Lightweight CNN-Based Multiclass Classification of Cervical Cytology Images for Computer-Aided Cervical Cancer Screening

Cervical cancer is one of the most important causes of cancer death among women worldwide, especially in areas where there is a shortage of access to skilled cytopathological screening. Manual inspection of Pap smear images is labor intensive and can be affected by inter-observer variability, thus the need for reliable automated diagnostic systems. In this study, the authors present a novel, lightweight Convolutional Neural Network (CNN) which is capable of multiclass classification of cervical cytology images from the SIPaKMeD database of 4,049 annotated cell images from five categories of cervical cells. To enhance the model robustness and generalization, image preprocessing methods like Adaptive Histogram Equalization and data augmentation techniques (e.g., rotation, flipping, zooming, and brightness adjustment) were used. The experimental results presented in terms of accuracy, precision, recall, F1-score, AUC and confusion matrix analysis showed high classification accuracy of 94.3% across all the classes. The performance of the proposed CNN was compared with the VGG16, ResNet50, DenseNet121 and MobileNetV2, and it was found that the proposed CNN yielded competitive or better results with much fewer trainable parameters and lower computational complexity. The proposed system can enable computer-aided cervical cancer screening, which could lead to early diagnosis of cervical cancer in a faster, cost-effective and scalable way in less-developed health care settings.