Machine Learning Allows for Distinguishing Precancerous and Cancerous Human Epithelial Cervical Cells Using...

Area of Science:

• Biophysics
• Cell Biology
• Medical Technology

Background:

• Atomic force microscopy (AFM) adhesion maps previously distinguished normal from cancerous/precancerous cervical cells using fractal dimension, but with limited precision between precancerous and cancerous stages (AUC 0.79).
• Accurate differentiation between premalignant and malignant cervical cells is clinically crucial for effective cancer management.

Purpose of the Study:

• To enhance the classification accuracy of precancerous versus cancerous human epithelial cervical cells.
• To investigate the efficacy of machine learning algorithms in analyzing AFM adhesion map data for improved cervical cancer screening.

Main Methods:

• Utilized high-resolution AFM adhesion maps from six precancerous and six cancerous human cervical cell lines.
• Applied machine learning, specifically the random forest decision tree algorithm, to analyze the adhesion map data.
• Validated classification robustness using K-fold cross-validation (K=500) and statistical significance testing (p < 0.0001).

Main Results:

• Machine learning analysis achieved significantly improved differentiation, yielding an AUC of 0.93, accuracy of 83%, sensitivity of 92%, and specificity of 78%.
• The enhanced sensitivity (92%) addresses a critical clinical need for improved cervical cancer screening.
• Results demonstrate statistically significant improvements over previous fractal dimension-only analyses.

Conclusions:

• Machine learning integration with AFM adhesion mapping provides a powerful tool for precise discrimination between precancerous and cancerous cervical cells.
• This approach holds significant potential for improving the sensitivity of cervical cancer screening, potentially complementing existing methods.
• Further clinical validation could lead to enhanced diagnostic capabilities for cervical neoplasia.