Cervical Cancer Image Classification Using CNN Transfer Learning
Deny Arifianto, Ali Suryaperdana Agoes · Advances in engineering research/Advances in Engineering Research · 2021
Cervical cancer is a major global public health problem, Indonesia is among top 3 countries in the world with the highest number of cervical cancer incidents.An early diagnosis for cervical cancer is one of the key approaches to prolong patient's life expectancy.The Papanicolaou (Pap smear) test is a cervical cancer screening test that has been widely utilized.Pap smear test is a tedious, labour-intensive, and time-consuming task, which leads to high inter operator's variability.A computer-based classification algorithm to assist the task has been proposed.In this paper we focus on the approaches using Convolutional Neural Network (CNN) to handle the classification task.Moreover, our proposal employs a parameter efficient model.Thus, the computational cost is greatly reduced.We use a transfer learning method for model adaptation.We trained the pre-trained SqueezeNet architecture with the three class of pap smear images dataset in caffe.The fine-tuning process was started with the initialization of the model's features to the object's broader spectrum.Then, the last layer output number was changed to fit the number of labels for cervical cancer class.