Dr. Freshteh Osmani
University of Birjand, IranTitle: Comparison of breast cancer recurrence prediction by random forest, artificial neural network and Cox regression modols
Abstract
Breast cancer is
one of the
most common cancers,
and also it is
the most
common type of malignancy in Iranian women that has been growing in
recent years. In patients with this disease, there is usually the risk of
recurrence. Many factors increase or decrease the risk of recurrence. A
variety of models
that can be noted to
detect and predict breast cancer
recurrence and factors affecting it, classical statistical methods such as
Cox Regression and new methods
such as Artificial Neural
Network and Random Forest. In this
study, using these three
techniques, recurrence of breast
cancer was predicted and
factors affecting relapse were detected. Models and prediction Accuracy are compared
using ROC curve. In this retrospective study, also called historic cohort
study, 342 patients with breast cancer with 13 features from each patient were
used.
Data of patients in the Cancer Research Center of Shohada Tajrish
hospital, have been recorded to track, and patients for at least 6
months after diagnosis were being monitored and follow-up was done for
them. To
develop predictive models
of breast cancer recurrence
and determination of factors affecting
recurrence, the three-layer Perceptron
Artificial Neural Network algorithms, Random Forest classification and
Cox Regression models were
fitted. Area under
the ROC for
the Artificial Neural Network,
Random Forest and Cox Regression
obtained 87%, 76%
and 65%, respectively. Artificial
Neural Network model classification accuracy is obtained 80%, against 60% for
random forest model and 33% for cox regression model. So, ANN model predicted
better than the two other models.
Biography
Freshteh Osmani completed his
Ph.D. at the age of 32 years from Tarbiat Modares University, Tehran, Iran. She is an assistant professor of
biostatistics and the head of the Student Research Committee of the dentistry
department. She has over 70 publications that have been cited over 250 times,
and his publication h-index is 11. He has been serving as an editorial board
and reviewer member of several journals.