Cervical cancer prediction using deformable kernel darknet-53 and depth wise separable convolutional neural networks.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Oncology

Background:

• Cervical cancer (CC) prediction is challenging due to data limitations and diverse clinical presentations.
• Early CC signs often lack distinct characteristics, complicating accurate identification.
• Inconsistent diagnostic methods and dataset variations hinder the development of generalized models.

Purpose of the Study:

• To develop an advanced AI framework for robust and accurate cervical cancer prediction.
• To address challenges of data heterogeneity and early-stage detection in CC diagnosis.
• To create a model with high generalization capabilities across different datasets.

Main Methods:

• Introduced Deformable Kernel Darknet-53 with Depth-Wise Separable Convolutional Neural Network (DK-D53-DWSCNNet).
• Employed Deformable Kernel Networks for Joint Image Filtering (DKNet-JIF) for image enhancement.
• Utilized an ensemble Darknet-53 CNN with Contextual Attention Network (D53-CNN-CAN) for segmentation.
• Integrated Geometric Algebra Transformers (GAT) with depth-wise separable CNNs for feature extraction.
• Optimized training using the Hyperbolic Sine Optimizer (HSO).

Main Results:

• Achieved 99.9% accuracy and 99.8% sensitivity on Herlev and SEER datasets.
• Demonstrated remarkable generalization capabilities across diverse datasets.
• The DK-D53-DWSCNNet effectively captured both edge and contextual features while reducing redundancy.

Conclusions:

• The DK-D53-DWSCNNet shows significant potential for improving early and robust cervical cancer prediction.
• The model's performance suggests its suitability for integration into clinical diagnostic procedures.
• This framework offers a promising solution for overcoming limitations in current CC diagnostic methods.