Detection and Classification of Cervical Cancer Using Optimized Deep Learning Approach

Given the global impact of the cervical cancer epidemic on peoples health, it is critical to have readily available and effective screening technologies. In order to effectively fight this disease, it is crucial to identify the groups who are most at risk. Our study aims to build a robust deep learning system tailored for the classification of cervical cancer using images acquired from Pap screenings; this will allow us to tackle this challenge head-on. Our approach improves upon previous methods of visual feature detection by using a deep learning model based on transfer learning called Squeeze-and-Excitation-ResNet152. The Deer Hunting Optimization method is used to optimise the network by modifying its hyper-parameters. We test our methods on eleven distinct disease sets, with a total of 8838 images distributed differently. Sources such as CRIC and SIPaKMeD were consulted for the acquisition of these images. In order to reduce dataset bias, we use a cost-sensitive loss function all through training. Impressive performance metrics were generated by the testing set analysis, which significantly outperformed the previous methods. Accuracy was 99.74%, precision was 98.98%, recall was 98.15%, specificity was 98.97%, and F1-Score was 99.06%. We can greatly enhance the identification of problems connected to cervical cancer with our technologies. One way to achieve this goal is to make it easier for medical professionals to make quick diagnosis.