Arrhythmias are among the diseases with high mortality rates worldwide, causing millions of deaths each year. This underscores the importance of real-time electrocardiogram (ECG) monitoring for timely heart disease diagnosis and intervention. Deep learning models, trained on ECG signals across twelve or more leads, are the predominant approach for automated arrhythmia detection in the AI-assisted medical field. While these multi-lead ECG-based models perform well in automatic arrhythmia detection, their complexity often restricts their use on resource-constrained devices. In this paper, we propose an efficient, lightweight arrhythmia classification model using a knowledge distillation technique to train a student model from a teacher model, tailored for embedded intelligence in wearable devices. The results show that the student model achieves 96.32% accuracy, which is comparable to the teacher model, with a remarkable compression ratio that is 1242.58 times smaller, outperforming other lightweight models. Enabled by the proposed model, we developed a wearable ECG monitoring system based on the STM32F429 Discovery kit and ADS1292R chip, achieving real-time arrhythmia detection on small wearable devices.

Bowel sounds, a reflection of the gastrointestinal tract’s peristalsis, are essential for diagnosing and monitoring gastrointestinal conditions. However, the absence of an effective, non-invasive method for assessing digestion through auscultation has resulted in a reliance on time-consuming and laborious manual analysis by clinicians. This study introduces an innovative deep learning-based method designed to automate and enhance the recognition of bowel sounds. Our approach integrates the Branchformer architecture, which leverages the power of self-attention and convolutional gating for robust feature extraction, with a self-supervised pre-training strategy. Specifically, the Branchformer model employs parallel processing of self-attention and convolutional gated Multi-layer Perceptron branches to capture both global and local dependencies in audio signals, thereby enabling effective characterization of complex bowel sound patterns. Furthermore, a self-supervised pre-training strategy is employed, leveraging a large corpus of unlabeled audio data to learn general sound wave representations, followed by fine-tuning on a limited set of bowel sound data to optimize the model’s recognition performance for specific tasks. Experimental results on public bowel sound datasets demonstrate the superior recognition performance of the proposed method compared to existing baseline models, particularly under data-limited conditions, thereby confirming the effectiveness of the self-supervised pre-training strategy. This work provides an efficient and automated solution for clinical bowel sound monitoring, facilitating early diagnosis and treatment of gastrointestinal disorders.

Fundus images play a crucial role in diagnosing ocular and systemic vascular diseases in clinical settings. However, these images often suffer from issues like uneven brightness and transmission distortion, which adversely impact their quality. To address these challenges, this paper proposes a novel approach called the Probability Conditional Guidance Generative Adversarial Network (PCGGAN). This method leverages the high-frequency components of the images to enhance their quality. The hypothesis behind this approach is that retaining the content and pathological information associated with the high-frequency components is vital during the enhancement process. To achieve this, the fundus image is decomposed into high-frequency and low-frequency components, with the high-frequency components probabilistically used as conditional guidance during model training. This training strategy enables the model to learn the distinctions between unconditional and conditional constraints, thereby enhancing its capability for targeted generation and generalization. The effectiveness of the proposed method is validated through both quantitative and qualitative experiments.

Scleral segmentation and scleral vessel segmentation are increasingly recognized as critical components in medical image analysis, with broad applications in ocular disease diagnosis and biometric identification. In particular, scleral vessel segmentation contributes significantly to the early detection of conditions such as diabetic retinopathy and glaucoma. However, the intricate structure of scleral vessels and the scarcity of high-quality annotated datasets continue to present major challenges. To address these issues, an ensemble deep learning-based approach is proposed, integrating three segmentation models—UNet, NestNet, and DeepLabv3_Plus—to evaluate and quantitatively analyze the unified task of scleral segmentation and scleral vessel segmentation. The input ocular images undergo preprocessing steps including denoising, contrast-limited adaptive histogram equalization, and cropping. Scleral regions are extracted to enhance the explicit representation of vascular structures. Experiments are conducted on the publicly available SBVPI dataset. The DeepLabv3_Plus model achieves the highest performance in scleral segmentation, with an accuracy of 0.9656, sensitivity of 0.9694, specificity of 0.9421, and a Dice coefficient of 0.9723. For scleral vessel segmentation, the same model achieves an accuracy of 0.9285 and a specificity of 0.9536. These results demonstrate the effectiveness of the proposed ensemble framework and highlight its potential for advancing scleral vessel segmentation research. Future work will focus on further model optimization and the exploration of clinical and real-world deployment scenarios. Received: 13 March 2025 | Revised: 2 July 2025 | Accepted: 25 July 2025 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support this work are available upon reasonable request to the first author, Yongbin Qi, at peninsulazz@163.com. Author Contribution Statement Yongbin Qi: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization, Project administration. Baochen Zhen: Investigation, Writing – review & editing. Jiaming Wang: Investigation. Shilin Zhao: Investigation. Yansuo Yu: Conceptualization, Methodology, Formal analysis, Resources, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition. Qiang Liu: Resources, Supervision, Funding acquisition.

The objective of this study is to analyze the potential preoperative factors affecting post-cataract visual acuity by machine learning, and establish a deep neural network to predict post-cataract visual acuity. 187 clinical data of patients were collected to investigate the relationship between preoperative factors and postoperative visual acuity. The impact of these preoperative factors on postoperative visual acuity was assessed through the coefficient of variation-based method, Pearson correlation coefficient-based method, random forest, and category boosting. Furthermore, we established a deep neural network to predict postoperative visual acuity, which validated the importance of the preoperative factors selected. It was found that axial length, cylinder, preoperative visual acuity, age, and keratometry reading two were critical preoperative factors associated with postoperative visual outcomes. Keratometry readings one, mean corneal radius, spherical equivalent, anterior chamber depth, intraocular lens, and reserved degree were relatively less critical preoperative factors. Experiments on postoperative visual acuity prediction show that machine learning methods based on preoperative factors achieve superior performance, with predictions based on critical preoperative factors outperforming those based on all preoperative factors. In addition, the proportion of the absolute error which is less than 0.225 between the predicted and the actual postoperative visual acuity based on the deep neural network is 78.95 percent. The study shows how to extract critical preoperative factors related to postoperative visual acuity through machine learning, and build a deep neural network to predict postoperative visual acuity using critical preoperative factors.