Click here to

  1. Home
  2. Meeting Program
  3. All Sessions
  4. Abstracts
Session: Data Science Robustness, Performance, and Data Harmonization [Return to Session]

Need for PET/CT Harmonization in Multi-Center Immunotherapy Radiomics Studies

K Strasek1*, R Jeraj1,2, M Namias3, D Valentinuzzi1, (1) University of Ljubljana, Faculty for Mathematics and Physics, Ljubljana,SI, (2) University of Wisconsin, Madison, WI, (3) Fundacion Centro Diagnostico Nuclear, Buenos Aires, B, AR

Presentations

SU-H430-IePD-F6-2 (Sunday, 7/10/2022) 4:30 PM - 5:00 PM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 6

Purpose: To quantify the importance and need for PET/CT harmonization in multi-center studies evaluating radiomics signature for prediction of efficacy of immunotherapy (iRADIOMICS).

Methods: Baseline [18F]-FDG PET/CT images of 30 non-small cell lung cancer patients imaged using a single PET/CT scanner were included in the study. De-harmonization was simulated by applying a Gaussian filter (FWHM=6mm) to get low-resolution images, and adding Poisson noise to get noisy images. Multiple datasets simulating harmonized and de-harmonized situations were constructed by combining varying proportions of original, low-resolution, and noisy images. From each dataset, six radiomic features, previously shown as being predictive of clinical outcome (iRADIOMICS), were extracted. The effect of de-harmonization was assessed by Mann-Whitney U-test (MWU), Cox proportional hazards regression analysis, and area under the ROC curve (AUC). The predictive value of the iRADIOMICS signature was compared to PD-L1 tumour proportion score (TPS) and iRECIST using ROC curve analysis.

Results: Harmonization mostly affected the predictive value of iRADIOMICS signature, even though not significantly. AUC for the harmonized (original) dataset was 0.89, for a de-harmonized (low-resolution) dataset was 0.87, and for a dataset containing a mix of harmonized and de-harmonized (original and low-resolution) datasets was 0.85. The addition of noisy images did not affect the predictive value of iRADIOMICS, compared to standard evaluation by TPS and iRECIST, which achieved AUC values of 0.60 and 0.79-0.86, respectively.

Conclusion: Multivariate iRADIOMICS remains a promising biomarker even if the scanners are not harmonized; only limited improvement in predictive power is achieved with harmonization.

Funding Support, Disclosures, and Conflict of Interest: Funding Support: ARRS

Keywords

Image Analysis, FDG PET, Feature Extraction

Taxonomy

IM- PET : Radiomics

Contact Email

Share:

Additional Links