Detection of Human Bladder Epithelial Cancerous Cells with Atomic Force Microscopy and Machine Learning.

Area of Science:

• Biomedical Engineering
• Oncology
• Nanotechnology

Background:

• Noninvasive bladder cancer detection is crucial.
• Atomic force microscopy (AFM) with machine learning shows promise for identifying bladder cancer cells from urine.
• Previous studies had limitations due to small patient cohorts and modest statistical significance.

Purpose of the Study:

• To corroborate the capability of AFM to identify bladder cancer cells.
• To enhance the accuracy and statistical significance of AFM-based bladder cancer detection.
• To establish a rigorous technical foundation for clinical development.

Main Methods:

• Utilized a controlled model system of genetically purified human bladder epithelial cell lines.
• Applied AFM to analyze cell adhesion maps and compare cancerous with nonmalignant cells.
• Processed AFM data using machine learning algorithms, including multi-channel imaging in Ringing mode.

Main Results:

• Achieved an area under the ROC curve (AUC) of 0.97 with 91% accuracy using standard AFM adhesion maps.
• Enhanced detection with multi-channel AFM imaging, reaching an AUC of 0.99 and 93% accuracy.
• Demonstrated statistically significant results (p < 0.0001) in the controlled model system.

Conclusions:

• AFM combined with machine learning is a highly accurate method for bladder cancer cell identification.
• Multi-channel AFM imaging further improves detection accuracy.
• This study provides strong technical validation for AFM-based bladder cancer detection, supporting future clinical applications.