Cervical cancer cell prediction and detection with high test accuracy based on a reliable automated system

Cervical cancer ranks as the fourth most prevalent cancer among women worldwide. The increasing incidence and mortality rates are largely attributed to limited screening infrastructure, a shortage of skilled medical personnel, and inadequate awareness-particularly in developing nations. Conventional diagnostic methods, including the Pap smear, HPV testing, histopathology, and visual inspection with acetic acid (VIA), are often prone to human error due to subjective interpretation by clinicians. To address these limitations, this study proposes a robust automated framework for cervical cancer detection and cervix type classification using advanced deep learning techniques with minimal data requirements. A total of 915 histopathology images and over 4,000 colposcopy images were collected from public repositories and local healthcare institutions. Several pre-trained convolutional models were evaluated for cervix type classification. The proposed system employs a lightweight GoogleNet architecture to identify the transformation zone, followed by an EfficientMobileNet model for cervix type classification and cancer detection. Experimental results demonstrate superior performance, achieving 96% accuracy for cervix type classification, 95% accuracy for cervical cancer detection, and 99% mean average precision (mAP) for region-of-interest (ROI) localisation. The proposed approach offers an effective and resource-efficient diagnostic solution, particularly beneficial for regions with constrained healthcare infrastructure and limited clinical expertise. The proposed model gives 97% recall, 97% specificity, 97% accuracy, 96% mAP, 97% F1-measure and 99.8% AUC, respectively, compared to existing approaches like VGG16, ResNet50, CNN, ConvNet_1, ConvNet_5 and ConvNet_10.