TY - JOUR
T1 - Reproducible radiomics through automated machine learning validated on twelve clinical applications
AU - Starmans, Martijn P. A.
AU - van der Voort, Sebastian R.
AU - Phil, Thomas
AU - Timbergen, Milea J. M.
AU - Vos, Melissa
AU - Padmos, Guillaume A.
AU - Kessels, Wouter
AU - Hanff, David
AU - Grunhagen, Dirk J.
AU - Verhoef, Cornelis
AU - Sleijfer, Stefan
AU - Bent, Martin J. van den
AU - Smits, Marion
AU - Dwarkasing, Roy S.
AU - Els, Christopher J.
AU - Fiduzi, Federico
AU - van Leenders, Geert J. L. H.
AU - Blazevic, Anela
AU - Hofland, Johannes
AU - Brabander, Tessa
AU - van Gils, Renza A. H.
AU - Franssen, Gaston J. H.
AU - Feelders, Richard A.
AU - de Herder, Wouter W.
AU - Buisman, Florian E.
AU - Willemssen, Francois E. J. A.
AU - Koerkamp, Bas Groot
AU - Angus, Lindsay
AU - van der Veldt, Astrid A. M.
AU - Rajicic, Ana
AU - Odink, Arlette E.
AU - Deen, Mitchell
AU - T., Jose M. Castillo
AU - Veenland, Jifke
AU - Schoots, Ivo
AU - Renckens, Michel
AU - Doukas, Michail
AU - de Man, Rob A.
AU - IJzermans, Jan N. M.
AU - Miclea, Razvan L.
AU - Vermeulen, Peter B.
AU - Bron, Esther E.
AU - Thomeer, Maarten G.
AU - Visser, Jacob J.
AU - Niessen, Wiro J.
AU - Klein, Stefan
N1 - 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).
PY - 2021
Y1 - 2021
N2 - Radiomics uses quantitative medical imaging features to predict clinical outcomes. Currently, in a new clinical application, findingthe 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 perapplication. To this end, we formulate radiomics as a modular workflow and include a large collection of common algorithms foreach component. To optimize the workflow per application, we employ automated machine learning using a random search andensembling. 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). Weshow that our framework has a competitive performance compared human experts, outperforms a radiomics baseline, and performssimilar or superior to Bayesian optimization and more advanced ensemble approaches. Concluding, our method fully automaticallyoptimizes 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.
AB - Radiomics uses quantitative medical imaging features to predict clinical outcomes. Currently, in a new clinical application, findingthe 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 perapplication. To this end, we formulate radiomics as a modular workflow and include a large collection of common algorithms foreach component. To optimize the workflow per application, we employ automated machine learning using a random search andensembling. 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). Weshow that our framework has a competitive performance compared human experts, outperforms a radiomics baseline, and performssimilar or superior to Bayesian optimization and more advanced ensemble approaches. Concluding, our method fully automaticallyoptimizes 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.
U2 - 10.48550/ARXIV.2108.08618
DO - 10.48550/ARXIV.2108.08618
M3 - Article
JO - arXiv preprint
JF - arXiv preprint
ER -