Can Radiomics of Dynamic PET Imaging with 11C-methionine Predict EGFR Amplification Status in Glioblastoma?

Authors

  • Gleb DANILOV Laboratory of Biomedical Informatics and Artificial Intelligence, National Medical Research Center for Neurosurgery named after N.N. Burdenko, 4th Tverskaya-Yamskaya Str. 16, Moscow, Russian Federation
  • Andrey POSTNOV Department of Neuroimaging, National Medical Research Center for Neurosurgery named after N.N. Burdenko, 4th Tverskaya-Yamskaya Str. 16, Moscow, Russian Federation
  • Diana KALAEVA Department of Neuroimaging, National Medical Research Center for Neurosurgery named after N.N. Burdenko, 4th Tverskaya-Yamskaya Str. 16, Moscow, Russian Federation
  • Nina VIKHROVA Department of Neuroimaging, National Medical Research Center for Neurosurgery named after N.N. Burdenko, 4th Tverskaya-Yamskaya Str. 16, Moscow, Russian Federation
  • Tatyana KOBYAKOVA Department of Neuroimaging, National Medical Research Center for Neurosurgery named after N.N. Burdenko, 4th Tverskaya-Yamskaya Str. 16, Moscow, Russian Federation

Keywords:

Glioblastoma, Radiomics, Positron Emission Tomography Computed Tomography, Artificial Intelligence, Epidermal growth factor receptor

Abstract

Introduction: Epidermal growth factor receptor (EGFR) amplification predicts poor survival in patients with brain gliomas. Purpose: This study aimed to evaluate whether EGFR amplification status can be predicted using radiomics data from dynamic PET scanning with 11C-methionine. Materials and Methods: We analyzed 31 PET/CT scans from 31 patients (7 men 22.6% and 24 women 77.4%, mean age 59 ± 10 years). Three datasets were used to predict EGFR amplification status via machine learning: 1) Radiomic features calculated as time series for each image biomarker; 2) Dynamic tumor-to-normal brain ratio (T/N) of radiopharmaceutical uptake - time series of T/N peak for 26 frames; 3) Static T/N - peak, max, and average T/N for static images. Results: Radiomics-based models achieved an average accuracy of 1.0 using k-nearest neighbors across thirty subsampling experiments. Despite this promising result, we approach it critically, considering significant methodological limitations of our study and similar works. These include a small sample size, lack of standardized regions of interest, and absence of reproducibility tests for the selected imaging biomarkers and resulting models. Conclusion: Further research should focus on reproducibility, which is crucial for ensuring the non-randomness, generalizability, and real-world value of our findings.

Downloads

Published

21.11.2024

How to Cite

1.
DANILOV G, POSTNOV A, KALAEVA D, VIKHROVA N, KOBYAKOVA T. Can Radiomics of Dynamic PET Imaging with 11C-methionine Predict EGFR Amplification Status in Glioblastoma?. Appl Med Inform [Internet]. 2024 Nov. 21 [cited 2024 Dec. 3];46(Suppl. 2):S1-S4. Available from: https://ami.info.umfcluj.ro/index.php/AMI/article/view/1072