InceptionNet-based Deep Learning Framework for Reliable Classification and Diagnosis of Cervical Cancer using Clinical Data Approaches

Cervical cancer is a major morbidity and mortality condition among women necessitating a better diagnostic approach to provide an early intervention. The focus of the present study is to identify and design a deep learning model, based on the InceptionNet architecture, able to classify and diagnose cervical cancer through clinical data in an accurate way. The high-class imbalance and the prevalence of negative biopsy cases among the labelled instances were used to preprocess the dataset and analyse it using exploratory data analysis, such as distribution and box plot evaluation of that important feature. We used InceptionNet, which is an architecture optimized to handle clinical tabular data, to distinguish the benign and malign outcome of the biopsies. Although training was done on comprehensive models, the results showed poor performance in identifying positive cases (sensitivity: 0%), and the area under ROC curve was also low (AUC = 0.46), a characteristic of minority class detection issues. These results indicate the necessity of better imbalance handling and feature engineering to better discriminate the diagnoses in realistic clinical practice.