Please use this identifier to cite or link to this item:
Full metadata record
DC FieldValueLanguage
dc.contributor.supervisorSingh, Ashima-
dc.contributor.authorDhillon, Arwinder-
dc.description.abstractHigh-dimensional datasets comprising genomic data, proteomic data, pathological images data and so on have taken noticeable toll now-days in healthcare research. Handling these datasets require appropriate knowledge about tools and techniques for efficient prediction like disease detection, survivability analysis, and biomarker identification etc. Researchers are working hard to achieve high accuracy in such predictions but the performance achieved in predictions for breast cancer patients is not sufficient due to high risk diseases like cancer. In this thesis, we proposed a framework called Genomic Pathological Extreme Learning Machine (GPELM), which is an invariant of Extreme Learning Machine (ELM). This package comprises of ensembling of six different models. These models consist ELM with Buckley James estimator (ELMBJ), Ensemble of ELMBJ (ELMBJEN), ELM with regularized Cox model (ELMCOX), Ensemble of ELMCOX (ELMCOXEN), ELM with gradient based boosting (ELMBOOST) and ELM with likelihood-based boosting (ELMCOXBOOST). The dataset has integrated genomic data (gene-expression, copy number alteration, DNA methylation, protein expression) and pathological images data for breast cancer survival prediction. These six models are present in the survELM package. GPELM is compared with other cox models and their related performance parameters comprising sensitivity, precision, accuracy, MCC, AUC, AUPR, hazard ratio and concordance index are calculated. GPELM has achieved 85% of accuracy for breast cancer survival prediction and there is 5% increase in each of the performance parameter taken into consideration. The purpose of this thesis is to predict the survival of breast cancer patients with best accuracy. GPELM gives a commendable contribution in survival prediction of breast cancer patients as very little increase in performance is considered significant for breast cancer survival. All the results achieved in the present research show the usefulness of GPELM for breast cancer survival prediction. GPELM package is also implemented and tested on lung cancer data. Therefore, the package can also be implemented on more disease datasets having simple clinical, genomic, images and integrated datasets.en_US
dc.subjectGenomic profilesen_US
dc.subjectPathological imagesen_US
dc.subjectIntegrative Analysisen_US
dc.subjectSurvival analysisen_US
dc.subjectBreast cancer survival predictionen_US
dc.subjectExtreme Learning Machineen_US
dc.subjectSurvival Modelsen_US
dc.titleGPELM: An Integrative Analysis for Breast Cancer Survival Predictionen_US
Appears in Collections:Masters Theses@CSED

Files in This Item:
File Description SizeFormat 
Arwinder Dhillon_801732006_MECSE.pdf2.52 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.