Strahlentherapie und Onkologie

, Volume 194, Issue 9, pp 824–834 | Cite as

Treatment-related features improve machine learning prediction of prognosis in soft tissue sarcoma patients

  • Jan C. Peeken
  • Tatyana Goldberg
  • Christoph Knie
  • Basil Komboz
  • Michael Bernhofer
  • Francesco Pasa
  • Kerstin A. Kessel
  • Pouya D. Tafti
  • Burkhard Rost
  • Fridtjof Nüsslin
  • Andreas E. Braun
  • Stephanie E. Combs
Original Article


Background and purpose

Current prognostic models for soft tissue sarcoma (STS) patients are solely based on staging information. Treatment-related data have not been included to date. Including such information, however, could help to improve these models.

Materials and methods

A single-center retrospective cohort of 136 STS patients treated with radiotherapy (RT) was analyzed for patients’ characteristics, staging information, and treatment-related data. Therapeutic imaging studies and pathology reports of neoadjuvantly treated patients were analyzed for signs of response. Random forest machine learning-based models were used to predict patients’ death and disease progression at 2 years. Pre-treatment and treatment models were compared.


The prognostic models achieved high performances. Using treatment features improved the overall performance for all three classification types: prediction of death, and of local and systemic progression (area under the receiver operatoring characteristic curve (AUC) of 0.87, 0.88, and 0.84, respectively). Overall, RT-related features, such as the planning target volume and total dose, had preeminent importance for prognostic performance. Therapy response features were selected for prediction of disease progression.


A machine learning-based prognostic model combining known prognostic factors with treatment- and response-related information showed high accuracy for individualized risk assessment. This model could be used for adjustments of follow-up procedures.


Biomarker Precision medicine Prognostic model Random forest Decision support systems 

Therapieinformationen verbessern auf maschinellem Lernen basierende prognostische Einschätzungen für Patienten mit Weichteilsarkomen


Hintergrund und Zielsetzung

Aktuelle prognostische Modelle für Patienten mit Weichteilsarkomen basieren primär auf Staginginformationen. Therapieinformationen werden dabei nicht berücksichtigt. Die Berücksichtigung solcher Daten könnte die Vorhersage verbessern.

Material und Methoden

Für eine retrospektive, monozentrische, strahlentherapeutisch behandelte Kohorte mit 136 Weichteilsarkompatienten wurden Patientencharakteristika, Staging und therapieassoziierte Daten erhoben. Potenzielle mit dem Therapieansprechen assoziierte Informationen von neoadjuvant behandelten Patienten wurden aus therapeutischen Magnetresonanztomographie(MRT)-Datensätzen und pathologischen Befunden erhoben. Auf Basis dieser Informationen wurden Random-Forest-Modelle für die Vorhersage des 2‑Jahres-Überlebens bzw. des Progresses generiert. Prätherapie- und Therapiemodelle wurden verglichen.


Die prognostischen Modelle zeigten insgesamt eine gute Vorhersagekraft. Die Hinzunahme von Therapieinformationen konnte die Vorhersageeffizienz der 3 Klassifikationen verbessern: Vorhersage des Versterbens sowie des lokalen und systemischen Progresses („area under the receiver operating or characteristic curve“ [AUC] je 0,87, 0,88 und 0,84). Strahlentherapieassoziierte Informationen wie das Planungszielvolumen und die Gesamtdosis hatten einen großen Einfluss auf die Vorhersagekraft. Mit dem Therapieansprechen assoziierte Informationen wurden für die Vorhersage des Progresses selektiert und zeigten so eine mögliche prognostische Bedeutung.


Auf maschinellem Lernen basierende prognostische Modelle zeigten eine hohe Genauigkeit für die Vorhersage des Überlebens und Krankheitsprogresses durch Einschluss von Informationen zur Therapie und zum Therapieansprechen. Diese Modelle könnten für die individuelle Risikoabschätzung in der Nachsorge verwendet werden.


Biomarker Präzisionsmedizin Prognostisches Modell Random Forest Entscheidungsunterstützungssystem 



Deutsches Konsortium für Translationale Krebsforschung (DKTK), Partner Site Munich (to JP, KK, SC), Alexander von Humboldt Foundation through German Federal Ministry for Education and Research, and the Bavarian Competence Network for Technical and Scientific High Performance Computing (to MB and BR), Allianz (to TG, BK, PT and AB).

Compliance with ethical guidelines

Conflict of interest

J.C. Peeken, T. Goldberg, C. Knie, B. Komboz, M. Bernhofer, F. Pasa, K.A. Kessel, P.D. Tafti, B. Rost, F. Nüsslin, A.E. Braun, and S.E. Combs declare that they have no competing interests.

Ethical standards

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the study.

Supplementary material

66_2018_1294_MOESM1_ESM.docx (64 kb)
Supplemental Tables and Supplemental Figure. Table S1: Performance assessment of six random forest-based prediction models. Table S2: Comparison of AUC values on training and test sets of six random forest-based prediction models. Figure S1: Precision/Recall curves for the spectrum of reliability scores for the pre-treatment models.


  1. 1.
    Gutierrez JC, Perez EA, Franceschi D et al (2007) Outcomes for soft-tissue sarcoma in 8249 cases from a Large State Cancer Registry. J Surg Res 141:105–114. CrossRefPubMedGoogle Scholar
  2. 2.
    Zagars GK, Ballo MT, Pisters PWT et al (2003) Prognostic factors for patients with localized soft-tissue sarcoma treated with conservation surgery and radiation therapy: an analysis of 1225 patients. Cancer 97:2530–2543. CrossRefPubMedGoogle Scholar
  3. 3.
    Pisters PW, Leung DH, Woodruff J, Shi W, Brennan MF (1996) Analysis of prognostic factors in 1,041 patients with localized soft tissue sarcomas of the extremities. J Clin Oncol 14:1679–1689. CrossRefPubMedGoogle Scholar
  4. 4.
    Ramanathan RC, A’Hern R, Fisher C, Thomas JM (1999) Modified staging system for extremity soft tissue sarcomas. Ann Surg Oncol 6:57–69CrossRefPubMedGoogle Scholar
  5. 5.
    Maki RG, Moraco N, Antonescu CR et al (2013) Toward better soft tissue sarcoma staging: building on american joint committee on cancer staging systems versions 6 and 7. Ann Surg Oncol 20:3377–3383. CrossRefPubMedPubMedCentralGoogle Scholar
  6. 6.
    Suit HD, Mankin HJ, Wood WC et al (1988) Treatment of the patient with stage M0 soft tissue sarcoma. J Clin Oncol 6:854–862. CrossRefPubMedGoogle Scholar
  7. 7.
    Edge SB, Compton CC (2010) The American Joint Committee on Cancer: the 7th edition of the AJCC cancer staging manual and the future of TNM. Ann Surg Oncol 17:1471–1474. CrossRefPubMedGoogle Scholar
  8. 8.
    Andrä C, Klein A, Dürr HR et al (2017) External-beam radiation therapy combined with limb-sparing surgery in elderly patients (>70 years) with primary soft tissue sarcomas of the extremities. Strahlenther Onkol 193:604–611. CrossRefPubMedGoogle Scholar
  9. 9.
    Kattan MW, Leung DHY, Brennan MF (2002) Postoperative nomogram for 12-year sarcoma-specific death. J Clin Oncol 20:791–796. CrossRefPubMedGoogle Scholar
  10. 10.
    Eilber FC, Brennan MF, Eilber FR et al (2004) Validation of the postoperative nomogram for 12-year sarcoma-specific mortality. Cancer 101:2270–2275. CrossRefPubMedGoogle Scholar
  11. 11.
    Amin MB, Edge S, Greene F et al (eds) (2017) AJCC cancer staging manual, 8th edn. Springer, ChamGoogle Scholar
  12. 12.
    Tufman A, Kahnert K, Kauffmann-Guerrero D et al (2017) Clinical relevance of the M1b and M1c descriptors from the proposed TNM 8 classification of lung cancer. Strahlenther Onkol 193:392–401. CrossRefPubMedGoogle Scholar
  13. 13.
    Abernethy AP, Etheredge LM, Ganz PA et al (2010) Rapid-learning system for cancer care. J Clin Oncol 28:4268–4274. CrossRefPubMedPubMedCentralGoogle Scholar
  14. 14.
    Lambin P, van Stiphout RGPM, Starmans MHW et al (2013) Predicting outcomes in radiation oncology–multifactorial decision support systems. Nat Rev Clin Oncol 10:27–40. CrossRefPubMedGoogle Scholar
  15. 15.
    Wardelmann E, Haas RL, Bovée JVMG et al (2016) Evaluation of response after neoadjuvant treatment in soft tissue sarcomas; the European Organization for Research and Treatment of Cancer-Soft Tissue and Bone Sarcoma Group (EORTC-STBSG) recommendations for pathological examination and reporting. Eur J Cancer 53(021):84–95. CrossRefPubMedGoogle Scholar
  16. 16.
    Breiman L (2001) Random forests. Mach Learn 45(1):5–32. CrossRefGoogle Scholar
  17. 17.
    Zimmer VA, Glocker B, Hahner N et al (2017) Learning and combining image neighborhoods using random forests for neonatal brain disease classification. Med Image Anal 42:189–199. CrossRefPubMedGoogle Scholar
  18. 18.
    Chen T, Cao Y, Zhang Y et al (2013) Random forest in clinical metabolomics for phenotypic discrimination and biomarker selection. Evid Based Complement Alternat Med. CrossRefPubMedPubMedCentralGoogle Scholar
  19. 19.
    Liu M, Xu X, Tao Y, Wang X (2017) An improved random forest method based on RELIEFF for medical diagnosis. IEEE International Conference on Computational Science and Engineering (CSE) and IEEE International Conference on Embedded and Ubiquitous Computing (EUC), pp 44–49Google Scholar
  20. 20.
    Kotti M, Duffell LD, Faisal AA, McGregor AH (2017) Detecting knee osteoarthritis and its discriminating parameters using random forests. Med Eng Phys 43:19–29. CrossRefPubMedPubMedCentralGoogle Scholar
  21. 21.
    Rastgoo M, Lemaître G, More O et al (2016) Classification of melanoma lesions using sparse coded features and random forests. Proceedings Volume 9785, Medical Imaging 2016: Computer-Aided Diagnosis; 97850 C. CrossRefGoogle Scholar
  22. 22.
    Statnikov A, Aliferis CF, Tsamardinos I, Hardin D, Levy S (2005) A comprehensive evaluation of multicategory classification methods for microarray gene expression cancer diagnosis. Bioinformatics 21:631–643. CrossRefPubMedGoogle Scholar
  23. 23.
    Eibe F, Hal MA, Ian H (2016) Witten the WEKA workbench. Online appendix for “data mining: practical machine learning tools and techniques”. Morgan Kaufmann, Burlington.Google Scholar
  24. 24.
    Vallières M, Freeman CR, Skamene SR, El Naqa I (2015) A radiomics model from joint FDG-PET and MRI texture features for the prediction of lung metastases in soft-tissue sarcomas of the extremities. Phys Med Biol 60:5471–5496. CrossRefPubMedGoogle Scholar
  25. 25.
    Abeshouse A, Adebamowo C, Adebamowo SN, Akbani R, Akeredolu T, Ally A et al (2017) Comprehensive and integrated genomic characterization of adult soft tissue sarcomas. Cell 171(8):950–965.e2. CrossRefGoogle Scholar
  26. 26.
    Sica GT (2006) Bias in research studies. Radiology 238:780–789. CrossRefPubMedGoogle Scholar
  27. 27.
    Wulfkuhle JD, Liotta LA, Petricoin EF (2003) Proteomic applications for the early detection of cancer. Nat Rev Cancer 3:267–275. CrossRefPubMedGoogle Scholar
  28. 28.
    Meyerson M, Gabriel S, Getz G (2010) Advances in understanding cancer genomes through second-generation sequencing. Nat Rev Genet 11:685–696. CrossRefPubMedGoogle Scholar
  29. 29.
    Vallieres M, Kumar A, Sultanem K, El Naqa I (2013) FDG-PET image-derived features can determine HPV status in head-and-neck cancer. Int J Radiat Oncol Biol Phys 87:467. CrossRefGoogle Scholar
  30. 30.
    Deist TM, Jochems A, van Soest J et al (2017) Infrastructure and distributed learning methodology for privacy-preserving multi-centric rapid learning health care: euroCAT. Clin Transl Radiat Oncol 4:24–31. CrossRefPubMedPubMedCentralGoogle Scholar
  31. 31.
    Peeken JC, Nüsslin F, Combs SE (2017) Radio-oncomics. Strahlenther Onkol. PubMedCrossRefGoogle Scholar
  32. 32.
    Wichmann H, Güttler A, Bache M et al (2014) Inverse prognostic impact of ErbB2 mRNA and protein expression level in tumors of soft tissue sarcoma patients. Strahlenther Onkol 190:912–918. CrossRefPubMedGoogle Scholar

Copyright information

© Springer-Verlag GmbH Germany, part of Springer Nature 2018

Authors and Affiliations

  • Jan C. Peeken
    • 1
    • 7
  • Tatyana Goldberg
    • 2
  • Christoph Knie
    • 1
  • Basil Komboz
    • 2
  • Michael Bernhofer
    • 3
  • Francesco Pasa
    • 4
    • 5
  • Kerstin A. Kessel
    • 1
    • 6
    • 7
  • Pouya D. Tafti
    • 2
  • Burkhard Rost
    • 3
  • Fridtjof Nüsslin
    • 1
  • Andreas E. Braun
    • 2
  • Stephanie E. Combs
    • 1
    • 6
    • 7
  1. 1.Department of Radiation Oncology, Klinikum rechts der IsarTechnical University of Munich (TUM)MunichGermany
  2. 2.Allianz SEMunichGermany
  3. 3.Department for Bioinformatics and Computational Biology, Informatik 12Technical University of Munich (TUM)GarchingGermany
  4. 4.Department of Computer Science, Informatik 9Technical University of Munich (TUM)GarchingGermany
  5. 5.Chair of Biomedical Physics, Department of PhysicsTechnical University of Munich (TUM)GarchingGermany
  6. 6.Institute of Innovative Radiotherapy (iRT), Department of Radiation Sciences (DRS)Helmholtz Zentrum MünchenNeuherbergGermany
  7. 7.Partner Site MunichDeutsches Konsortium für Translationale Krebsforschung (DKTK)MunichGermany

Personalised recommendations