Deep-learning ultrasound system for precise diagnosis of bladder tumors

Construction of a Deep Learning-Based Precise Diagnostic Framework for Bladder Tumors Using Ultrasound

This project will try a deep-learning system that reads ultrasound images to automatically segment bladder tumors, estimate T-stage, and predict tumor grade for adults with suspected bladder masses.

Quick facts

• Study type: Observational
• Enrollment: 400 (estimated)
• Ages: 18 Years to 85 Years
• Sex: All
• Sponsor: Peking University First Hospital (Academic / other)
• Drugs / interventions: chemotherapy
• Locations: 1 site (Beijing)
• Trial ID: NCT07111364 on ClinicalTrials.gov

What this trial studies

Researchers will build and train a deep-learning model using ultrasound images collected from patients with suspected bladder masses, combining retrospective and prospective cohorts. The system is designed to perform automated tumor segmentation, predict T-stage, and estimate pathological grade, with surgical pathology serving as the reference standard. The aim is to improve the objectivity, accuracy, and efficiency of ultrasound-based diagnosis while reducing dependence on operator experience. Patients with adequate preoperative abdominal or transrectal ultrasound images who undergo surgery will be included, while those with recent treatments, indwelling devices, or poor image quality will be excluded.

Who should consider this trial

Good fit: Adults (≥18 years) with a suspected bladder mass on ultrasound who are scheduled for surgical treatment and can undergo adequate abdominal or transrectal ultrasound are ideal candidates.

Not a fit: Patients older than 85, unable to obtain technically adequate ultrasound images, with recent bladder therapies, indwelling urinary devices, or without pathologically confirmed urothelial carcinoma are unlikely to benefit.

Why it matters

Potential benefit: If successful, the tool could give faster, more objective ultrasound-based tumor assessments that help guide treatment decisions and optimize use of clinical resources.

How similar studies have performed: Deep-learning models applied to ultrasound and other imaging modalities have shown promising results for tumor detection and grading, but ultrasound-based DL specifically for bladder T-staging and grade prediction remains relatively novel and less validated.

Eligibility criteria

Inclusion Criteria:① Suspected bladder mass detected by abdominal ultrasound (age ≥18 years);② Patients scheduled for surgical treatment of bladder tumors.

Exclusion Criteria:

* Age \>85 years;

  * Patients unable to undergo abdominal/transrectal ultrasound (e.g., uncooperative individuals, technically inadequate images);

    * History of bladder tumor surgery, radiotherapy, chemotherapy, or systemic therapy within 3 months; ④ Patients with indwelling medical devices (e.g., double-J ureteral stents, urinary catheters);

      * Failure to undergo bladder tumor surgery within 2 weeks post-ultrasound; ⑥ Non-urothelial carcinoma or pathologically unconfirmed diagnoses.

Where this trial is running

Beijing

• Department of Urology, Peking University First Hospital — Beijing, China (Recruiting)

Study contacts

• Study coordinator: Zheng Zhang

Conditions Deep Learning; Ultrasound; Bladder Cancer