Reproducible radiomics through automated machine learning validated on twelve clinical applications

Martijn P. A. Starmans, Sebastian R. van der Voort, Thomas Phil, Milea J. M. Timbergen, Melissa Vos, Guillaume A. Padmos, Wouter Kessels, David Hanff, Dirk J. Grunhagen, Cornelis Verhoef, Stefan Sleijfer, Martin J. van den Bent, Marion Smits, Roy S. Dwarkasing, Christopher J. Els, Federico Fiduzi, Geert J. L. H. van Leenders, Anela Blazevic, Johannes Hofland, Tessa BrabanderRenza A. H. van Gils, Gaston J. H. Franssen, Richard A. Feelders, Wouter W. de Herder, Florian E. Buisman, Francois E. J. A. Willemssen, Bas Groot Koerkamp, Lindsay Angus, Astrid A. M. van der Veldt, Ana Rajicic, Arlette E. Odink, Mitchell Deen, Jose M. Castillo T., Jifke Veenland, Ivo Schoots, Michel Renckens, Michail Doukas, Rob A. de Man, Jan N. M. IJzermans, Razvan L. Miclea, Peter B. Vermeulen, Esther E. Bron, Maarten G. Thomeer, Jacob J. Visser, Wiro J. Niessen, Stefan Klein

Research output: Contribution to journalArticleAcademic

90 Downloads (Pure)

Abstract

Radiomics uses quantitative medical imaging features to predict clinical outcomes. Currently, in a new clinical application, finding
the optimal radiomics method out of the wide range of available options has to be done manually through a heuristic trial-anderror process. In this study we propose a framework for automatically optimizing the construction of radiomics workflows per
application. To this end, we formulate radiomics as a modular workflow and include a large collection of common algorithms for
each component. To optimize the workflow per application, we employ automated machine learning using a random search and
ensembling. We evaluate our method in twelve different clinical applications, resulting in the following area under the curves: 1)
liposarcoma (0.83); 2) desmoid-type fibromatosis (0.82); 3) primary liver tumors (0.80); 4) gastrointestinal stromal tumors (0.77);
5) colorectal liver metastases (0.61); 6) melanoma metastases (0.45); 7) hepatocellular carcinoma (0.75); 8) mesenteric fibrosis
(0.80); 9) prostate cancer (0.72); 10) glioma (0.71); 11) Alzheimer’s disease (0.87); and 12) head and neck cancer (0.84). We
show that our framework has a competitive performance compared human experts, outperforms a radiomics baseline, and performs
similar or superior to Bayesian optimization and more advanced ensemble approaches. Concluding, our method fully automatically
optimizes the construction of radiomics workflows, thereby streamlining the search for radiomics biomarkers in new applications.
To facilitate reproducibility and future research, we publicly release six datasets, the software implementation of our framework,
and the code to reproduce this study.
Original languageEnglish
JournalMedical Image Analysis
DOIs
Publication statusPublished - 2021

Bibliographical note

Funding
Martijn P. A. Starmans and Jose M. Castillo T. acknowledge funding from the research program STRaTeGy with project numbers 14929, 14930, and 14932, which is (partly) financed by the Netherlands Organization for Scientific Research (NWO). Sebastian R. van der Voort acknowledges funding from the Dutch Cancer Society (KWF project number EMCR
2015-7859). Part of this study was financed by the Stichting Coolsingel (reference number 567), a Dutch non-profit foundation. This study is supported by EuCanShare and EuCanImage
(European Union’s Horizon 2020 research and innovation programme under grant agreement Nr. 825903 and Nr. 952103, respectively).

Fingerprint

Dive into the research topics of 'Reproducible radiomics through automated machine learning validated on twelve clinical applications'. Together they form a unique fingerprint.

Cite this