Volume 05 · Issue 01 · 2025

With the rapid advancement of digitalization and intelligence in the medical field, a plethora of cutting-edge technologies are gradually being applied to revolutionize healthcare. In the medical data security privacy protection and artificial intelligence encryption computing, blockchain stands out due to its inherent characteristics of traceability, tamper-proofing, and high credibility. Although blockchain technology has been applied in various industries, its application in the medical field needs more driving force, and its development needs to be standardized. This clinical practice guideline is developed following the World Health Organization’s recommended process, adopting Grading of Recommendations Assessment, Development and Evaluation in assessing evidence quality. Considering the integration of blockchain and medical scenarios, we focus on the value and implementability of practical medical applications and provide the guidance on the construction and application of medical blockchain. This practice guideline includesd 10 potential medical application scenarios and usage frameworks. It is worth highlighting that electronic medical record sharing, drug and device anti-counterfeiting, medical digital intellectual property protection, and public health management are considered to be the most easily implemented and effective medical scenarios. The recommendations in this guideline were formulated based on the consideration of stakeholder values and preferences, resource utilization, feasibility, and acceptability, may have a profound impact on the construction of medical blockchain-related scenarios in China and internationally.

Practice Guidance Registration for Transparency (PREPARE) website (http://www.guidelines-registry.cn) Registration No. PREPARE-2023CN637.

This editorial presents an optimistic yet cautious perspective on the development, deployment, and regulation of large language models (LLMs) in the field of medicine. It is essential to strike a balance between embracing the benefits of artificial intelligence-driven solutions and preserving the human touch that is vital for providing compassionate care. The exponential growth of medical data has paved the way for the integration of LLMs into healthcare, offering unprecedented opportunities to enhance clinical decision-making and alleviate physicians’ workloads. Recently, LLMs have exhibited remarkable potential across various clinical scenarios, including streamlining diagnostic processes, optimizing radiology reports, and providing personalized treatment recommendations. However, the implementation of LLMs in healthcare is not without its challenges. Issues such as the scarcity of high-quality annotated data, privacy concerns, and the risk of generating misleading or overconfident information are significant hurdles that must be addressed. Moreover, while LLMs can replace certain basic tasks traditionally performed by humans, it is crucial to recognize that senior clinicians play an irreplaceable role in complex decision-making and providing emotional support to patients. By harnessing the power of LLMs to augment human capabilities while maintaining essential human elements within healthcare, we might shape a future where artificial intelligence and human intelligence coexist harmoniously. Prioritizing ethical development and deployment for artificial intelligence, empowering healthcare professionals, and safeguarding patient privacy will be key to realizing the full potential of LLMs in revolutionizing healthcare delivery. Through ongoing research, collaboration, and adaptation, responsible integration of LLMs holds promise for elevating both quality and accessibility globally, ultimately creating a more efficient, personalized, and patient-centric healthcare system.

Accurately identifying and localizing the five subtypes of intracranial hemorrhage (ICH) are crucial steps for subsequent clinical treatment; however, the lack of a large computed tomography (CT) dataset with annotations of the categorization and localization of ICH considerably limits the development of deep learning-based identification and localization methods. We aimed to construct this large dataset and develop a deep learning-based model to identify and localize the five ICH subtypes, including intraventricular hemorrhage (IVH), intraparenchymal hemorrhage (IPH), subdural hemorrhage (SDH), subarachnoid hemorrhage (SAH), and epidural hemorrhage (EDH), in non-contrast head CT scans.

Based on the public Radiological Society of North America (RSNA) 2019 dataset, we constructed a large CT dataset named RSNA 2019+ that was annotated for bleeding localization of the five ICH subtypes by three radiologists. An improved YOLOv8 architecture with the bidirectional feature pyramid network was proposed and trained using the RSNA 2019+ training dataset and evaluated on the RSNA 2019+ test dataset. The public CQ500, and two private datasets collected from the Xinqiao and Sunshine Union Hospitals, respectively, were also annotated to perform multicenter validation. Furthermore, the performance of the deep learning model was compared with that of four radiologists. Multiple performance metrics, including the average precision (AP), precision, recall and F1-score, were used for performance evaluation. The McNemar and chi-squared tests were performed, and the 95% Wilson confidence intervals were given for the precision and recall.

There were 175,125; 4,707; 8,259; and 3,104 bounding boxes after annotation on the RSNA 2019+; CQ500+; and the PD 1 and PD 2 datasets, respectively. With an intersection-over-union threshold of 0.5, the APs of IVH, IPH, SAH, SDH and EDH are 0.852, 0.820, 0.574, 0.639, and 0.558, respectively, yielding a mean average precision (mAP) of 0.688 for our proposed deep learning model on the RSNA 2019+ test dataset. For the multicenter validation involving the three external datasets, the mAPs for CQ500, PD1, and PD2 were 0.594, 0.734, and 0.66, respectively, which is comparable to those of radiologist with eight years of experience in head CT interpretation.

The deep learning model developed from the constructed RSNA 2019+ dataset exhibited good potential in identifying and localizing the five ICH subtypes in CT slices and has the potential to assist in the clinical diagnosis.

Intracranial aneurysm imaging artificial intelligence (AI) products have entered the clinical implementation phase, but the application status of them in Chinese hospitals remains unclear. A nationwide survey was conducted to explore the current status of intracranial aneurysm imaging AI products in hospitals across China.

Delphi method was used to develop a questionnaire, which was then distributed to the radiologists across China between September 3rd and 10th, 2023. Independent predictors of the adoption of these AI products, radiologists’ attitudes, concerns and knowledge about these AI products were evaluated using logistic regression. Participants were categorized into seven groups based on Chinese geographical regions to compare the performance of these AI products in different geographical regions.

After 3 rounds of Delphi discussion by 29 radiologists, the questionnaire was derived. A total of 961 radiologists from 777 different hospitals in 31 provinces across China completed the questionnaire. Among these hospitals, 45.4% (353/777) had introduced intracranial aneurysm imaging AI products. The most commonly reported concern with these AI products was poor specificity (265/446, 59.4%). The majority of respondents had basic (310/961, 42.0%) or intermediate (331/961, 44.9%) knowledge of AI products and they held positive attitudes (913/961, 95.0%) towards using them. Those who had received AI training were more likely to possess a higher level of knowledge about AI (odds ratio (OR) = 1.80, P = 0.04). For regional comparison, respondents in Central China and East China gave the highest ratings to the accuracy (OR = 2.41, P = 0.048 vs. OR=2.36, P = 0.02) and specificity (OR = 2.34, P = 0.046 vs. OR = 2.37, P = 0.02) of these AI products.

The intracranial aneurysm imaging AI products may be widely used in Chinese hospitals but vary by clinical scenarios and geographic position.

Accurate measurement of left ventricular ejection fraction (LVEF) is crucial in diagnosing and managing cardiac conditions. Deep learning (DL) models offer potential to improve the consistency and efficiency of these measurements, reducing reliance on operator expertise.

The aim of this study was to develop an innovative software-hardware combined device, featuring a novel DL algorithm for the automated quantification of LVEF from 2D echocardiographic images.

A dataset of 2,113 patients admitted to the Affiliated Hospital of Qingdao University between January and June 2023 was assembled and split into training and test groups. Another 500 patients from another campus were prospectively collected as external validation group. The age, sex, reason for echocardiography and the type of patients were collected. Following standardized protocol training by senior echocardiographers using domestic ultrasound equipment, apical four-chamber view images were labeled manually and utilized for training our deep learning framework. This system combined convolutional neural networks (CNN) with transformers for enhanced image recognition and analysis. Combined with the model that was named QHAutoEF, a 'one-touch’ software module was developed and integrated into the echocardiography hardware, providing intuitive, real-time visualization of LVEF measurements. The device’s performance was evaluated with metrics such as the Dice coefficient and Jaccard index, along with computational efficiency indicators. The dice index, intersection over union, size, floating point operations per second and calculation time were used to compare the performance of our model with alternative deep learning architectures. Bland-Altman analysis and the receiver operating characteristic (ROC) curve were used for validation of the accuracy of the model. The scatter plot was used to evaluate the consistency of the manual and automated results among subgroups.

Patients from external validation group were older than those from training group ((60±14) years vs. (55±16) years, respectively, P < 0.001). The gender distribution among three groups were showed no statistical difference (43 % vs. 42 % vs. 50 %, respectively, P = 0.095). Significant differences were showed among patients with different type (all P < 0.001) and reason for echocardiography (all P <0.001 except for other reasons). QHAutoEF achieved a high Dice index (0.942 at end-diastole, 0.917 at end-systole) with a notably compact model size (10.2 MB) and low computational cost (93.86 G floating point operations (FLOPs)). It exhibited high consistency with expert manual measurements (intraclass correlation coefficient (ICC) =0.90 (0.89, 0.92), P < 0.001) and excellent capability to differentiate patients with LVEF ≥60 % from those with reduced function, yielding an area under the operation curve (AUC) of 0.92 (0.90–0.95). Subgroup analysis showed a good correlation between QHAutoEF results and manual results from experienced experts among patients of different types (R = 0.93, 0.73, 0.92, respectively, P <0.001) and ages (R = 0.92, 0.94, 0.89, 0.91, 0.81, respectively, P <0.001).

Our software-hardware device offers an improved solution for the automated measurement of LVEF, demonstrating not only high accuracy and consistency with manual expert measurements but also practical adaptability for clinical settings. This device might potentially support clinicians and augment clinical decision.

Monkeypox has become a significant public health concern owing to the recent epidemics and associated morbidity. The treatment is limited by the availability of drugs, especially in endemic communities. Computational methods can facilitate the discovery and development of new and effective therapies that are affordable. This study was aimed at identifying potential drug candidates from the SuperNatural chemical library against monkeypox virus essential proteins using computational methods.

We identified 7 highly conserved essential proteins involved in monkeypox virus (MPXV) replication, infectivity, and propagation as potential therapeutic targets. A library of 447 orally administrable drug-like compounds from the SuperNatural database was screened against the proteins for potential binders/ligands associations using virtual screening and molecular dynamics simulations.

Our search identified hit compounds that mimicked the tecovirimat binding pose and outperformed it in binding affinity. Notably, resveratrol-3-O-D-glucopyranoside showed significant binding affinity to the viral protein F13L, a key protein involved in MPXV transmission. Extensive molecular dynamics simulations showed stable interactions between resveratrol-3-O-β-D-glucopyranoside and F13L, and other hit compounds with their respective targets.

Although the predicted interactions require further experimental validation, our results suggested that the identified compounds could be promising therapeutic candidates for the development of novel monkeypox drugs. These findings might underscore the significance of natural compounds in drug discovery and lay the foundation for developing novel antivirals against monkeypox.

Hypertension is a critical medical condition that increases the risks of many fatal diseases. Early detection of hypertension can be crucial to lead a healthy life. Machine learning (ML) can be useful for the early prediction of a patient’s likelihood of having a blood pressure abnormality and preventing it. Explainable artificial intelligence (XAI) is a state-of-the-art ML toolset that helps us understand and explain the prediction of an ML model. This research aims to build an automatic blood pressure anomaly detection system with maximum accuracy using the fewest features and learn why a model arrived at a particular result using XAI.

This study utilized the "Blood Pressure Data for Disease Prediction" dataset from Kaggle. Data were collected from medical reports of random participants in 2019 based on the presence of blood pressure abnormality, chronic kidney disease, and adrenal and thyroid disorders. We have used several ML algorithms (extreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM), decision tree (DT), and logistic regression (LR)) to predict blood pressure abnormality based on patient’s data. Principal component analysis (PCA) and recursive feature elimination (RFE) algorithms were used as feature optimizers. Key outcome metrics included receiver operating characteristic (ROC) curve analysis and accuracy. Additional performance measurement techniques, such as precision, recall, specificity, F1-score, and kappa were calculated to identify the model with the best performance. Moreover, several XAI methods, namely permutation feature importance (PFI), partial dependence plots (PDP), Shapley additive explanations (SHAP), and local interpretable model-agnostic explanations (LIME) were implemented for additional exploration of our best model.

The combination of RFE and XGBoost provides the most significant results. The results of the study show that the algorithm has an AUC of 0.95, indicating good discriminatory power in detecting abnormal blood pressure. The accuracy, precision, recall, specificity, F1-score, and kappa scores were 91.50%, 88.64%, 92.65%, 92.27%, 90.83%, and 0.8, respectively. According to the XAI experiment, the genetic pedigree coefficient and hemoglobin level in a patient contribute the most to blood pressure abnormality prediction. Adrenal and thyroid diseases, as well as chronic kidney illness, have an impact on the projections. Existing research backs up this conclusion.

Compared to previous studies on this dataset, our results would be superior, and the use of XAI shed new light on our model’s prediction. This study would provide new insight into blood pressure detection in the medical profession.

Corneal stiffness abnormalities play an important role in the onset and progression of keratoconus. However, the limited availability of specialty devices for measuring corneal stiffness restricts their application in clinical practice. This study aimed to develop a machine learning (ML) model that can predict corneal stiffness based on ocular structures and investigate its efficacy in diagnosing keratoconus.

This retrospective study enrolled healthy individuals and keratoconus patients at the Zhongshan Ophthalmic Center from June 2018 to June 2021. Eleven features, including ocular structural parameters, intraocular pressure (IOP), and age were used to train ML regression models for predicting the stiffness parameter at first applanation (SP-A1) and the Corvis biomechanical index for Chinese populations (cCBI) measured by a Corvis ST device. Mean absolute errors (MAEs) and the area under the receiver operating characteristic curve (AUC) were used to evaluate the performance of the models. The diagnostic efficacy of the predicted SP-A1 and cCBI for keratoconus was evaluated by the AUC, net reclassification index (NRI), and integrated discrimination improvement (IDI).

A total of 1,523 eyes were involved, of which 601 were diagnosed with keratoconus. The MAEs of the SP-A1 prediction were similar in cross-validation (8.95 mmHg/mm) and testing (10.65 mmHg/mm). The R² value for the SP-A1 prediction exceeded 0.7, indicating that the performance was clinically acceptable. The AUC for the cCBI prediction was 0.935 (95% CI 0.906-0.963). The top three predictors for SP-A1 and cCBI were IOP, keratometry, and central corneal thickness. The addition of the predicted SP-A1 and cCBI significantly improved model performance in diagnosing keratoconus, with NRI of 0.607 (95% CI 0.367-0.812) and 0.188 (95% CI −0.022-0.398), and IDI of 0.028 (95% CI 0.006-0.048) and 0.045 (95% CI 0.018-0.072), respectively.

Our models predicted SP-A1 and cCBI relatively accurately in keratoconus and normal corneas. Moreover, the predicted SP-A1 and cCBI values significantly contributed to the diagnosis of keratoconus. These models could provide a potential alternative for evaluating corneal stiffness and thus facilitate keratoconus screening.

Lung diseases pose a significant threat to public health worldwide, resulting in a substantial number of fatalities. Diseases such as chronic obstructive pulmonary disease and lung cancer constitute two of the three deadliest diseases worldwide, contributing to over 3 million deaths annually. This study offered a comparative analysis of different diagnostic techniques used for lung pathologies from an engineering standpoint. The review concentrated on intelligent detection methods, including electronic nose, computer vision (CV), or image processing, and biosensors such as graphene-field effect transistor (FET). The E-nose-based detection technique uses electronic sensors to recognize volatile organic compounds (VOCs) in the exhaled breath. These VOCs can aid in the diagnosis of lung pathologies such as pneumonia. The CV processing method involves the application of advanced imaging techniques and machine learning algorithms to scrutinize and diagnose lung pathologies and ventilator-associated pneumonia (VAP). Lastly, biosensors employ the exceptional properties of these materials to identify specific biomarkers in biological samples. This information can be used to diagnose lung pathologies and VAP. This study examined the current state-of-the-art methods and offers a comprehensive analysis of their advantages and disadvantages from an engineering perspective. The study underscored the potential of these techniques to enhance the diagnosis of lung pathologies and VAP and presents the advances in the field of smart biomedical applications. Additionally, it emphasized the necessity for further research to optimize their performance and clinical usefulness.