Machine Learning

Researchers are evaluating a new deep learning model that uses ultrasound video data to estimate blood volume before surgery. This prospective, single-center observational study focuses on adult patients scheduled for surgery, aiming to provide an accurate, non-invasive way to assess blood volume, which is important to prevent complications during surgery. The study is led by Shanghai 6th Peoples Hospital and addresses the current lack of practical methods for direct blood volume measurement. Participants will undergo preoperative ultrasound scans capturing videos of four major blood vessels the Internal Jugular Vein, Subclavian Vein, Inferior Vena Cava, and Common Carotid Artery. The true blood volume is calculated using a clinical method involving hemoglobin concentration changes before and after acute normovolemic hemodilution. The collected ultrasound videos will be used to train and validate a deep learning model that combines convolutional and temporal analysis techniques to estimate blood volume from the ultrasound data. During the study, participants will have ultrasound video clips taken shortly before surgery. Researchers will analyze these images alongside blood volume values determined by the clinical method. The main outcome measured is the accuracy of the blood volume estimate within 30 minutes before surgery. The study does not involve treatment changes but focuses on data collection and model validation. Participation lasts through the preoperative period up to surgery, with no additional follow-up specified.

Researchers are investigating the use of transcranial direct current stimulation tDCS combined with cognitive training to address symptoms in adolescents with autism spectrum disorder ASD. ASD is characterized by challenges in communication, social skills, repetitive behaviors, and emotional distress triggered by environmental changes. This study aims to determine if baseline resting-state EEG and clinical data can predict which individuals respond to tDCS treatment, helping to improve intervention strategies. Participants will receive active tDCS over 10 sessions during two weeks, with one session per day on consecutive working days. Each session includes 20 minutes of tDCS combined with an online cognitive training program consisting of five exercises targeting information processing speed and executive function. After treatment, participants will be classified as responders or non-responders based on changes in social responsiveness scores. During the study, assessments will measure changes in social communication and repetitive behaviors using the Social Responsiveness Scale, cognitive function through various CANTAB tests, and neurophysiological measures at the start and immediately after the intervention. Researchers will monitor behavioral changes and executive function outcomes to evaluate the impact of the combined treatment over the study period ending in 2026.

Pregnant women with aplastic anemia AA, a rare blood disorder causing bone marrow failure and low blood counts, face serious risks including heavy bleeding, infections, and complications for their babies like preterm birth and restricted growth. This research aims to develop and validate a tool to predict adverse pregnancy outcomes in women with AA, helping guide early clinical decisions and improve health for mothers and infants. The study combines retrospective and prospective data collection across multiple centers to address the current lack of comprehensive data in this area. The study involves observing pregnant women diagnosed with AA either before or during their pregnancy. Researchers will gather baseline information and diagnostic data, then follow participants regularly through questionnaires, phone calls, video consultations, online platforms, and in-person visits. They will record treatments, other health conditions, and pregnancy outcomes, aiming to build a prediction model for adverse outcomes. Participants will be monitored from their first hospital visit during pregnancy until 42 days after delivery. Data collected includes clinical assessments and neonatal Apgar scores shortly after birth. The study will provide detailed information on the health and risks for mothers and babies affected by AA during pregnancy, helping improve care strategies. The total participation duration varies depending on the timing of enrollment and delivery.

Researchers are conducting a real-world, observational study to evaluate the efficacy and safety of immune checkpoint inhibitors ICIs as the first-line treatment for advanced malignant tumors. The study includes patients with stage IV solid tumors such as non-small-cell lung carcinoma, stomach, breast, and urinary system cancers. The goal is to build a database combining immunohistochemistry, gene detection, imaging data, and survival outcomes to develop predictive models for immunotherapy biomarkers and treatment responses. The study collects various data points including immunohistochemistry results, gene testing, imaging, overall survival, progression-free survival, and objective response rate. It also monitors immune-related adverse events over time. This observational study does not involve experimental treatments but instead observes patients receiving ICIs as part of their standard care. The study spans at least two years for some outcome measures to assess long-term effects and safety. Participants will undergo regular assessments including survival evaluations and monitoring for treatment-related side effects. Data on progression-free survival will be collected at six months, with overall survival and immune-related adverse events followed for up to two years. The study aims to validate the predictive value of biomarker systems for immunotherapy and improve understanding of treatment effects in advanced cancer patients. Total participation duration will vary depending on individual treatment and follow-up schedules.

Researchers are developing an artificial intelligence tool to predict adverse outcomes in patients with acute pulmonary embolism using data from CT pulmonary angiography. This observational study collects clinical, laboratory, and CT scan information from patients admitted with acute pulmonary embolism to better understand and forecast complications within 30 days of hospital admission. The goal is to improve early risk assessment by comparing the new AI-based prediction model with established risk systems. The study divides participants into two groups a derivation cohort used to create and test a logistic regression model, and a validation cohort to confirm the models consistency. No interventions or treatments are given as part of the study. Data collected include CT parameters, echocardiography results, and blood markers such as cardiac troponin I and NT-pro BNP. The models accuracy is assessed by comparing predicted outcomes with actual adverse events. Participants will have their clinical and imaging data analyzed from hospital admission, with follow-up to track treatment-emergent adverse events within 30 days and up to 2 years. Researchers monitor the incidence of these adverse events to evaluate the AI tools predictive value. The entire participation involves data collection without experimental treatment, and the study spans several years to gather long-term safety information.

Atopic dermatitis AD is a long-lasting skin condition marked by repeated rashes and itching, which seriously impacts patients quality of life and creates a significant economic burden. Researchers are evaluating an Artificial Intelligence assistant decision-making system AIADMS app that integrates the Treat to Target T2T strategy to improve disease control and clinical outcomes for patients with moderate to severe AD. This project aims to develop, train, and test the AIADMS app through clinical trials to verify its effectiveness and safety in managing AD. The AIADMS app is designed for both patients and clinicians with features such as reminders, photo uploads, self-evaluation, and data management. The app uses a deep learning model trained on over 10,000 clinical images to assess AD severity and clinical signs. Participants in the treatment group will use the app for disease management and be followed up at scheduled times up to 12 months after treatment, while the control group will receive routine face-to-face care. The app supports Android and iOS and provides automated assessments using tools like EASI, POEM, SCORAD, PP-NRS, and DLQI to help patients and clinicians monitor progress. Participants will have follow-up visits at 2 weeks, 4 weeks, 8 weeks, 12 weeks, 6 months, and 12 months after treatment. During these visits, evaluations using five treatment objectives will be performed. Data collected will include clinical scores, economic consumption, and satisfaction evaluations. The primary outcome is the overall efficiency rate of treatment objectives at 12 weeks. The study also involves tracking safety and patient adherence, with the entire study period lasting up to 12 months.

This research aims to investigate whether artificial intelligence AI can detect imaging features typical of Intensive Care Unit-acquired Weakness ICUAW using neuromuscular ultrasound. The study focuses on evaluating if AI-based image analysis can identify and monitor ICU patients with ICUAW and whether these AI results correlate with muscle weakness severity, visual muscle echogenicity grading, and 30- and 90-day patient outcomes. ICUAW is a common neuromuscular complication in critically ill patients, often difficult to assess due to patient sedation and limited cooperation during clinical exams. Participants will undergo non-invasive neuromuscular ultrasound of peripheral muscles in the upper and lower limbs. The ultrasound images will be processed using AI, specifically Convolutional Neural Networks, to classify muscle weakness severity. Explainable AI techniques will also be used to highlight the areas within the ultrasound images that contribute to the AIs decisions, helping to understand muscle changes. The study includes groups of critically ill patients with and without ICUAW as well as healthy controls. During the study, researchers will assess muscle echogenicity abnormalities by ultrasound on Day 14 and measure ICUAW severity through various scales. Additional outcomes such as ventilation duration, hospital stay length, survival, frailty, and overall recovery will be evaluated at 30 and 90 days. Data collection involves clinical examinations, scoring systems, and AI image analysis to improve diagnosis and monitoring of muscle weakness in ICU patients.

Researchers are evaluating a facial image-based artificial intelligence AI algorithm designed to screen for coronary artery disease CAD in high-risk community populations, including those with diabetes, hypertension, or aged over 65. The study aims to verify if the AI can accurately identify high-risk and low-risk groups by comparing the actual CAD prevalence, and to compare the AI screening detection rate with natural detection rates in a real-world cohort. Participants are divided into two groups. One group undergoes AI-based facial image screening followed by Coronary Computed Tomography Angiography CCTA to confirm CAD diagnosis. The other group represents a real-world setting without the AI screening and is observed to track natural CAD detection and major adverse cardiovascular events MACE over a six-month follow-up period. During the study, all participants receive monitoring for six months to assess CAD presence and cardiovascular outcomes, including heart attacks and revascularization procedures. Researchers will measure differences in CAD prevalence between AI-identified risk groups, detection rates of CAD, and the incidence of cardiovascular events and deaths. This follow-up period helps evaluate the AI screenings performance compared to usual detection methods.

This research aims to develop and evaluate an AI-assisted nursing care robot called the E-Nursing Assistant to help address workforce shortages and heavy workloads faced by nurses. The study focuses on improving nursing work efficiency and quality by reducing nurses workload using this technology. The trial is led by National Taiwan University Hospital and involves both nursing staff and patients in a clinical ward setting. The study tests the programming framework of the E-Nursing Assistant, which performs functions like guiding patients to specific ward locations, providing equipment usage instructions, playing educational care videos, reminding patients about examination precautions, offering health education on specimen collection, and answering medical questions through an expert-developed Q&A system. This intervention is implemented in the ward for nursing care tasks that do not involve patient safety. Participants will be observed regarding nursing staff workload and stress levels before and after the intervention, as well as the time and frequency spent on nursing tasks. Feedback is gathered through interviews. Outcomes include measuring workload-related stress in nursing staff, system usability of the robot, and the number of patients served. The study includes adults aged 18 and older and involves both nurses actively working in wards and patients or caregivers willing to accept the intervention.

Breast phyllodes tumor PT is a rare type of breast tumor classified into benign, borderline, and malignant categories based on specific tissue features. Malignant PTs have a high chance of coming back locally and spreading to other parts of the body. Early and accurate diagnosis along with identifying treatment targets is important to improve patient outcomes. This research focuses on using artificial intelligence AI to combine clinical, imaging, and genetic data to help diagnose and predict the prognosis of breast PT. The study collects high-quality data from nearly a thousand patients with breast PT, including various medical images such as ultrasound, mammography, CT, and MRI, along with tissue gene sequencing. Researchers aim to build a detailed multi-omics database and develop an AI-based system that can support early diagnosis and predict how the tumor may progress. This system is designed to assist personalized treatment decisions and address care differences across regions. Participants diagnosed with breast PT will contribute their imaging and tissue data. Researchers will assess outcomes like diagnostic sensitivity, false-negative and false-positive rates, and accuracy over five years using statistical measures such as the receiver operating characteristic curve. The study involves no treatment interventions but focuses on observation and data analysis to improve diagnostic tools. The research is sponsored by Sun Yat-Sen Memorial Hospital of Sun Yat-Sen University and started in March 2023, with an expected completion by the end of 2027.