Randomized trial evaluates deep learning models for cervical cytology classification, suggesting AI's role in cancer detection.
Key Points
The study aims to evaluate multiple deep learning architectures for the automated classification of cervical cytology images into diagnostic categories.
Comprehensive evaluation of eight CNN architectures including AlexNet, VGG-16, and EfficientNet-B0.
Incorporation of Vision Transformer (ViT-16) for enhanced feature representation.
All models were trained under identical conditions for fair comparisons.
ViT-16 achieved the highest test accuracy of 95.88% and specificity of 0.9864.
EfficientNet-B0 and DenseNet-121 showed 94.33% and 93.30% accuracy, respectively.
ViT-16 excelled in classifying challenging categories like SCC and HSIL.