This paper presents a comprehensive review of loss functions and performance metrics in deep learning, highlighting key developments and practical insights across diverse application areas. We begin by outlining fundamental considerations in classic tasks such as regression and classification, then extend our analysis to specialized domains like computer vision and natural language processing including retrieval-augmented generation. In each setting, we systematically examine how different loss functions and evaluation metrics can be paired to address task-specific challenges such as class imbalance, outliers, and sequence-level optimization. Key contributions of this work include: (1) a unified framework for understanding how losses and metrics align with different learning objectives, (2) an in-depth discussion of multi-loss setups that balance competing goals, and (3) new insights into specialized metrics used to evaluate modern applications like retrieval-augmented generation, where faithfulness and context relevance are pivotal. Along the way, we highlight best practices for selecting or combining losses and metrics based on empirical behaviors and domain constraints. Finally, we identify open problems and promising directions, including the automation of loss-function search and the development of robust, interpretable evaluation measures for increasingly complex deep learning tasks. Our review aims to equip researchers and practitioners with clearer guidance in designing effective training pipelines and reliable model assessments for a wide spectrum of real-world applications.

Digital transformation is reshaping the healthcare field by streamlining diagnostic workflows and improving disease management. Within this transformation, Digital Twins (DTs), which are virtual representations of physical systems continuously updated by real-world data, stand out for their ability to capture the complexity of human physiology and behavior. When coupled with Artificial Intelligence (AI), DTs enable data-driven experimentation, precise diagnostic support, and predictive modeling without posing direct risks to patients. However, their integration into healthcare requires careful consideration of ethical, regulatory, and safety constraints in light of the sensitivity and nonlinear nature of human data. In this review, we examine recent progress in DTs over the past seven years and explore broader trends in AI-augmented DTs, focusing particularly on movement rehabilitation. Our goal is to provide a comprehensive understanding of how DTs bolstered by AI can transform healthcare delivery, medical research, and personalized care. We discuss implementation challenges such as data privacy, clinical validation, and scalability along with opportunities for more efficient, safe, and patient-centered healthcare systems. By addressing these issues, this review highlights key insights and directions for future research to guide the proactive and ethical adoption of DTs in healthcare.

Crop and weed discrimination in natural field environments is still challenging for implementing automatic agricultural practices, such as weed control. Some weed control methods have been proposed. However, these methods are still restricted as they are implemented under controlled conditions. The development of a sound weed control system begins by recognizing the crop and the different weed plants presented in the field. In this work, a classification approach of Zea mays L. (Crop), narrow-leaf weeds (NLW), and broadleaf weeds (BLW) from multi-plant images are presented. Moreover, a large image dataset was generated. Images were captured in natural field conditions, in different locations, and growing stages of the plants. The extraction of regions of interest (ROI) is carried out employing connected component analysis (CCA), whereas the classification of ROIs is based on Convolutional Neural Networks (CNN) and compared with a shallow learning approach. To measure the classification performance of both methods, accuracy, precision, recall, and F1-score metrics were used. The best alternative for the weed classification task at early stages of growth and in natural corn field environments was the CNN-based approach, as indicated by the 97% accuracy value obtained.

Automatic sign language recognition is a challenging task in machine learning and computer vision. Most works have focused on recognizing sign language using hand gestures only. However, body motion and facial gestures play an essential role in sign language interaction. Taking this into account, we introduce an automatic sign language recognition system based on multiple gestures, including hands, body, and face. We used a depth camera (OAK-D) to obtain the 3D coordinates of the motions and recurrent neural networks for classification. We compare multiple model architectures based on recurrent networks such as Long Short-Term Memories (LSTM) and Gated Recurrent Units (GRU) and develop a noise-robust approach. For this work, we collected a dataset of 3000 samples from 30 different signs of the Mexican Sign Language (MSL) containing features coordinates from the face, body, and hands in 3D spatial coordinates. After extensive evaluation and ablation studies, our best model obtained an accuracy of 97% on clean test data and 90% on highly noisy data.

Background: Oral diseases such as caries, gingivitis, and periodontitis are highly prevalent worldwide and often arise from plaque. This study focuses on detecting three plaque stages—new, mature, and over-mature—using state-of-the-art YOLO architectures to enhance early intervention and reduce reliance on manual visual assessments. Methods: We compiled a dataset of 531 RGB images from 177 individuals, captured via multiple mobile devices. Each sample was treated with disclosing gel to highlight plaque types, then preprocessed for lighting and color normalization. YOLOv9, YOLOv10, and YOLOv11, in various scales, were trained to detect plaque categories, and their performance was evaluated using precision, recall, and mean Average Precision (mAP@50). Results: Among the tested models, YOLOv11m achieved the highest mAP@50 (0.713), displaying superior detection of over-mature plaque. Across all YOLO variants, older plaque was generally easier to detect than newer plaque, which can blend with gingival tissue. Applying the O’Leary index indicated that over half of the study population exhibited severe plaque levels. Conclusions: Our findings demonstrate the feasibility of automated plaque detection with advanced YOLO models in varied imaging conditions. This approach offers potential to optimize clinical workflows, support early diagnoses, and mitigate oral health burdens in low-resource communities.

Three-dimensional (3D) reconstruction from images has significantly advanced due to recent developments in deep learning, yet methodological variations and diverse application contexts pose ongoing challenges. This systematic review examines the state-of-the-art deep learning techniques employed for image-based 3D reconstruction from 2019 to 2025. Through an extensive analysis of peer-reviewed studies, predominant methodologies, performance metrics, sensor types, and application domains are identified and assessed. Results indicate multi-view stereo and monocular depth estimation as prevailing methods, while hybrid architectures integrating classical and deep learning techniques demonstrate enhanced performance, especially in complex scenarios. Critical challenges remain, particularly in handling occlusions, low-texture areas, and varying lighting conditions, highlighting the importance of developing robust, adaptable models. Principal conclusions highlight the efficacy of integrated quantitative and qualitative evaluations, the advantages of hybrid methods, and the pressing need for computationally efficient and generalizable solutions suitable for real-world applications.