The integration of deep learning techniques into sports performance analysis has significantly advanced athlete monitoring, motion tracking, and predictive modelling. These advancements have significantly improved the ability to assess performance, optimize training strategies, and reduce injury risks. However, despite notable progress, challenges remain in standardizing methodologies, ensuring model reliability, and enhancing real-time application across various sports disciplines. We conducted a systematic literature search of Web of Science Core Collection (WOS), China National Knowledge Infrastructure (CNKI), and Association for Computing Machinery Digital Library (ACM DL) for relevant studies published from 2015 to 2024, with no language restrictions. Eligible studies were those that explicitly applied deep learning techniques (such as convolutional and recurrent neural networks) to sports performance analysis tasks (e.g., action recognition and classification, motion detection and tracking, injury prediction) and reported their methodology and performance metrics. Key data, including sport type, application domain, and model type, were extracted for narrative synthesis. A total of 51 studies met the inclusion criteria, covering a broad range of individual and team sports. Deep learning techniques in sports performance analysis were chiefly employed for action recognition, object detection and multi-target tracking, target classification, and performance or injury prediction. CNNs were the most common models for visual recognition tasks, while RNNs (including LSTMs) were frequently used for temporal sequence data. Most studies reported improved performance outcomes with deep learning; however, we observed considerable variability in data quality, model validation approaches, and cross-sport generalizability. Deep learning has demonstrated transformative potential in optimizing sports performance analysis by providing automated, data-driven insights. Future research should prioritize integrating multi-modal data sources, refining real-time analytics, and improving the adaptability of deep learning techniques across different sports contexts to support more precise and data-driven performance assessments.

Steganography is the art and practice of communication using hidden messages. The least significant bits (LSB) based method is the well-known type of steganography in the spatial domain. Usually, achieving the larger embedding capacity in LSB-based methods requires a large number of LSB bits modification which indirectly reduces the visual quality of stego-image and increases the risk of steganalysis detection attacks. In this study, we propose a novel steganography method with data mapping strategy which can reduce the number of bits modification per pixel. In the proposed method, four secret data bits are mapped with the four most significant bits of a cover pixel. Furthermore, the only two LSBs of a pixel are modified to indicate the mapping strategy. Experimental results show that the proposed method is able to achieve 3.48% larger embedding capacity while enhancing the visual quality (i.e., peak signal to noise ratio (PSNR) 3.73 dB) and reducing the modification of 0.76 bits per pixel. Moreover, the proposed method provides security against basic Regular and Singular groups (RS) steganalysis and histogram steganalysis detection attacks.

Dental impression tray is frequently used in dentistry to record the patient’s oral structure for clinical oral diagnosis and treatment planning. Manual procedure of taking impressions is costly, time-consuming, and additionally, no research has been done on selecting dental impression tray from dental arch images using computer vision in real-life scenarios. In this spirit, an intelligent model is proposed based on computer vision and machine learning to select appropriate dental impression trays from maxillary arch images. A dataset of 52 patients’ maxillary arch images have been acquired and various sets of features such as colors, textures, and shapes of the images were extracted to better characterize the maxillary arch images. Considering the importance of the features in describing the maxillary arch object and to improve the classification performance, a method based on multi-feature fusion with ensemble classifier is proposed. Besides, the performance of a deep learning based multilayer perceptron neural network is also investigated. The proposed multi-feature fusion with ensemble classifier attained 92.31% precision, 91.75% recall, 91.75% accuracy, respectively, on the dataset, which clearly establishes the feasibility of the proposed model. An illustration of a real-life application of the proposed model is also provided.

Background/Objectives: Patients with syndromic craniosynostosis (SC) pose a significant challenge for post-operational outcomes due to the variability in craniofacial deformities and gain-of-function characteristics. This study aims to develop validated predictive tools using stable cranial base variables to predict changes in the midfacial region and explore the craniofacial morphology among patients with SC. Methods: This study involved 17 SC patients under 12 years old, 17 age-matched controls for morphological analysis, and 21 normal children for developing craniofacial predictive models. A stable cranial base and changeable midfacial variables were analyzed using the Mann–Whitney U test. Pearson correlation identified linear relationships between the midface and cranial base variables. Multicollinearity was checked before fitting the data with multiple linear regression for growth prediction. Model adequacy was confirmed and the 3-fold cross-validation ensured results reliability. Results: Patients with SC exhibited a shortened cranial base, particularly in the middle cranial fossa (S-SO), and a sharper N-S-SO and N-SO-BA angle, indicating a downward rotation and kyphosis. The midface length (ANS-PNS) and zygomatic length (ZMs-ZTi) were significantly reduced, while the midface width (ZFL-ZFR) was increased. Regression models for the midface length, width, and zygomatic length were given as follows: ANS-PNS = 23.976 + 0.139 S-N + 0.545 SO-BA − 0.120 N-S-BA + 0.078 S-SO-BA + 0.051 age (R² = 0.978, RMSE = 1.058); ZFL-ZFR = −15.618 + 0.666 S-N + 0.241 N-S-BA + 0.155 S-SO-BA + 0.121 age (R² = 0.903, RMSE = 3.158); and ZMs-ZTi = −14.403 + 0.765 SO-BA + 0.266 N-S-BA + 0.111 age (R² = 0.878, RMSE = 3.720), respectively. Conclusions: The proposed models have potential applications for midfacial growth estimation in children with SC.