الفهرس 
MOTIVATIONOnly 14 pages are availabe for public view
 |  | from | 96 |  |  |
| | | from | 96 | | |
Abstract
Liver Cancer is one of the main causes of cancer-related deaths worldwide. Due to the extreme heterogeneity of this disease, its prognosis and management are still not yet standardized. Different treatment modalities are available, however, the patient’s response to each of them varies; Therefore, it’s critical to establish a model to help physicians individualize the management of this aggressive tumor.
This work presents one of the first investigations into personalizing liver cancer treatment for patients with genotype 4 using their clinical and genetic data. The aim of this study is to build a model that can predict the best treatment for a liver cancer patient with genotype 4, which could mitigate the impact of the disease and overall improve the patient’s prognosis.
Firstly, we proposed two pipelines, a Single-Model pipeline and a Multi-Model pipeline, for recommending the best liver cancer treatment for patients and therefore potentially improve their prognosis. The data of 1427 Egyptian patients with liver cancer was analyzed. We studied the performance of six regression methods in predicting a patient’s survival months after undergoing a certain treatment. The best performing method which was Random Forest was used in building the models in the proposed pipelines. The performance of both pipelines was compared, the Single-Model pipeline proved to be more reliable than the Multi-Model pipeline with 80.6% successful treatment recommendations which is higher than the prediction accuracy achieved in existing work. Moreover, we also prove the crucial importance of genetic data and their effect on patients’ prognosis and response to treatment.
Secondly, we built a model to predict the HCC patients’ survivability and extract important features that affect their prognosis. Three feature selection methods and seven data mining classification techniques were studied and compared. The model’s features were selected using Wrapper Subset method and built using Random Forest. The model’s predictive accuracy was 87.54%. Moreover, important genetic features such as p53 gene exon 6 and 9 mutations proved to have a significant impact on the patient’s overall prognosis.