CNN based method for classifying cervical cancer cells in pap smear images

Area of Science:

• Medical Imaging
• Computational Biology
• Oncology

Background:

• Cervical cancer diagnosis relies on early detection, which is challenging due to asymptomatic early stages.
• Machine learning and deep learning show promise in classifying cancerous cells in medical images.
• Current methods often involve segmentation of whole slide images (WSI) before cell classification, where segmentation accuracy impacts overall performance.

Purpose of the Study:

• To develop and evaluate a CNN-based method for accurate classification of cervical cancer cells.
• To investigate a direct classification approach for WSI cervical cell clusters, bypassing the need for segmentation.
• To compare the performance of the proposed method against existing benchmarks.

Main Methods:

• Utilized transfer learning with pre-trained convolutional neural network (CNN) models.
• Applied deep learning approaches to extract features for classifying Pap smear images.
• Developed a direct classification method for WSI cervical cell groups, eliminating the segmentation step.

Main Results:

• Achieved high classification accuracy: 96.74% for WSI patches and 97.55% for full-cell images on the SIPaKMeD dataset.
• Attained 90.42% accuracy on the Herlev dataset.
• Demonstrated that direct classification of WSI cervical cell groups is effective and segmentation is not strictly necessary for high performance.

Conclusions:

• The proposed CNN-based method accurately distinguishes between cancerous and non-cancerous cervical cells.
• Direct classification of WSI cervical cell groups offers a viable and efficient alternative to segmentation-based approaches.
• This approach holds potential for improving early cervical cancer detection through automated image analysis.