Performance Comparison of Conventional Neural Networks and Deep Learning Network For Cervical Cancer Diagnosis

Authors

  • Chandra Prabha R
  • Seema Singh

Abstract

Cervical cancer is the fourth-most common cause for death from cancer in women. Efforts are being made to develop more efficient techniques for the detection of cancer at the initial stage. Conventional methods require expert pathologists to examine the biopsy slide and classify it. In this regard few concerns have risen such as the deficiency of expert pathologists, lack of technical support to doctors and also lack of awareness among women especially in rural areas. Hence there is a requirement for an effective and accurate system that detects cervical cancer which can be used by health worker to detect cancer at initial stage (as a part of basic health check-up). This paper describes and compares two techniques for the cervical cancer diagnosis. The first   method involves extraction of key features from complex cytology images using image processing algorithm followed by a neural network classifier with back propagation algorithm using MATLAB tool. The major challenge faced in this method is extracting the key features from complex images with overlapping cells, which is further used by neural network for classification. The other method is based on deep learning that uses inception neural network with tensor flow. A comparative analysis is presented for the same image database which is created with a Bangalore based pathology laboratory. The database is of 460 images of which 197 images are cancerous and 263 are non-cancerous images. The analysis proved that deep learning method was able to provide better classification results.

Downloads

Published

2020-01-27

Issue

Section

Articles