A new investigational diagnostic imaging technique based on analyzing nanoscale resolution maps of cell surfaces performed favorably compared to what might be expected with cystoscopy in detecting bladder cancer.
A new investigational diagnostic imaging technique based on analyzing nanoscale resolution maps of cell surfaces performed favorably compared to what might be expected with cystoscopy in detecting bladder cancer, researchers reported in the Proceedings of the National Academy of Sciences (Dec. 3, 2018 [Epub ahead of print]).
When evaluating five cells per patient urine sample, the modality had 94% diagnostic accuracy in the identification of individuals who did or did not have cancer, according to study author Igor Sokolov, PhD, of Tufts University, Boston.
The technique, which involves machine-learning analysis of images obtained using atomic force microscopy (AFM), offered a statistically significant improvement in diagnostic accuracy versus cystoscopy, according to the investigators, who described a study that involved analysis of cells from bladder cancer patients and control subjects.
While further research is needed, this noninvasive imaging technique might one day provide an alternative diagnostic approach that avoids the discomfort and expense associated with cystoscopy, and may have utility in cancers of the upper urinary tract or urethra, Dr. Sokolov said in an interview.
“The methodology is applicable to pretty much any cancer where we can get cells without tissue biopsy. It’s definitely much more preferable to biopsy for the patient-if you spit, spill, or cough out some cells, we can analyze them,” he told Urology Times.
In the present study, Dr. Sokolov and colleagues evaluated urine samples from 25 patients with pathologically confirmed bladder cancer and 43 individuals with no present evidence of bladder cancer. The cells were collected similar to what’s done in voided urine cytology, according to investigators.
The cells were evaluated with AFM, in which the movements of a tiny cantilever are recorded as it passes over the surface-in this case, the surface of a cell. Those measurements are displayed as images that reflect certain physical properties, such as height, and force of adhesion between the sample and the tip of the cantilever that is in contact with the sample.
The AFM images were analyzed using computer-based machine learning methods designed to recognize different objects and patterns, according to Dr. Sokolov.
Next: Accuracy of the modality was 94%The accuracy of the modality was 94%, with sensitivity/specificity pairs of 81/98% and 91/82% in example analyses, Dr. Sokolov and colleagues reported in the journal.
The AFM-based noninvasive analysis was significantly better than what has been reported for cystoscopy, according to results of a statistical analysis. Using area under the receiver operating characteristic curve (AUC) for classification, AFM and cystoscopy had AUCs of 0.92 and 0.77, respectively (p<.05), the analysis shows.
Results of the AFM-based modality also compare favorably with the accuracy of currently used noninvasive modalities for detecting bladder cancer based on urine samples, such as immunocytochemistry, genetic analysis using fluorescence in situ hybridization, and NMP22 biomarker evaluation, which have reported sensitivities of 20% to 80% in this setting, according to Dr. Sokolov.
The research was supported in part by the National Science Foundation and Norris Cotton Cancer Center, Dartmouth College. Tufts University has applied for a patent related to the atomic force microscopy machine learning technology, invented by Dr. Sokolov and Milos Miljkovic, PhD, a co-author of the paper.