Cervical Pap Smear Screening and Cancer Detection Using Deep Neural Network
Munakala Lohith, Soumi Bardhan, Oishila Bandyopadhyay · 2023
Cervical cancer is considered as a common type of cancer among Indian female population. One of the main causes of cervical cancer deaths is late diagnosis. Early detection of the disease helps to ensure a faster recovery. Pap smear is the common screening method to identify the suspicious cases of cervical cancer at the initial stage. Hence, the development of an automated tool for fast screening of Pap smear images would be beneficial. In this work, an integrated convolutional neural network (CNN)-based approach is being proposed to detect and classify cervical cell abnormalities as per Bethesda system. The CNN-based YOLOv5(L) architecture is used for cell classification in multi-cell pap smear images. The proposed model is fine-tuned and tested on an exhaustive pap smear dataset SIPaKMED with five different classes of cells (parabasal, metaplastic, dyskeratotic, koilocytotic, and superficial intermediate). It can accurately predict the respective class of different types of cells present in the multi-cell test sample. The model has achieved an average mAP score of 0.59, which appears much better than other similar approaches.