An observational study in Kidney (Renal Cell) Cancer, Ureteropelvic Junction Obstruction and Upper Urinary Tract Carcinoma, sponsored by Chinese PLA General Hospital. Not yet recruiting. Open to participants aged 18 Years and older. Per ClinicalTrials.gov, last updated 2026-11-17.
Sponsored by Chinese PLA General Hospital · Observational
Help build ClinicalTrials.gg — short surveys, remote interviews or testing, in your language. Take part →
This study aims to evaluate whether an artificial intelligence (AI) based software system can assist surgeons in identifying key anatomical structures during minimally invasive upper urinary tract surgery.
During these procedures, surgeons operate while viewing images from laparoscopic endoscope. Key anatomical structures, such as the kidney, renal artery and vein, ureter, renal pelvis, inferior vena cava, and left reproductive vein, as well as surgical targets such as tumors, may sometimes be difficult to distinguish because of bleeding, surgical smoke, tissue movement, changes in tissue appearance, or obstruction by surgical instruments. The study team has developed an AI-based image recognition system that analyzes surgical video in real time and identifies and outlines relevant anatomical structures on the screen during surgery.
The study has two main goals (primary endpoints):
The study will include adults aged 18 years or older who undergo one of five types of minimally invasive upper urinary tract surgery: partial nephrectomy, radical nephrectomy, radical nephroureterectomy, inferior vena cava tumor thrombectomy, or pyeloplasty.
The study will be conducted at 11 hospitals in China. In the first phase, surgical videos recorded between 2015 and 2025 will be used to develop, validate, and test the AI system. In the second phase, approximately 100 patients will be prospectively enrolled from internal centers, and the AI-generated anatomical recognition results will be displayed on a monitor during surgery. Overall, approximately 1,600 retrospective surgical cases and 100 prospective cases are planned for inclusion.
The AI system is designed solely as an assistive tool for anatomical recognition. It does not make clinical decisions, provide autonomous surgical instructions, or control any surgical equipment. Surgeons remain fully responsible for all clinical and surgical decisions, and the AI display can be turned off at any time.
All participants will receive standard clinical care. Participation in the study will not involve any additional medications, surgical procedures, or blood tests. Identifying personal information will be removed from the surgical videos and clinical data used for the study, and all study data will be stored securely in accordance with applicable data protection requirements.
As secondary objectives, the study will compare the performance of the AI system with that of surgeons with different levels of surgical experience using a set of challenging surgical images or video frames. The study will also evaluate surgeons' experience with the system, including its usability, clarity of display, and whether the AI-generated information interferes with or affects the surgical workflow.
The time perspective of this study is both retrospective and prospective . In the retrospective phase, surgical videos and routinely collected clinical data from patients who underwent eligible minimally invasive upper urinary tract procedures between January 2015 and May 2025 will be used for dataset construction, annotation, AI model development, internal validation, and multicenter external testing. In the prospective phase, eligible patients will be consecutively enrolled in a real-world surgical setting to evaluate the real-time deployment, technical performance, stability, and clinical usability of the AI-assisted anatomical recognition system.
2,039 studies on the registry are indexed under Carcinoma, Renal Cell; 396 are open to participants now.
This study's planned enrollment of 1,700 is above the median of 151 across 337 observational studies indexed under Carcinoma, Renal Cell.
Browse Carcinoma, Renal Cell studies →Chinese PLA General Hospital is the lead sponsor of 573 studies on the registry; 209 are open to participants now.
Counted across the registry records on this site, refreshed daily.
Adults aged 18 years or older who undergo minimally invasive kidney surgery at one of 11 participating hospitals in China, covering five procedures: partial nephrectomy, radical nephrectomy, radical nephroureterectomy, inferior vena cava tumor thrombectomy with radical nephrectomy, and pyeloplasty. The population includes patients operated between January 2015 and May 2025 whose intraoperative videos and perioperative records are analyzed retrospectively (approximately 1,600 planned cases), and patients enrolled consecutively during the prospective real-time validation phase (approximately 100 cases). Eligible patients have intraoperative video suitable for anatomical identification and perioperative clinical data sufficient to record baseline characteristics and study outcomes. Patients with emergency critical conditions, severely incomplete or low-resolution video, or refusal of data use are excluded.
Go through the criteria one by one — yes, no or not sure — and take the list to your doctor or the study team. Kept in this browser only.
Exclusion Criteria:
An AI-based software system that analyzes the live surgical video feed during minimally invasive upper urinary tract surgery and performs real-time identification and delineation of key anatomical structures and surgical targets, including the kidney, renal artery and vein, ureter, renal pelvis, inferior vena cava, tumors, and adjacent organs. The AI-generated results are overlaid on a surgical monitor to provide intraoperative visual assistance to the surgeon. The system functions solely as an assistive visualization tool; it does not make clinical decisions, provide autonomous surgical guidance, or control any surgical equipment. All surgical decisions remain under the responsibility of the operating surgeon, and the AI display can be turned off at any time.
Dice coefficient
The Dice similarity coefficient will quantify the overlap between the AI-generated segmentation mask and the expert-reviewed reference mask for each target structure in surgical video frames. Dice = 2 x overlap area/(predicted area + reference area), with areas measured as pixel counts. Values range from 0 to 1: 0 indicates no overlap and 1 indicates perfect overlap. Higher values indicate greater agreement with the reference annotation. Results will be reported overall and for each target structure.
Time frame: From enrollment to the end of surgery at 3 days
Intersection over Union, IoU
Intersection over Union will quantify the overlap between the AI-generated segmentation mask and the expert-reviewed reference mask for each target structure. IoU = intersection area/union area, where the intersection contains pixels shared by both masks and the union contains pixels belonging to either mask. Values range from 0 to 1: 0 indicates no overlap and 1 indicates perfect overlap. Higher values indicate greater segmentation agreement with the reference annotation. Results will be reported overall and for each target structure.
Time frame: From enrollment to the end of surgery at 3 days
Precision
Precision will measure the proportion of anatomical structure identifications made by the AI system that are correct according to expert-reviewed reference annotations. Precision = TP/(TP + FP), where TP is the number of correctly identified structures and FP is the number of incorrect or spurious identifications. Values range from 0 to 1. A value of 1 means that all reported positive identifications are correct; lower values indicate a greater proportion of false-positive identifications. Results will be reported overall and for each target structure.
Time frame: From enrollment to the end of surgery at 3 days
Recall
Recall will measure the proportion of target structures present in the expert-reviewed reference annotations that are correctly identified by the AI system. Recall = TP/(TP + FN), where TP is the number of correctly identified target structures and FN is the number of reference-positive target structures that are missed or incorrectly identified. Values range from 0 to 1. A value of 1 means that all reference-positive targets are correctly identified; lower values indicate a greater proportion of missed targets. Results will be reported overall and for each target structure.
Time frame: From enrollment to the end of surgery at 3 days
F1 score
The F1 score will summarize precision and recall for anatomical structure identification against expert-reviewed reference annotations. It is calculated as their harmonic mean: F1 = 2 x precision x recall/(precision + recall), equivalently 2TP/(2TP + FP + FN). TP, FP, and FN denote correct, false-positive, and missed target identifications, respectively. Values range from 0 to 1. Higher values indicate better combined precision and recall; a value of 1 indicates perfect precision and recall. Results will be reported overall and for each target structure.
Time frame: From enrollment to the end of surgery at 3 days
Mean inference latency
Inference latency will be measured as the elapsed time from the model receiving a preprocessed video frame to completion of its anatomical recognition output. Mean inference latency will be calculated by summing the recorded per-frame inference times and dividing by the number of evaluated frames during each prospective procedure. The unit is milliseconds per frame. Lower values indicate faster inference and a shorter processing delay; higher values indicate slower inference.
Time frame: From enrollment to the end of surgery at 1 day
Processing frame rate, FPS
Processing frame rate will measure the number of surgical video frames processed by the AI system per second during prospective intraoperative use. FPS = number of frames with completed AI processing/elapsed wall-clock time of the monitored processing period in seconds. The rate will be calculated for each evaluated procedure and expressed in frames per second. Higher values indicate greater processing throughput; lower values indicate fewer processed frames per second.
Time frame: From enrollment to the end of surgery at 1 day
System crash rate
A system crash will be defined as unexpected software termination or freezing that prevents further AI output and requires a restart. The outcome will be calculated as the number of prospective procedures with at least one crash divided by the total number of procedures in which AI deployment was attempted, multiplied by 100. Each procedure will be counted once regardless of the number of crashes. Planned shutdowns and voluntary switches to the original surgical display without software failure will not count as crashes. Values range from 0% to 100%; lower values indicate a smaller proportion of procedures affected by crashes.
Time frame: From enrollment to the end of surgery at 1 day
an investigator-developed Likert scale questionnaire
After each evaluated prospective surgical procedure, surgeons will complete the investigator-developed Artificial Intelligence System Usability and Surgeon Experience Questionnaire. Usability, trustworthiness, perceived clinical value, and visual interference will each be rated on a five-point Likert-type scale and reported separately. Each domain has a minimum score of 1 and a maximum score of 5. For usability, trustworthiness, and perceived clinical value, 1 indicates very low and 5 indicates very high. Higher scores indicate more favorable ratings. For visual interference, 1 indicates no interference and 5 indicates very severe interference. Higher scores indicate greater visual interference.
Time frame: From enrollment to the end of surgery in 2 hours
NASA-TLX score
Surgeons' perceived workload will be assessed after each evaluated procedure using the original weighted National Aeronautics and Space Administration Task Load Index (NASA-TLX). Six dimensions are rated from 0 to 100: mental demand, physical demand, temporal demand, performance, effort, and frustration. Fifteen pairwise comparisons determine each dimension's weight, ranging from 0 to 5; the weights sum to 15. The overall score is calculated as the sum of each dimension's rating multiplied by its weight, divided by 15. The total score ranges from 0 to 100; higher scores indicate greater perceived workload and lower scores indicate less perceived workload.
Time frame: From enrollment to the end of surgery in 2 hours
Performance in human-AI comparison
The artificial intelligence system and human annotators will identify anatomical structures on the same set of 100 to 300 challenging intraoperative images randomly sampled from the external validation set. Their outputs will be compared with expert-reviewed reference annotations. Accuracy is the percentage of correct determinations. Sensitivity is the percentage of reference-positive determinations correctly identified, and specificity is the percentage of reference-negative determinations correctly identified. Each metric will be reported separately for the artificial intelligence system and for three human groups: experts (more than 100 relevant procedures), intermediate-experience physicians (50 to 100 relevant procedures), and novices (residents, students, or surgical assistants).
Time frame: From enrollment to the end of surgery at 2 weeks
Time to correct anatomical structure identification in surgical videos: human-AI comparison
Time to correct anatomical identification will be compared among the AI system, expert surgeons, intermediate-experience surgeons, and novice participants using the same surgical video clips from the external test set. Time will be measured from each target structure's first identifiable appearance, predefined by independent experts, until its location is correctly indicated. Human participants will indicate the structure by clicking or pointing; AI identification will be recorded when the system first correctly indicates the structure's location. Clips will be presented at normal speed without pausing or replay, and AI will process frames online, including processing latency. Results will be expressed in seconds per target structure for each group. Targets not identified by clip end will be right-censored in time-to-event analyses. Lower values indicate faster identification.
Time frame: From enrollment to the end of surgery at 2 weeks
No study locations are listed for this record.
Start is estimated for Nov 2026; primary completion for Dec 2026, 1 month later.
Of 97 non-industry observational studies under Carcinoma, Renal Cell, half reached primary completion within 50 months of starting (the median), counting those still running; a quarter within 26 and three quarters within 241. This study's planned 1 month is below that.
Counting every sponsor, half of 139 studies reached it within 50 months; a quarter within 33 and three quarters within 94.
How these figures are worked out, and which studies are left out of them.
From the registry record's own update history. This site started tracking changes on Sep 25, 2026; for anything earlier, see the record history on ClinicalTrials.gov ↗
This study is not yet recruiting, as verified in Nov 2026. You cannot join it, but the record below documents what was studied.
Get an email when the registry record changes — status, dates, results — or when someone posts here.
Sign in to followQuestions and observations about this study, from anyone following it. Not medical advice, and not a channel to the study team — their contact details are on the registry record.
Sign in to join the discussion. Reading takes no account; posting does. You choose a display name, and a pseudonym is the default.
Nothing here yet. If you are running this trial, taking part in it, or weighing whether to, this is the place to say so.
Chinese PLA General Hospital →