Site Editor

Thomas Flaig, MD

Advertisement
Advertisement

Novel Machine Learning Approach to Medical Decision-Making in Bladder Cancer

By: Emily Rhode
Posted: Thursday, December 15, 2022

It may be possible to avoid invasive cystoscopy when gene-expression signatures are available for risk assessment and monitoring of patients with bladder cancer. Findings presented in the journal BioData Mining outlined the results of a proposed new analysis pipeline for the modeling of disease risk and outcomes from biomarkers related to the prediction and treatment of bladder cancer. Mauro Nascimben, a PhD student at the Università del Piemonte Orientale, Novara, Italy, and colleagues demonstrated that analysis pipelines using Uniform Manifold Approximation and Projection (UMAP) methodology may more accurately predict bladder cancer outcomes. The researchers also employed machine learning procedures to show the effectiveness of preprocessing in patient condition predictions while identifying a subgroup of biomarkers that may help estimate prognosis in these patients.

“The numerical experiments in the current investigation testing three distinct preprocessing sequences based on single or double discretizations helped to discriminate more effectively six possible patients’ outcomes given a bladder cancer gene-expression data set from a cross-sectional study,” the authors concluded.

In this study, the authors used gene-expression data from the bladder cancer biomarkers of 386 patients with bladder cancer. Preprocessing data discretization was performed using log-z, uniform, or normal data mapping. Tree ensemble embedding identified extremely randomized trees as the foremost model, which was then dimensionally reduced and clustered before applying the matrices to two experimental conditions: complete and partial embedding of the gene-expression data. The complete embedding model used t-distributed stochastic neighbor embedding (t-SNE) and UMAP to produce a prognostic map of population patterns in patients with bladder cancer. The partial embedding model used 75% of the data to train the tree ensemble for the purposes of checking model reliability.

The researchers noted that several UMAP parameters influenced the results, and therefore they recommend that future researchers pay particular attention to the UMAP process.

Disclosure: The authors reported no conflicts of interest.


By continuing to browse this site you permit us and our partners to place identification cookies on your browser and agree to our use of cookies to identify you for marketing. Read our Privacy Policy to learn more.