Machine learning helps identifying volume-confounding effects in radiomics

Alberto Traverso; Michal Kazmierski; Ivan Zhovannik; Mattea Welch; Leonard Wee; David Jaffray; Andre Dekker; Andrew Hope

doi:10.1016/j.ejmp.2020.02.010

Machine learning helps identifying volume-confounding effects in radiomics

Alberto Traverso^*, Michal Kazmierski^*, Ivan Zhovannik, Mattea Welch, Leonard Wee, David Jaffray, Andre Dekker, Andrew Hope

^*Corresponding author for this work

Research output: Contribution to journal › Article › Academic › peer-review

19 Citations (Web of Science)

Abstract

Purpose: Highlighting the risk of biases in radiomics-based models will help improve their quality and increase usage as decision support systems in the clinic. In this study we use machine learning-based methods to identify the presence of volume-confounding effects in radiomics features.

Methods

841 radiomics features were extracted from two retrospective publicly available datasets of lung and head neck cancers using open source software. Unsupervised hierarchical clustering and principal component analysis (PCA) identified relations between radiomics and clinical outcomes (overall survival). Bootstrapping techniques with logistic regression verified features' prognostic power and robustness.

Results

Over 80% of the features had large pairwise correlations. Nearly 30% of the features presented strong correlations with tumor volume. Using volume-independent features for clustering and PCA did not allow risk stratification of patients. Clinical predictors outperformed radiomics features in bootstrapping and logistic regression.

Conclusions

The adoption of safeguards in radiomics is imperative to improve the quality of radiomics studies. We proposed machine learning (ML) - based methods for robust radiomics signatures development.

Original language	English
Pages (from-to)	24-30
Number of pages	7
Journal	Physica Medica: European journal of medical physics
Volume	71
DOIs	https://doi.org/10.1016/j.ejmp.2020.02.010
Publication status	Published - Mar 2020
Event	International Conference on the Use of Computers in Radiation Therapy (ICCR) / International Conference on Monte Carlo Techniques for Medical Applications (MCMA) - Montreal, Canada Duration: 17 Jun 2019 → 21 Jun 2019

Keywords

Radiomics
Machine learning
Predictions
Lung
Head and neck
TUMOR VOLUME
HETEROGENEITY

Access to Document

10.1016/j.ejmp.2020.02.010Licence: CC BY-NC-ND

Cite this

@article{a0f91f51324d4ebd817a3a09755c208e,

title = "Machine learning helps identifying volume-confounding effects in radiomics",

abstract = "Purpose: Highlighting the risk of biases in radiomics-based models will help improve their quality and increase usage as decision support systems in the clinic. In this study we use machine learning-based methods to identify the presence of volume-confounding effects in radiomics features.Methods841 radiomics features were extracted from two retrospective publicly available datasets of lung and head neck cancers using open source software. Unsupervised hierarchical clustering and principal component analysis (PCA) identified relations between radiomics and clinical outcomes (overall survival). Bootstrapping techniques with logistic regression verified features' prognostic power and robustness.ResultsOver 80% of the features had large pairwise correlations. Nearly 30% of the features presented strong correlations with tumor volume. Using volume-independent features for clustering and PCA did not allow risk stratification of patients. Clinical predictors outperformed radiomics features in bootstrapping and logistic regression.ConclusionsThe adoption of safeguards in radiomics is imperative to improve the quality of radiomics studies. We proposed machine learning (ML) - based methods for robust radiomics signatures development.",

keywords = "Radiomics, Machine learning, Predictions, Lung, Head and neck, TUMOR VOLUME, HETEROGENEITY",

author = "Alberto Traverso and Michal Kazmierski and Ivan Zhovannik and Mattea Welch and Leonard Wee and David Jaffray and Andre Dekker and Andrew Hope",

year = "2020",

month = mar,

doi = "10.1016/j.ejmp.2020.02.010",

language = "English",

volume = "71",

pages = "24--30",

journal = "Physica Medica: European journal of medical physics",

issn = "1120-1797",

publisher = "ELSEVIER SCI LTD",

note = "International Conference on the Use of Computers in Radiation Therapy (ICCR) / International Conference on Monte Carlo Techniques for Medical Applications (MCMA) ; Conference date: 17-06-2019 Through 21-06-2019",

}

TY - JOUR

T1 - Machine learning helps identifying volume-confounding effects in radiomics

AU - Traverso, Alberto

AU - Kazmierski, Michal

AU - Zhovannik, Ivan

AU - Welch, Mattea

AU - Wee, Leonard

AU - Jaffray, David

AU - Dekker, Andre

AU - Hope, Andrew

PY - 2020/3

Y1 - 2020/3

N2 - Purpose: Highlighting the risk of biases in radiomics-based models will help improve their quality and increase usage as decision support systems in the clinic. In this study we use machine learning-based methods to identify the presence of volume-confounding effects in radiomics features.Methods841 radiomics features were extracted from two retrospective publicly available datasets of lung and head neck cancers using open source software. Unsupervised hierarchical clustering and principal component analysis (PCA) identified relations between radiomics and clinical outcomes (overall survival). Bootstrapping techniques with logistic regression verified features' prognostic power and robustness.ResultsOver 80% of the features had large pairwise correlations. Nearly 30% of the features presented strong correlations with tumor volume. Using volume-independent features for clustering and PCA did not allow risk stratification of patients. Clinical predictors outperformed radiomics features in bootstrapping and logistic regression.ConclusionsThe adoption of safeguards in radiomics is imperative to improve the quality of radiomics studies. We proposed machine learning (ML) - based methods for robust radiomics signatures development.

AB - Purpose: Highlighting the risk of biases in radiomics-based models will help improve their quality and increase usage as decision support systems in the clinic. In this study we use machine learning-based methods to identify the presence of volume-confounding effects in radiomics features.Methods841 radiomics features were extracted from two retrospective publicly available datasets of lung and head neck cancers using open source software. Unsupervised hierarchical clustering and principal component analysis (PCA) identified relations between radiomics and clinical outcomes (overall survival). Bootstrapping techniques with logistic regression verified features' prognostic power and robustness.ResultsOver 80% of the features had large pairwise correlations. Nearly 30% of the features presented strong correlations with tumor volume. Using volume-independent features for clustering and PCA did not allow risk stratification of patients. Clinical predictors outperformed radiomics features in bootstrapping and logistic regression.ConclusionsThe adoption of safeguards in radiomics is imperative to improve the quality of radiomics studies. We proposed machine learning (ML) - based methods for robust radiomics signatures development.

KW - Radiomics

KW - Machine learning

KW - Predictions

KW - Lung

KW - Head and neck

KW - TUMOR VOLUME

KW - HETEROGENEITY

U2 - 10.1016/j.ejmp.2020.02.010

DO - 10.1016/j.ejmp.2020.02.010

M3 - Article

C2 - 32088562

VL - 71

SP - 24

EP - 30

JO - Physica Medica: European journal of medical physics

JF - Physica Medica: European journal of medical physics

SN - 1120-1797

T2 - International Conference on the Use of Computers in Radiation Therapy (ICCR) / International Conference on Monte Carlo Techniques for Medical Applications (MCMA)

Y2 - 17 June 2019 through 21 June 2019

ER -