. Beyond its strong performance, the proposed approach establishes a foundation for future artifact-aware systems that could assist in image quality control, preprocessing, and bias auditing within clinical diagnostic pipelines.
- 6 November 2025
- Published by Springer Nature
- p. 239-250
AI-Based Data Augmentation for Cleft Lip Assessment: Generating Realistic Synthetic Videos to Improve Clinical Outcomes
- 6 November 2025
- Published by Springer Nature
- p. 239-250
AI-Based Data Augmentation for Cleft Lip Assessment: Generating Realistic Synthetic Videos to Improve Clinical Outcomes
- 6 November 2025
- Published by Springer Nature
- p. 445-456
A Comparative Study of Machine Learning Models for Two-Tier Android Malware Classification with Dynamic Behavioral Analysis
- 6 November 2025
- Published by Springer Nature
- p. 445-456
A Comparative Study of Machine Learning Models for Two-Tier Android Malware Classification with Dynamic Behavioral Analysis
- Open Access
- 11 September 2025
- Published by Springer Nature in Research Square
Sun-Exposure and Lesion Location Bias in Deep Learning Models for Skin Cancer Detection
Background Deep learning (DL) models have demonstrated high performance in classifying skin lesions from dermoscopic images. However, the influence of photoexposure-related factors, such as level of exposure due to the anatomical site and skin phototype, on classification performance remains understudied. Investigating these factors is essential not only to understand their potential impact on bias in model predictions, but also to explore their potential role as risk indicators for skin cancer. Objective This study aims to assess the impact of skin phototype and anatomical-site–related photoexposure on the performance of DL models for malignancy detection, with a focus on potential sources of bias. Methods DL models are trained on widely used public dermoscopic image datasets. Performance is then evaluated on a recently published dataset of 60 patients from the University Hospital of North Norway, which includes dermoscopic images and clinical data on skin phototype and anatomical site to assess their impact on model performance. Results Preliminary analysis suggests that model performance varies between subgroups, with reduced precision observed in lesions chronically photoexposed. The reactions that cause red and painful skin are associated with better model performance, in addition to being mostly benign lesions. However, this result is skewed by the low incidence of malignant cases. Conclusions The findings highlight that individual sun-related behaviours and skin characteristics can influence the reliability of DL-based skin lesion identification. These results underscore the importance of evaluating model robustness across diverse patient profiles and may guide future efforts to define healthier sun exposure habits for the population.
- 11 September 2025
- Published by Springer Nature in Research Square
Open AccessSun-Exposure and Lesion Location Bias in Deep Learning Models for Skin Cancer Detection
Background Deep learning (DL) models have demonstrated high performance in classifying skin lesions from dermoscopic images. However, the influence of photoexposure-related factors, such as level of exposure due to the anatomical site and skin phototype, on classification performance remains understudied. Investigating these factors is essential not only to understand their potential impact on bias in model predictions, but also to explore their potential role as risk indicators for skin cancer. Objective This study aims to assess the impact of skin phototype and anatomical-site–related photoexposure on the performance of DL models for malignancy detection, with a focus on potential sources of bias. Methods DL models are trained on widely used public dermoscopic image datasets. Performance is then evaluated on a recently published dataset of 60 patients from the University Hospital of North Norway, which includes dermoscopic images and clinical data on skin phototype and anatomical site to assess their impact on model performance. Results Preliminary analysis suggests that model performance varies between subgroups, with reduced precision observed in lesions chronically photoexposed. The reactions that cause red and painful skin are associated with better model performance, in addition to being mostly benign lesions. However, this result is skewed by the low incidence of malignant cases. Conclusions The findings highlight that individual sun-related behaviours and skin characteristics can influence the reliability of DL-based skin lesion identification. These results underscore the importance of evaluating model robustness across diverse patient profiles and may guide future efforts to define healthier sun exposure habits for the population.
- Open Access
- 20 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 150756-150771
Evaluating Time Series Classification Models for Nocturnal Hypoglycemia: From Predictive Performance to Environmental Impact
Type 1 Diabetes (T1D) is an autoimmune condition that results in an insulin deficiency. People with T1D require the administration of exogenous insulin to maintain target glucose levels. However, insulin therapy can cause hypoglycemic episodes, which occur when blood glucose levels fall below 70 mg/dL. Nocturnal Hypoglycemia (NH) occurs while the individual is asleep and can lead to different clinical complications. Developing predictive approaches to predict NH before sleep could reduce these episodes and mitigate acute complications. While numerous models exist for Time Series Classification (TSC), their use for NH prediction remains limited. This study evaluates 14 different TSC models for NH prediction, assessing their performance by evaluating classification metrics, computational time, and environmental impact (measured by energy consumption and CO2 emissions). The approaches include distance-based, convolutional-based, deep learning, dictionary-based, feature-based, shapelet-based, and interval-based methods. We employed glucose data from 52 individuals with T1D. Experimental results showed that interval-based and feature-based approaches achieved the best predictive performance, obtaining the highest Area Under the Curve Operator (AUCROC) of 0.703. Additionally, both demonstrated low environmental impact due to their short computational time. However, substantial differences in environmental impact were observed depending on the approach. Distance-based methods and deep learning approaches exhibited the highest environmental impact. This paper provides key insights into the effectiveness of TSC models for NH prediction, highlighting the trade-off between model performance and environmental impact.
- 20 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 150756-150771
Open AccessEvaluating Time Series Classification Models for Nocturnal Hypoglycemia: From Predictive Performance to Environmental Impact
Type 1 Diabetes (T1D) is an autoimmune condition that results in an insulin deficiency. People with T1D require the administration of exogenous insulin to maintain target glucose levels. However, insulin therapy can cause hypoglycemic episodes, which occur when blood glucose levels fall below 70 mg/dL. Nocturnal Hypoglycemia (NH) occurs while the individual is asleep and can lead to different clinical complications. Developing predictive approaches to predict NH before sleep could reduce these episodes and mitigate acute complications. While numerous models exist for Time Series Classification (TSC), their use for NH prediction remains limited. This study evaluates 14 different TSC models for NH prediction, assessing their performance by evaluating classification metrics, computational time, and environmental impact (measured by energy consumption and CO2 emissions). The approaches include distance-based, convolutional-based, deep learning, dictionary-based, feature-based, shapelet-based, and interval-based methods. We employed glucose data from 52 individuals with T1D. Experimental results showed that interval-based and feature-based approaches achieved the best predictive performance, obtaining the highest Area Under the Curve Operator (AUCROC) of 0.703. Additionally, both demonstrated low environmental impact due to their short computational time. However, substantial differences in environmental impact were observed depending on the approach. Distance-based methods and deep learning approaches exhibited the highest environmental impact. This paper provides key insights into the effectiveness of TSC models for NH prediction, highlighting the trade-off between model performance and environmental impact.
- Open Access
- 8 April 2025
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 165, 104821
Transfer learning for a tabular-to-image approach: A case study for cardiovascular disease prediction
- 8 April 2025
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 165, 104821
Open AccessTransfer learning for a tabular-to-image approach: A case study for cardiovascular disease prediction