@misc{10902/14573,
year = {2018},
month = {6},
url = {http://hdl.handle.net/10902/14573},
abstract = {Currently, according to some estimates diagnostic errors contribute to approximately 10 percent of patient deaths, which explains the interest in incorporating new technologies into this process. With the significant developments made in the last decades, Artificial Intelligence (AI) has arguably become a good candidate for this task. Specifically, our project involves using Machine Learning (a subfield of AI) to improve childhood cáncer treatment and diagnosis.
We will focus on Acute Lymphoblastic Leukemia (ALL), the most common type of pediatric cancer with around 300 yearly diagnoses in Australia. In ALL, relapse decreases the chances of survival as standard chemotherapy becomes ineffective. Thus, modified therapy is needed for patients with high risk of relapse. In this project, we explore the use of Machine Learning to accurately predict relapse, which would ultimately result in better treatment. This study will be based on a dataset containing genetic information, which will be the basis for predictions, and cancer outcome (relapse/mortality) of about 150 patients.
The low number of samples, combined with the high proportion of non-relapse cases, means there are very few relapse examples, which complicates finding meaningful patterns to make accurate predictions. Furthermore, high dimensionality creates difficulties when trying to achieve generalizable solutions. To address these challenges, we explore the use of biased classifiers, particularly sparse linear methods; dimensionality reduction techniques both supervised (univariate feature selection, LDA) and unsupervised (PCA, Autoencoders); ensemble approaches, especially bagging and undersampling/oversampling methods (ENN).},
title = {Artificial intelligence for cancer treatment and diagnosis},
author = {Martínez de la Pedraja García, Juan},
}