CRIS

Permanent URI for this communityhttps://scripta.up.edu.mx/handle/20.500.12552/1

Browse

Settings

Author: search.filters.author.Renza, DiegoHas files: Yes

Search Results

Now showing 1 - 4 of 4
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    A Deep-Learning and Bio-inspired Vision Model-Based Approach for Automatic Coronary Arteries Segmentation
    (IEEE, 2025)
    López-Figueroa, Alberto
    ;
    Brieva, Jorge  
    ;
    Moya-Albor, Ernesto  
    ;
    Gomez-Coronel, Sandra L.
    ;
    Renza, Diego
    The accurate segmentation of coronary arteries from X-ray angiograms is critical for the diagnosis and treatment of cardiovascular diseases. Yet, it remains a challenging task due to low image contrast and complex vessel structures. This paper introduces a novel hybrid methodology that combines a bio-inspired vision model, the Steered Hermite Transform (SHT), with a deep learning architecture for robust vessel segmentation. We leverage the SHT to decompose each angiogram into a rich, multi-resolution set of 15 feature maps that capture local image structures at different scales and orientations. These Hermite coefficients, along with the original image, form a 16-channel input tensor used to train a U-Net. This approach enables the network to learn from an enhanced feature space that explicitly represents vessel-like patterns. Evaluated on a public dataset of 134 coronary angiograms, our model demonstrates outstanding performance, achieving an Area Under the Curve (AUC) of 0.9872. The results confirm that enriching the input of a deep neural network with SHT coefficients significantly improves its ability to identify and segment complex vascular networks accurately. ©The authors ©IEEE.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Complexity-Driven Adversarial Validation for Corrupted Medical Imaging Data
    (MDPI AG, 2026)
    Renza, Diego
    ;
    Brieva, Jorge  
    ;
    Moya-Albor, Ernesto  
    Distribution shifts commonly arise in real-world machine learning scenarios in which the fundamental assumption that training and test data are drawn from independent and identically distributed samples is violated. In the case of medical data, such distribution shifts often occur during data acquisition and pose a significant challenge to the robustness and reliability of artificial intelligence systems in clinical practice. Additionally, quantifying these shifts without training a model remains a key open problem. This paper proposes a comprehensive methodological framework for evaluating the impact of such shifts on medical image datasets under artificial transformations that simulate acquisition variations, leveraging the Cumulative Spectral Gradient (CSG) score as a measure of multiclass classification complexity induced by distributional changes. Building on prior work, the proposed approach is meaningfully extended to twelve 2D medical imaging benchmarks from the MedMNIST collection, covering both binary and multiclass tasks, as well as grayscale and RGB modalities. We evaluate the metric analyzing its robustness to clinically inspired distribution shifts that are systematically simulated through motion blur, additive noise, brightness and contrast variation, and sharpness variation, each applied at three severity levels. This results in a large-scale benchmark that enables a detailed analysis of how dataset characteristics, transformation types, and distortion severity influence distribution shifts. Thus, the findings show that while the metric remains generally stable under noise and focus distortions, it is highly sensitive to variations in brightness and contrast. On the other hand, the proposed methodology is compared against Cleanlab’s widely used Non-IID score on the RetinaMNIST dataset using a pre-trained ResNet-50 model, including both class-wise analysis and correlation assessment between metrics. Finally, interpretability is incorporated through class activation map analysis on BloodMNIST and its corrupted variants to support and contextualize the quantitative findings. ©The authors ©MDPI.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Challenges and Advances in Digital Processing of Fetal Phonocardiography Signal: A Review
    (Springer Nature Switzerland, 2025)
    Moya-Albor, Ernesto  
    ;
    Brieva, Jorge  
    ;
    Gomez-Coronel, Sandra L.
    ;
    Renza, Diego
    This chapter presents a state-of-the-art review of different investigations focused on Fetal Phonocardiography (fPCG). fPCG signals allow the identification of the fetus’s cardiac alterations during pregnancy through a noninvasive and secure approach. However, fPCG signals present some challenges, for example: very weak signal sources, high levels of noise, source mixing, and significant signal attenuation. This work provides a review of available fPCG datasets and the methods proposed for source separation, extraction, and filtering of fPCG signals, as well as the methods for estimating fetal heart rate (fHR) and detecting fetal Heart Sounds (fHS). Additionally, since it is sometimes necessary to transmit or store fPCG signals, the chapter also discusses signal compression approaches and applications involving fPCG signals. ©The authors ©Springer.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Publication,
    Edge-Enhanced Knowledge Distillation System for Diabetic Retinopathy Lesions Computer-Aided Diagnosis
    (Springer Nature Switzerland, 2025)
    Lopez-Figueroa, Alberto
    ;
    Jacome-Herrera, Sebastian
    ;
    Moya-Albor, Ernesto  
    ;
    Renza, Diego
    ;
    Brieva, Jorge  
    This work addresses the challenge of deploying computationally intensive Deep Learning (DL) models for Diabetic Retinopathy (DR) lesion detection in clinical settings, particularly on resource-constrained edge devices. DR is a significant global health issue and a leading cause of preventable blindness, making early and accessible detection crucial. We propose a proof-of-concept system utilizing Knowledge Distillation (KD) to create a tiny, efficient DL model for DR lesion detection, specifically designed for embedding into retinal scanners via the NVIDIA Jetson Nano platform. Our novel approach employs a KD framework where a pre-trained Inception-v3 model acts as the ‘teacher,’ fine-tuned on fundus image data. This teacher model distills its knowledge into a compact ‘student’ model based on the MobileNet-v2 architecture, which is trained on a small, synthetically generated dataset optimized through an iterative distillation process using a custom loss function combining Kullback-Leibler divergence and Categorical Cross-Entropy. This method significantly reduces model size and computational requirements while maintaining high diagnostic accuracy, comparable to larger, state-of-the-art models. By enabling real-time, on-device analysis, this embedded AI solution enhances data privacy, ensures consistent performance, and improves the accessibility of advanced DR screening, particularly in remote or underserved healthcare environments. This makes AI-assisted DR detection feasible for widespread clinical adoption directly within scanning devices. ©The authors ©Springer.