Skip to content

Multiple-instance learning for classification of urinary cytology slides: retrospective model development in a pilot cohort and prospective validation

Multiple-instance learning for classification of urinary cytology slides: retrospective model development in a pilot cohort and prospective validation - MIL-UC Pilot

Status
Active, not recruiting
Phases
Unknown
Study type
Observational
Source
DRKS
Registry ID
DRKS00039151
Enrollment
100
Registered
2026-01-29
Start date
2025-11-14
Completion date
Unknown
Last updated
2026-08-03

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

C67 C65 C66

Interventions

Group 1: AI training and evaluation using archived urinary cytology slides from patients with histologically confirmed urothelial carcinoma of the urinary bladder or upper urinary tract, in whom urina

Sponsors

Klinik für Urologie und Urochirurgie der Universitätsmedizin Mannheim
Lead Sponsor

Eligibility

Sex/Gender
All
Age
18 Years to No maximum

Inclusion criteria

Inclusion criteria: -Archived urinary cytology samples from routine clinical diagnostics with an available cytopathological report. -Availability of a urological procedure performed in temporal proximity to the urinary cytology to allow determination of a reliable ground truth (GT). -Positive GT (urothelial carcinoma): --Histological confirmation of urothelial carcinoma of the urinary bladder (after transurethral resection of the bladder [TUR-B] or cystectomy), or --Histological confirmation of urothelial carcinoma of the upper urinary tract (after ureterorenoscopy [URS] with biopsy or after nephroureterectomy). -Negative GT (no urothelial carcinoma): --Negative histological findings, or --Documented endoscopic exclusion of urothelial carcinoma (e.g. cystoscopy, URS, holmium laser enucleation of the prostate [HoLEP], transurethral resection of the prostate [TUR-P], DJ stent insertion or exchange, urethrotomy) with systematic inspection of the urinary bladder and/or the upper urinary tract. -Time interval between urinary cytology and the histological or endoscopic reference examination = 1 month.

Exclusion criteria

Exclusion criteria: -Evidence of significant bacteriuria at the time of urinary cytology (bacterial count > 10³/mL in preoperative urine culture). -Patients who received intravesical BCG therapy within the past 3 months -Urinary cytology samples obtained from neobladders. -Washed urine cytology samples with a sampling site not corresponding to the tumor location (e.g. bladder wash cytology in upper tract urothelial carcinoma or renal pelvis wash cytology in bladder carcinoma). -Absence of a histological reference and no endoscopic procedure available to confirm a negative ground truth. -Patients without current evidence of urothelial carcinoma who had been diagnosed with urothelial carcinoma within the preceding two years, in order to minimize the risk of an incorrect ground truth (exception: at least two documented negative cystoscopies within this period). -Patients without evidence of tumor in the surgical specimen following oncological cystectomy, in order to minimize the risk of an incorrect ground truth. -Histological or endoscopic reference examinations performed more than one month apart from the urinary cytology.

Design outcomes

Primary

MeasureTime frame
What: Diagnostic performance of the AI model for the detection of urothelial carcinoma based on urinary cytology slides. When: Assessed during model training and validation, including model tuning and evaluation within a five-fold cross-validation framework on the retrospective dataset and external prospective validation on an independent dataset collected between April 2025 and January 2026. How: Evaluation of AI-based slide-level predictions against the reference standard (ground truth) defined by histological or endoscopic findings, with performance quantified by the area under the receiver operating characteristic curve (AUC-ROC).

Secondary

MeasureTime frame
1. Sensitivity and specificity of the AI model for the detection of urothelial carcinoma, assessed separately in the internal cross validation and in the external prospective validation. 2. Comparison of the diagnostic performance of the AI model with archived cytopathological findings from routine clinical practice in the retrospective pilot cohort. 3. Assessment of the interpretability of AI decisions using heatmaps (top 50 image patches with the highest attention scores), evaluated by an experienced cytopathologist in a blinded manner and stratified by AI outcome groups (true positive, false positive, false negative).

Countries

Germany

Contacts

Public ContactKarl-Friedrich Kowalewski

Klinik für Urologie und Urochirurgie der Universitätsmedizin Mannheim

karl-friedrich.kowalewski@umm.de06213832201

Outcome results

None listed

Source: DRKS (via WHO ICTRP) · Data processed: Aug 10, 2026