Reproducibility of radiomic features using network analysis and its application in Wasserstein k-means clustering.

Jung Hun Oh,Joseph O Deasy,Vaios Hatzoglou,Harini Veeraraghavan,Usman Mahmood,Aditya P Apte,Aditi Iyer,Amita Shukla-Dave,Nancy Y Lee,Nadeem Riaz,Evangelia Katsoulakis,Yao Yu,Allen Tannenbaum,Maryam Pouryahya

doi:10.1117/1.jmi.8.3.031904

Abstract

.Purpose: The goal of this study is to develop innovative methods for identifying radiomic features that are reproducible over varying image acquisition settings.Approach: We propose a regularized partial correlation network to identify reliable and reproducible radiomic features. This approach was tested on two radiomic feature sets generated using two different reconstruction methods on computed tomography (CT) scans from a cohort of 47 lung cancer patients. The largest common network component between the two networks was tested on phantom data consisting of five cancer samples. To further investigate whether radiomic features found can identify phenotypes, we propose a -means clustering algorithm coupled with the optimal mass transport theory. This approach following the regularized partial correlation network analysis was tested on CT scans from 77 head and neck squamous cell carcinoma (HNSCC) patients in the Cancer Imaging Archive (TCIA) and validated using an independent dataset.Results: A set of common radiomic features was found in relatively large network components between the resultant two partial correlation networks resulting from a cohort of lung cancer patients. The reliability and reproducibility of those radiomic features were further validated on phantom data using the Wasserstein distance. Further analysis using the network-based Wasserstein -means algorithm on the TCIA HNSCC data showed that the resulting clusters separate tumor subsites as well as HPV status, and this was validated on an independent dataset.Conclusion: We showed that a network-based analysis enables identifying reproducible radiomic features and use of the selected set of features can enhance clustering results.

Highlights

Radiomics enables an in-depth measurement of tumor phenotypes by quantifying imaging signals from radiologic images that can reflect key information of signatures associated with patient outcomes.[1,2] Recently, the close connection of radiomics with machine learning has accelerated the development of new imaging features and radiomic outcomes modeling, showing the potential of using radiomics to build predictive models of individual cancer outcomes.[3,4] Despite such great progress in radiomics in recent years, the development of computational techniques to identify repeatable and reproducible radiomic features remains challenging and relatively unadvanced.[5]
To investigate whether radiomic features found by the regularized partial correlation network analysis can identify phenotypes, we further propose a method, the network-based Wasserstein k-means (NWK) clustering algorithm, to identify subgroups of tumors, employing the Wasserstein distance as a cost function in the conventional k-means algorithm
We demonstrated the potential of network-based approaches to identify reproducible radiomic features on data obtained using two different reconstruction kernels on lung cancer computed tomography (CT) scans

Summary

Introduction

Radiomics enables an in-depth measurement of tumor phenotypes by quantifying imaging signals from radiologic images that can reflect key information of signatures associated with patient outcomes.[1,2] Recently, the close connection of radiomics with machine learning has accelerated the development of new imaging features and radiomic outcomes modeling, showing the potential of using radiomics to build predictive models of individual cancer outcomes.[3,4] Despite such great progress in radiomics in recent years, the development of computational techniques to identify repeatable and reproducible radiomic features remains challenging and relatively unadvanced.[5] This has led to the lack of success of many radiomic models in subsequent external validation on independent data, impairing the reliability of the models.[6,7] One of the reasons for this is likely due to the susceptibility of radiomic features to image reconstruction and acquisition parameters.[8,9] Since radiomic features are computed via multiple tasks including imaging acquisition, segmentation, and feature extraction, the selection of parameters present in each step may affect the stability of features computed.[10] As such, prior to model building, the development of radiomic features with high repeatability and high reproducibility as well as the development of tools that can identify such features is more likely to be urgently needed in the field of radiomics

Objectives

Methods

Results

Discussion

Conclusion