Back to 2026 Abstracts
Predictive Capabilities of a Machine Learning-Driven, Nanoscale Imaging Platform for Non-Muscle Invasive Bladder Cancer Diagnosis in a Prospective, Cross-Sectional, Case-Controlled Study
Amir Monemian, PhD1, Jean Pham, MS1, Son Pham, MS1, Zhiying Xie, PhD1, Nadia Makarova, PhD1, Louis Liou, MD, PhD2, Emily Huang, MD3, Nnamdi Onochie, BS3, Matthew Mossanen, MD, MPH3.
1Cellens, Inc., Boston, MA, USA, 2Emerson Hospital, Concord, MA, USA, 3Mass General Brigham, Boston, MA, USA.
BACKGROUND: Surveillance for non-muscle invasive bladder cancer (NMIBC) requires frequent surveillance cystoscopies resulting in high costs, discomfort, and procedural risk. A reliable non-invasive rule-out test could reduce negative procedures and triage those who truly need cystoscopy. We developed a novel nanoscale urothelial cell imaging method that uses atomic force microscopy with machine learning to detect recurrence of NMIBC.
METHODS: In a prospective, cross-sectional, case-controlled study, we enrolled 95 surveillance patients with prior NMIBC; 38 had recurrence confirmed by histology, 54 had no recurrence confirmed by negative cystoscopy, and 3 lacked a final diagnosis. Urothelial cell surfaces were scanned with a nanoscale cantilever to quantify biophysical properties (e.g., adhesion). Surface features underwent a standardized data cleaning pipeline. A finalized gradient-boosting model produced a binary patient-level output (“cancer signature detected” vs “not detected”) with thresholds prespecified for a highly sensitive use case. Performance was compared against cystoscopy, cytology, and cytology.
RESULTS: Of 95 enrolled, 76 patients (80%) were evaluable (27 recurrence, 49 non-recurrence); 19 were non-evaluable due to inadequate urothelial cellularity (n=8), insufficient medical records (n=5), or covered with other particles (n=6). Preliminary patient-level performance on the 76-patient analysis set showed an
AUC of 88.1 (95% CI: 79.1-97.0). All 27 positive recurrences on both low grade and high-grade patients were detected with our method.
CONCLUSIONS: Preliminary data for this nanoscale urothelial cell imaging method is promising, demonstrating an ability to detect recurrences where urine cytology fails, especially for low-grade disease. This may help urologists better identify patients with recurrent bladder cancer for cystoscopic evaluation while improving access to care and patients' quality of lives.
Back to 2026 Abstracts
