Bladder cancer is one of the leading causes of death in the U.S., and while early diagnosis is critical to increasing survival rates, current diagnostic and monitoring methods including cystoscopy, biopsy, and tumor resection, are invasive and costly. A Tufts University-led team of scientists has now developed a noninvasive approach that uses atomic force microscopy (AFM) and machine learning to identify cancerous cells in patients’ urine, by mapping cell surface features and topography at the nanometer scale. The researchers say their technology represents the first time that AFM has been used to help diagnose disease, and suggest that as well as being more accurate compared with cystoscopy, the method could feasibly be applied to detect other forms of cancer or noncancerous abnormalities in cells collected from body fluids, as well as monitor cells’ reactions to drugs.
“By introducing a noninvasive diagnostic method that is more accurate than the invasive visual examination, we could significantly decrease the cost and inconvenience to patients,” said Igor Sokolov, Ph.D., professor of mechanical engineering and biomedical engineering at Tufts University School of Engineering. “All that is needed is a urine sample, and not only could we more effectively monitor patients after treatment, we could also more easily screen healthy individuals who may have a family history of the disease, and potentially detect the grade of cancer development. Determining the efficiency of early screening and grade detection is a separate, important task of our future research. ”
Dr. Sokolov is the lead author of the researchers’ published paper in the Proceedings of the National Academy of Sciences (PNAS), which is titled, “Noninvasive diagnostic imaging using machine-learning analysis of nanoresolution images of cell surfaces: Detection of bladder cancer.”
Bladder cancer is one of the world’s most common forms of cancer, with an estimated 81,190 cases and 17,240 deaths in the U.S. in 2018. The 5-year survival rate is as high as 95% when the disease is diagnosed early, but drops to 10% for patients with metastatic disease. The current gold-standard method for diagnosis includes cystoscopy (passing a camera into the bladder through the urethra), along with biopsy so that cells can be examined, and subsequent tumor resection. The recurrence rate for treated patients is 50–80%, so invasive, expensive cystoscopy exams still need to be carried out every 3–6 months to monitor patients.
“The requirement for frequent cystoscopy makes bladder cancer the most expensive cancer per patient to diagnose, monitor, and treat,” the authors stated. Cystoscopy also displays limited accuracy for some grades and locations of bladder tumors, they suggested. What’s needed is a patient-friendly approach that is less costly and more accurate. “A low-cost, accurate, effective, and noninvasive test will greatly expand participation of patients in screening and early detection evaluation programs because it will decrease the patient discomfort and potential postprocedural complications, while it can assist in improving diagnosis, monitoring, and surveillance, acting as adjunctive to cystoscopy and/or eliminating unnecessary cystoscopies.”
The Tufts University-led scientists have now developed a diagnostic approach that uses AFM and machine learning methods to image and analyze cells extracted from patients’ urine. AFM involves scanning over a surface with a cantilever, which is deflected as it passes over surface contours. Every deflection is recorded, to create a topographical map with a resolution of fractions of a nanometer. How the cantilever is deflected can also indicate physical properties of the sample, such as the adhesion force between the AFM probe and the sample surface.
The scientists applied the technique to analyze urine samples from 43 control individuals without evidence of bladder cancer, and 25 patients with pathologically confirmed bladder cancer. The analyzed sets of surface parameters, including roughness, directionality, and fractal properties, derived from each image. Using machine learning enabled a more accurate recognition of these parameters.
The results confirmed that cells extracted from the urine of bladder cancer patients demonstrated different features from those of control patient cells, which could be identified using the AFM and machine learning approach. Encouragingly, when used to examine just five cells’ per urine sample, the new technique demonstrated greater than 90% sensitivity, compared with 20–80% sensitivity for currently available noninvasive diagnostic approaches that use urine samples. Existing noninvasive tests include biochemical evaluation of the biomarker NMP22, genetic analysis using fluorescence in situ hybridization, or immunocytochemistry. The specificity of the AFM-based approach was also comparable to that of the other tests, at 82–98%.
The authors say that while their method will need evaluating on samples from much larger numbers of patients, it could feasibly be integrated seamlessly into current clinical practice. Another benefit is that the AFM technique requires only a small number of randomly chosen cells. “As we have shown, the accuracy of our AFM method is higher than that of currently used clinical standard, cystoscopy, and the currently used noninvasive methods such as VUC [voided urine cytology] and biochemical evaluation,” they wrote. “When introduced, it will help facilitate screening, reduce overdiagnosis (currently a substantial problem), and consequently, the number of unnecessary and costly medical procedures … The described approach can be extended to detect other cancers and other nonmalignant cell abnormalities as well as to the detection of cell reaction to various drugs (nanopharmacology). Therefore, we expect the described method may be a new direction of biomedical imaging.”
“AFM has been around for more than 30 years, but this is the first time it has shown promise for clinical diagnostics,” noted Dr. Sokolov. “The accuracy appears to be better than the current clinical standard for bladder cancer diagnosis, but we will need to test the method on a larger cohort of patients before it can be introduced into clinical practice. We are hopeful that AFM could ultimately be applied to the detection of other tumor types, such as gastrointestinal, colorectal, and cervical cancers.”
