A two stage deep supervised learning model with inter-layer weight sharing for multi class medical image classification

Cervical cancer is a lethal condition that develops in the uterine cervix of sexually active women. Despite years of research, it remains an immense global risk that can be detected early with the help of a Papanicolaou Smear (Pap smear) test. However, manual Pap smear screening under a microscope is subjective, and the process is challenging for the health care professionals. This work aims to develop an efficient computer-assisted prediction model based on Pap smear images for the identification of cervical cancer using Convolution Neural Network (CNN). We have introduced an efficient two-stage classification model employing the Deep Learning (DL) approach named ChannelMixer, specifically designed to analyze Pap smear images. This approach effectively classifies a dataset labelled with five distinct classes, namely Koilocytotic, Dyskeratotic, Metaplastic, Superficial Intermediate and Parabasal, into two final categories, i.e., Abnormal and Normal cells. The presented model attains an impressive accuracy of $$96.95\%$$ across the five-class classification and $$98.68\%$$ in the two-class category, as determined through a rigorous five-fold cross-validation process. A comprehensive assessment of the proposed model encompassing diverse evaluation metrics, including accuracy, precision, F1-score, recall, loss analysis. The ChannelMixer DL model introduced in this study surpasses state-of-the-art approaches and exhibits better performance compared to other well-known pre-trained Convolution Neural Network models.