Enhancing cervical cancer diagnosis with ensemble learning and shark optimization algorithm: comparative study of CT and MRI in cervical cancer diagnosis

Cervical cancer, one of the most common female cancers, can be detected with computed tomography (CT) and magnetic resonance imaging (MRI). Computer-aided diagnosis (CAD) methods based on artificial intelligence have been widely explored to improve traditional screening methods for cervical cancer detection. This study aims to compare the accuracy of CT and MRI in diagnosing cervical cancer using a novel methodology that combines the Large Vision Model (LVM) and InternImage, which reduces the misclassification of cervical tumors, especially in benign and malignant cases. InternImage (based on InceptionV3) extracts pre-trained deep features, making it more sensitive to tumor-specific patterns. At the same time, LVM focuses on fine-grained spatial features, helping to classify early changes in cervical pathology. In the Shark Optimization Algorithm (SOA), the procedure dynamically selects the optimal weight parameter, avoiding overreliance on a single model. This application improves generalization across different CT and MRI datasets. The performance of the proposed model is evaluated on two new datasets, KAUH-CCTD and KAUH-CCMD, collected from King Abdullah University Hospital (KAUH) in Jordan. The proposed model classified images into three categories: benign, malignant, and normal. The proposed model achieved the best performance in diagnosing CT images, with an accuracy of 98.49%, while achieving an accuracy of 92.92% in diagnosing MRI images. CT imaging, especially MRI, can detect tumor extension into the cervical stroma, which could change treatment approaches. Additionally, imaging plays a crucial role in monitoring treatment and patient progress to detect early disease relapses.