Frontiers in Oncology (May 2021)

Targeting Treatment Resistance in Head and Neck Squamous Cell Carcinoma – Proof of Concept for CT Radiomics-Based Identification of Resistant Sub-Volumes

  • Marta Bogowicz,
  • Matea Pavic,
  • Oliver Riesterer,
  • Oliver Riesterer,
  • Tobias Finazzi,
  • Helena Garcia Schüler,
  • Edna Holz-Sapra,
  • Leonie Rudofsky,
  • Lucas Basler,
  • Manon Spaniol,
  • Andreas Ambrusch,
  • Martin Hüllner,
  • Matthias Guckenberger,
  • Stephanie Tanadini-Lang

DOI
https://doi.org/10.3389/fonc.2021.664304
Journal volume & issue
Vol. 11

Abstract

Read online

PurposeRadiomics has already been proposed as a prognostic biomarker in head and neck cancer (HNSCC). However, its predictive power in radiotherapy has not yet been studied. Here, we investigated a local radiomics approach to distinguish between tumor sub-volumes with different levels of radiosensitivity as a possible target for radiation dose intensification.Materials and MethodsOf 40 patients (n=28 training and n=12 validation) with biopsy confirmed locally recurrent HNSCC, pretreatment contrast-enhanced CT images were registered with follow-up PET/CT imaging allowing identification of controlled (GTVcontrol) vs non-controlled (GTVrec) tumor sub-volumes on pretreatment imaging. A bi-regional model was built using radiomic features extracted from pretreatment CT in the GTVrec and GTVcontrol to differentiate between those regions. Additionally, concept of local radiomics was implemented to perform detection task. The original tumor volume was divided into sub-volumes with no prior information on the location of recurrence. Radiomic features from those sub-volumes were then used to detect recurrent sub-volumes using multivariable logistic regression.ResultsRadiomic features extracted from non-controlled regions differed significantly from those in controlled regions (training AUC = 0.79 CI 95% 0.66 - 0.91 and validation AUC = 0.88 CI 95% 0.72 – 1.00). Local radiomics analysis allowed efficient detection of non-controlled sub-volumes both in the training AUC = 0.66 (CI 95% 0.56 – 0.75) and validation cohort 0.70 (CI 95% 0.53 – 0.86), however performance of this model was inferior to bi-regional model. Both models indicated that sub-volumes characterized by higher heterogeneity were linked to tumor recurrence.ConclusionLocal radiomics is able to detect sub-volumes with decreased radiosensitivity, associated with location of tumor recurrence in HNSCC in the pre-treatment CT imaging. This proof of concept study, indicates that local CT radiomics can be used as predictive biomarker in radiotherapy and potential target for dose intensification.

Keywords