Detection of malignancy associated changes in cervical cell nuclei using feed-forward neural networks

Area of Science:

• Oncology
• Computational Biology
• Digital Pathology

Background:

• Normal cells near precancerous lesions exhibit subtle DNA distribution changes, termed Malignancy Associated Changes (MACs).
• Distinguishing these altered cells from truly normal cells is crucial for early cancer detection.

Purpose of the Study:

• To compare the effectiveness of neural networks and linear discriminant analysis in classifying Malignancy Associated Changes (MACs).
• To evaluate the performance of automated cell classifiers for detecting precancerous changes.

Main Methods:

• 53 nuclear features were extracted from 25,360 normal-appearing cells across 344 slides.
• Feed-forward neural networks and linear discriminant analysis were used to design cell classifiers.
• Classifiers were validated on a library of 93,494 cells from 395 slides using jackknifing.

Main Results:

• On a cell-by-cell basis, neural networks achieved a 72.5% correct classification rate, outperforming linear discriminant analysis (61.6%).
• When applied to a larger dataset, neural networks reached 76.2% accuracy, compared to 67.6% for the discriminant function.
• These findings demonstrate the superior performance of neural networks in identifying MACs.

Conclusions:

• Neural networks provide a more accurate method for detecting Malignancy Associated Changes (MACs) than traditional statistical models.
• Automated cell classification using AI shows significant potential for improving the accuracy of precancerous lesion detection in pathology.