Abdominal Surgical Diseases, Head and Neck Surgical Diseases, Musculoskeletal Surgical Diseases, Pelvic Surgical Diseases, Surgical Neoplasms, Thoracic Surgical Diseases, Urologic Surgical Diseases
Conditions
Keywords
Minimally invasive surgery, Surgical video dataset, Surgical artificial intelligence, Endoscopic surgery, Robotic surgery
Brief summary
This is a prospective multicenter patient registry study. We continuously collect full-length intraoperative surgical videos from thoracoscope, laparoscope, hysteroscope, transcervical resectoscope, cystoscope, prostate resectoscope, arthroscope, intervertebral foramen endoscope, otorhinolaryngology endoscope and endoscopic surgical robots, accompanied by inpatient medical records, preoperative imaging data and 5-year postoperative follow-up data. All imaging data will be standardized and de-identified to construct a large-scale standardized surgical video dataset. The dataset will be applied for training, verification and optimization of surgical video foundation large model, serving for surgical teaching, intraoperative operation quality control and basic medical AI research. We will also explore the correlation between intraoperative surgical details and postoperative prognosis to improve the standard specifications of minimally invasive surgery. No clinical intervention will be imposed on participants throughout the whole research.
Interventions
None listed
Sponsors
Chinese Academy of Sciences
Beihang University
Beijing Digital Precision Medicine Technology Co., Ltd.
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to 65 Years
Healthy volunteers
No
Inclusion criteria
1. Patients aged ≥ 18 years old hospitalized to receive minimally invasive endoscopic, laparoscopic or robotic surgical treatment for diseases of various body systems; 2. Complete full-length intraoperative surgical videos can be recorded during operation, with complete medical records and preoperative imaging data; 3. Participants fully understand the study, voluntarily sign written informed consent, and agree that their de-identified intraoperative images and clinical data can be used for scientific research.
Exclusion criteria
1. Minors under 18 years of age; 2. Patients with incomplete intraoperative videos or missing clinical imaging documents; 3. Patients with consciousness disturbance or mental disorders who cannot sign informed consent independently; 4. Subjects who refuse to participate in the study and disapprove the use of their medical data for research; 5. Patients who are predicted to be unavailable for long-term postoperative follow-up.
Design outcomes
Primary
| Measure | Time frame |
|---|---|
| Completion rate of qualified intraoperative surgical imaging data | Immediately after each surgery |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Completeness rate of long-term postoperative clinical follow-up | 3 months, 1 year, 3 years and 5 years after surgery | — |
| Reusability rate of annotated key anatomical structures in videos | From completion of data warehousing and annotation, the reusability rate will be analyzed within 3 months, assessed up to 3 months after annotation completion. | — |
| Feasibility rate (%) of surgical video dataset applied in different clinical AI research scenarios | After full construction of the surgical video dataset, scenario feasibility assessment will be finished within 6 months, assessed up to 6 months after dataset construction. | Three core application scenarios are predefined: 1) training of surgical computer vision AI models; 2) validation of intraoperative surgical recognition algorithms; 3) surgical skill assessment and teaching research. An expert review panel consisting of at least 3 attending surgeons and 2 medical AI researchers independently evaluates whether the dataset has sufficient sample size, annotation completeness and video quality to support each scenario. Feasibility proportion is calculated as: (Number of scenarios the dataset is suitable for / Total predefined scenarios) × 100%. |
| Accuracy percentage (%) of AI-based surgical procedure identification on annotated surgical videos | After completion of data warehousing and annotation, AI surgical procedure identification accuracy testing will be conducted within 3 months, assessed up to 3 months post annotation completion. | After all surgical videos are imported into the data warehouse and manually annotated by experienced surgeons to generate gold-standard procedure labels, the surgical video analysis AI model automatically outputs predicted surgical procedure categories for each video clip. Each AI-predicted label is compared against the manual gold-standard annotation label. Identification accuracy is calculated by the formula: (Number of video clips with correctly predicted surgical procedures / Total number of tested video clips) × 100%. |
Countries
China
Contacts
CONTACTKUNSHAN HE
Outcome results
None listed