Noninvasive bladder cancer diagnostics via machine learning analysis of nanoscale surface images of epithelial cells extracted from voided urine samples - PROJECT SUMMARY/ABSTRACT Bladder cancer is common cancer with an estimated 81,190 new cases and 17,240 deaths in 2018 (with > 500,000 survivors) only in the US. The gold standard for diagnosis of bladder cancer includes an invasive optical bladder examination (cystoscopy) and tumor resection for pathology examination. Because of a high recurrence rate of this cancer (50-80%), frequent (once every 3-6-12 months) costly and invasive cystoscopy exams are required to monitor patients for recurrence and/or progression to a more advanced stage. It makes bladder cancer the most expensive cancer to monitor/follow up and treat per patient. Moreover, the invasive nature of the current standard of care, cystoscopy, causes rather low compliance of patient to follow this procedure. There is an urgent unmet need for a bladder cancer screening and monitoring test, which will be noninvasive, rapid, objective, reproducible, easy to perform and interpret, and highly accurate. Such a test will reduce the need in frequent cystoscopies and greatly expand the participation of patients in screening and early detection programs because it decreases the patient discomfort and post-procedural complications. Here we propose to develop such a test for identification of the presence of bladder cancer and its aggressiveness (grade). It will be based on non-invasive analysis of individual cells extracted from urine (extraction technology already exists in hospitals for voided urine cytology tests, (VUC) the current standard-of- care, a non-invasive examination of cells in urine used to assist with cancer diagnosis and surveillance). A novel modality of Atomic Force Microscopy (AFM) will be used for nanoscale imaging of cells extracted from urine, mapping/imaging of the physical properties of the cell surface. The collected images will further be analyzed using machine-learning methods and novel advanced statistical approaches to identify a “digital signature” of cancer. The proposed technology is fundamentally different from previously studied urine biomarkers and all existing physical methods because it is based on the analysis of physical properties of the cell surface, not cell bulk or presence of biochemical markers or genetic analysis. Our strong preliminary results demonstrate the feasibility of the proposed approach, its presumed superiority compared to the currently used non-invasive methods, and lead us to the central hypothesis that bladder cancer can be identified by analyzing a small number of cells randomly chosen from urine samples, with a low sampling error. This is a substantial departure from VUC tests, which require a visual analysis of many cells. Supported by the preliminary data, we propose (1) to optimize and expand the method, (2) to define the accuracy of cancer detection on a large cohort of patients, and (3) to assess the accuracy of identification of aggressiveness (low versus high grade) of bladder cancer. Our long-term goal is to develop a non-invasive clinical method for accurate detecting of presence and monitoring bladder cancer as well as many other cancers, in which cells can be extracted from easily accessible bodily fluids without the need for tissue biopsy (e.g urine-bladder & upper urinary tract cancer, stool- colorectal cancer, sputum-aerodigestive cancer, cervical smears-cervical cancer etc.), using methods based on the analysis of physical characteristics of the cell surface. The proposed research, which is the first step in pursuit of this overarching goal.
