Swin-GA-RF: genetic algorithm-based Swin Transformer and random forest for enhancing cervical cancer classification

Area of Science:

• Medical Imaging
• Computational Biology
• Artificial Intelligence

Background:

• Cervical cancer presents a growing global health concern, necessitating advancements in early detection methods.
• Computer vision, especially Swin transformers, shows potential in medical image analysis, outperforming traditional methods.
• Accurate classification of cervical cells in Pap smear images is vital for timely intervention.

Purpose of the Study:

• To develop and evaluate a novel approach, Swin-GA-RF, for improved cervical cancer cell classification.
• To leverage the Swin transformer's capabilities combined with genetic algorithm feature selection and random forest classification.
• To enhance the performance of cervical cancer screening using the SIPaKMeD1 dataset.

Main Methods:

• Utilized the Swin transformer for feature extraction from cervical cell images.
• Employed a genetic algorithm (GA) for optimal feature selection.
• Integrated a random forest classifier, replacing the standard softmax layer, and applied data augmentation.

Main Results:

• Swin-GA-RF demonstrated superior performance compared to other Swin transformers and pre-trained CNN models.
• Achieved high accuracy (99.012% binary, 98.808% five-class) and F1-scores in both classification tasks.
• The Adam optimizer yielded the best results for the Swin-GA-RF model.

Conclusions:

• The proposed Swin-GA-RF approach is highly effective for cervical cancer classification from Pap smear images.
• This method shows significant potential for improving the accuracy and efficiency of early cervical cancer diagnosis.
• Swin-GA-RF offers a valuable advancement for cervical cancer screening programs and women's health initiatives.