Deep and Domain Specific Feature-Based Cervical Cancer Classification Using Support Vector Machine Optimized With Particle Swarm Optimization
Ritesh Maurya, Lucky Rajput, Satyajit Mahapatra · IEEE Access · 2024
Cervical cancer presents a significant threat to the global healthcare system and its early detection is a challenging task. Cervical cancer is detected through manual evaluation of Pap smear images by expert pathologists. Computer-aided diagnosis (CAD) systems developed using machine learning and deep learning-based algorithms present a promising solution. This work proposes a novel approach by combining the deep features extracted from MobileNetV1 model with the domain-specific shape features, texture features and color features for cervical cancer classification. The noisy features have been eliminated by utilising the concept of mutual information whereas, a support vector machine optimized with Particle Swarm Optimization (PSO) was used for the classification. The proposed methodology is evaluated on the publicly available Sipakmed dataset, comprising 4049 sample images across five different classes. The suggested approach obtains an accuracy of 97.9% in Pap Smear cell image categorisation task. Thus, is presents a promising approach for the cervical cell categorisation task.