Data Augmentation and Pretrained Transfer Learning Approaches for Automated Cervical Cancer Detection

Abstract

Cervical cancer is still the top cause of death in women around the world, especially in places where early screening and diagnosis services are hard to get to. Thanks to progress in artificial intelligence (AI) and medical images, automatic recognition systems can help doctors make more accurate diagnoses and do less work by hand. This research shows a full system that uses extra data and pre-trained transfer learning models to automatically sort medical images of cervical cancer into different types. To deal with the problems that come with having small and uneven datasets, a full data enhancement process is set up. This adds changes in direction, brightness, and contrast to the datasets to help the models generalise better. A number of cutting-edge convolutional neural networks (CNNs), such as ResNet50, VGG16, VGG19, and MobileNet, are fine-tuned on the new dataset. Performance measures like accuracy, precision, recall, and F1-score are used to judge the models. It was shown in experiments that the fine-tuned ResNet50 model works better than others, with an F1-score of 93.9% and an accuracy of 94.8%. This proves that transfer learning works in the medical imaging area. The suggested method also cuts down on training time by a large amount and improves classification performance while using very few computing resources. This study shows how mixing enhancement with pre-trained models improves stability and scaling. This provides a useful way to help doctors find cervical cancer early and accurately. Multi-modal data fusion and real-time application in healthcare situations will be looked into in more detail in the future..