A hybrid compound scaling hypergraph neural network for robust cervical cancer subtype classification using whole slide cytology images.

Cervical cancer is a major cause of mortality among women, particularly in low-income countries with insufficient screening programs. Manual cytological examination is time-consuming, error-prone and subject to inter-observer variability. Automated and robust classification of the whole slide cytology images for cervical cancer is essential for detecting precancerous and malignant lesions. We propose a novel deep learning framework, the Compound Scaling Hypergraph Neural Network model (CSHG-CervixNet), for robust classification of cervical cancer subtypes. The model integrates a Compound Scaling Convolutional Neural Network (CSCNN) with a k-dimensional Hypergraph Neural Network (kd-HGNN) architecture. CSCNN balances the network's depth, width, and resolution, supporting effective feature representation with minimal computational overhead. kd-HGNN captures higher-order relationships between the features, and its propagation mechanism ensures better feature diffusion across distant nodes. The model is evaluated on the benchmark Sipakmed dataset and achieves an accuracy of 99.31%, with a macro-averaged precision of 98.97%, recall of 99.38%, and F1-score of 99.34%, demonstrating its robustness in cervical cancer subtype classification. Pathologists and other medical experts will find this study helpful in distinguishing cervical cancer subtypes so that targeted treatment may be provided and effective disease management is made possible.