An Integrated Knowledge Distillation Approach with Lightweight Neural Network for Cervical Cancer Classification

Cervical cancer is one of the most common and deadly cancers that affects women across global. Frequent screening of the cervix can help to avoid the growth of pre-cancerous cancer by helping early detection and subsequent treatment. The Papsmear test is widely applied for screening the early detection of cervical cancer; however, cervical cancer is labor-intensive, time-consuming as well as susceptible to human mistakes. A lightweight neural network (LNN)-based integrated knowledge distillation strategy has been developed using a model for cervical cancer classification. By using data from a skilled teacher, strategies to train an LNN model for the student learner, this model provides an effective architecture. This study selects an LNN in order to generate a highly effective model through a few parameters and computations. Moreover, SPIkAMeD is the most extensively used publicly available dataset for classifying cervical cancer. The proposed model achieved a computational time of 80.1 s, an accuracy of 99.1%, a precision of 98.655%, a recall of 99.11%, an F1-score of 98.97% and an AUC of 0.99. A confusion matrix and k-fold validation were used for additional evaluation, demonstrating the model’s effectiveness. This integrated model may increase the efficacy and cost-effectiveness of a computer-aided diagnostic (CAD) system for cervical cancer screening and prevention.