Restricted Boltzmann Machine and its Potential to Better Predict Cancer Survival
Journal Title: Biomedical Journal of Scientific & Technical Research (BJSTR) - Year 2018, Vol 6, Issue 1
Abstract
Traditional methods to predict cancer survival include Competing-Risk Regression and Cox Proportional Hazards Regression; both require the hazard of input variables to be proportionate, limiting the use of non-proportionate measurements on miRNA inhibitors and inflammatory cytokines. They also require imputation at missing data before prediction, adding fallible workloads to the clinical practitioners. To get around the two requirements, we applied Restricted Boltzmann Machine (RBM) to two patient datasets including the NCCTG lung cancer dataset (228 patients, 7 clinicopathological variables) and the TCGA Glioblastoma (GBM) miRNA sequencing dataset (211 patients, 533 mRNA measurements) to predict the 5-year survival. RBM has achieved a c-statistic of 0.989 and 0.826 on the two datasets, outperforming Cox Proportional Hazards Regression that achieved 0.900 and 0.613, respectively.The rapid development of new computational methods and tools for data analysis and building predictive models has enabled more precise cancer prognosis. Use the prediction of the risk of recurrence after liver transplantation as an example, in the recent years, researchers and clinical practitioners have discovered several clinicopathological variables that play an important role, such as the number and size of tumors and the level of alpha-fetoprotein (AFP). There were a few significant studies of patients who underwent primary liver transplantation; to name a few, Mazzaferro et al. [1] studied a cohort of 1,018 patients at three tertiary centers in Italy and achieved an average c-statistic of 0.780 on predicting 5-year risk of HCC-related death using 2 variables including: a) Sum of tumor number and size and b) Logarithmic level of AFP. Ling et al. [2] studied a cohort of 1,010 patients extracted from the China Liver Transplant Registry database and achieved a c-statistic of 0.767 on predicting 2-year risk of HCC recurrence using 4 variables including: a) Cold Ischemia Time, b) Tumor Burden, c) Differentiation And d) Afp. Mehta et al. [3] studied a cohort of 1,061 patients at 3 academic transplant centers including the University of California-San Francisco; Mayo Clinic, Rochester; and Mayo Clinic, Jacksonville, and achieved a c-statistic of 0.82 on predicting 5-year risk of HCC recurrence using 3 variables including: a) Micro Vascular Invasion b) Afp and c) The sum of the Largest Viable Tumor Diameter and Number of Viable Tumors on Explant. Agopian et al. [4] studied a cohort of 865 patients at University of California, Los Angeles from 1984 to 2013 and achieved a c-statistic of 0.85 on predicting 5-year risk of HCC recurrence using 8 variables including: a) Nuclear Grade, b) Macrovascular Invasion, c) Milan Criteria, d) Nonincidental and Radiologic Maximum Diameter e) Microvascular Invasion, f) Neutrophil-Lymphocyte Ratio, g) Afp and h) Total Cholesterol. The first study is using Competing-Risks Regression, while the second, third and fourth studies are using Cox Proportional-Hazards Regression to analyze the impact of potential factors on patients’ recurrence or survival. With Competing-Risks Regression, one focuses on the cumulative incidence function that indicates the probability of the event of interest happens before a certain time, while in Cox Proportional Hazards Regression, one instead focuses on the survival function that indicates the probability or survival beyond a certain time. Although both the Competing-Risk Regression and Cox Proportional Hazards Regression method have been widely adopted in biomedical research for investigating the association between the recurrence time and survival time of patients and one or more predictor variables, there are requirements to satisfy for the two methods to work correctly
Authors and Affiliations
Ruibang Luo, Wen Ma, Tak-Wah Lam
Keywords
Related Articles
- EP ID EP588462
- DOI 10.26717/BJSTR.2018.06.001305
- Views 168
- Downloads 0
Two Cases of 45,X/47,Xxx Turner Mosaicism With Different Growth Patterns Possibly Caused By Different Growth Hormone Therapies
We reported two Japanese girls with 45,X/47,XXX Turner mosaicism, who showed different growth patterns possible caused by different growth hormone (GH) therapies. The first case was diagnosed with 45,X/47,XXX mosaicism a...
Vermitechnology Based Tribal Women Empowerment for Economic Development in Himachal Pradesh
A novel technique of converting decomposable organic wastes into valuable manure (compost) through earthworm activity is a faster and beneficial process. The earthworms are used as the natural bioreactors for making deco...
Retention of Probase Hot Versus the Conventional Heat- Cured Acrylic Resin Denture Bases
This study compared the retention of proBase hot versus a conventional heat-cured acrylic resin denture bases. Two completely edentulous patients of U-shaped and V-shaped palatine vault. Primary and secondary impressions...
Painful Cutaneous and Subcutaneous Tumors Accompanied with Vascularized Appearance using High-Resolution Ultrasound in Dermatology: The Acronym “England” or “Lend an Egg”
The acronym “ENGLAND” or “LEND AN EGG” is often used to recall these painful tumors in dermatology. “LEND AN EGG” tumors show: leiomyoma, eccrine spiradenoma, neuroma, dermatofibroma, angiolipoma, neurilemmoma, endometri...
Kienbock Disease, a Tertiary Care Experience from the Developing World
Introduction: Kienbock’s disease is a form of osteonecrosis of lunate, first described in 1910 by Robert Kienbock. It occurs most commonly in men aged 20 to 40 years of age. Its exact etiology is still under study but mo...