Development of a decision support system based on clean architecture for cervical cancer diagnosis based on pap smear images and artificial intelligence techniques

Cervical cancer is a type of cancer that arises from the uncontrolled growth and proliferation of cervical cells and has a higher treatment success rate compared to other cancer types once diagnosed. This thesis aims to evaluate the effectiveness of deep learning models in diagnosing cervical cancer. Experiments were conducted using the SIPaKMeD dataset, various image preprocessing techniques, and deep learning models. Blur, Bilateral, Blur-Bilateral, Median-Blur, and Gaussian-Blur filtering techniques were applied, and their performances were examined. The study tested and compared the performances of Inception-ResNet-V2, MobileNetV2, NASNetLarge, ResNet101, and VGG19 models. The results showed that Inception-ResNet-V2 and MobileNetV2 models stood out with high accuracy rates. Specifically, the Blur filter was found to be the most effective with an average F1-score of 0.8088. In the ROC curve analysis, the Inception-ResNet-V2 model had the highest AUC value, indicating the best classification capability. The web-based decision support system developed within the scope of the thesis helps users analyze Pap smear results and make diagnoses. Its user-friendly interface allows healthcare professionals and other users to easily input data and interpret the results. This system can be used as a reliable and effective diagnostic tool in clinical applications. This study demonstrates the effectiveness of deep learning models and various preprocessing techniques in diagnosing cervical cancer and makes significant contributions by offering a decision support system for healthcare professionals and other users.