- author = "Hasseeb Azzawi",
- title = "Computational models and approaches for lung cancer diagnosis",
- school = "Deakin University",
- year = "2019",
- address = "Australia",
- month = oct,
- keywords = "genetic algorithms, genetic programming, gene expression programming",
-
URL = "
https://hdl.handle.net/10536/DRO/DU:30157091",
-
URL = "https://dro.deakin.edu.au/ndownloader/files/37464802/1",
- size = "128 pages",
-
abstract = "Cancer is a main problem of public health and remains
the main cause of death worldwide. Lung cancer is the
most prevalent cancer which threatens the human's life
more than any other malignant cancer. Over the past
decades, scientists applied different diagnosis methods
in order to predict the type of cancer before causing
symptoms. Moreover, they have developed new strategies
for the prediction of lung cancer. Among the most
interesting and challenging tasks for physicians and
biological researchers the attractive one is to find
the effective and accurate tools to diagnose the cancer
before it spreads or metastasizes to regional lymph
nodes.
The microarray technology based gene expression profiling, which is one of the high throughput technologies, is great helpful to the diagnosis of lung cancer in effective way. Also, it is provides substantial advances to our understanding of lung cancer biology at the molecular level. However, when using microarray data for cancer diagnosis it is important to analyse the microarray data because of the characteristics it has such as: a large number of gene expressions with a small number of samples, noise in the expressions, and high correlations among gene expressions. Consequently, diagnosis tasks need computational models and approaches to analyse data and extract patterns or knowledge that assists in achieving satisfactory diagnosis results. To this end, machine learning methods were the proper choices. However, the direct application of these methods might lead to over-fitting problems.
In the recent past, researchers have developed many algorithms that identify molecular signatures to solve the over-fitting problems. Unfortunately, the existing algorithms are suffering from some limitations like unsatisfactory results, low performance, dependency on the training sets, and lack of generalization. The questions of how to effectively diagnose lung cancer patients still remain unanswered. Therefore, novel methods that are able to identify robust genes to improve the performance of diagnosis are greatly needed.
This need was the reason why we developed novel lung cancer diagnostic algorithms in this research. These algorithms address the issues mentioned above and provide better lung cancer diagnosis performance. The evaluations of our proposed approaches show that these algorithms are effective in increasing the reliability of lung cancer patient diagnosis. Also, prediction performance evaluations and comparisons between proposed approaches and other representative machine learning methods demonstrate the superior performance of our proposed approach. Furthermore, statistical evaluations show the effectiveness and stability of the proposed algorithms.",
- notes = "CRICOS Provider Code: 00113B",
