Summary: | 碩士 === 國立宜蘭大學 === 資訊工程學系碩士班 === 106 === Machine learning has been applied in many fields so far. In medical diagnosis, machine learning can be used in practical scenarios: help doctors to diagnose, predict diseases, administer treatment, predict the life expectancy, and even perform surgery. As far as the disease treatment is concerned, cancer, the second leading cause of death globally, is undoubtedly the most challenging one.
To slow down the impact of cancer on human beings, corresponding predictions based on a large amount of clinical information have been integrated with doctors' medical diagnosis to find the best solutions for patients. Survival analysis, one of the most common categories in medical science, means the method to predict the survival time of cancer patients according to the existing information. In the past, common algorithms for survival analyses include decision trees, Bayes classifier, logistic regression, support vector machine, random forest and so on. Some of them in fact used neural networks for forecasting but the outcomes might be bad due to the inefficiency of computers at that time. Now, neural networks have been applied to the GPU to solve complicated problems. Different from previous neural networks, current neural networks have more hidden layers and we've got a new term "deep neural network."
This paper uses lung cancer patients' body data retrieved from the public database SEER (Surveillance, Epidemiology and End Result Program) of the National Cancer Institute to predict the survival time. We furthermore compare the result with other algorithms, including decision trees, random forest, support vector machine, Bayes classifier and logistic regression. The analysis shows that the deep neural network obviously outperforms others.
|