Nuklearmedizin 2022; 61(02): 174
DOI: 10.1055/s-0042-1746049
Abstracts | NuklearMedizin 2022
WIS-Vortrag
Radiomics

Ability of F18-FDG PET radiomics and machine learning in predicting KRAS mutation status in therapy-naive lung adenocarcinoma

Authors

  • R. Zhang

    1   Klinikum Bayreuth, Klinik und Institut für Nuklearmedizin, Bayreuth

  • K. Shi

    2   Inselspital Bern, Nuklearmedizin, Bern

  • W. Hohenforst-Schmidt

    3   Sana Klinikum Hof, Pneumologie, Hof

  • C. Steppert

    4   REGIOMED Klinikum Coburg, Pneumologie, Coburg

  • C. Schmidkonz

    5   Universitätsklinikum Erlangen, Nuklearmedizin, Erlangen

  • A. Atzinger

    5   Universitätsklinikum Erlangen, Nuklearmedizin, Erlangen

  • A. Hartmann

    6   Universitätsklinikum Erlangen, Pathologie, Erlangen

  • M. Vieth

    7   Klinikum Bayreuth, Pathologie, Bayreuth

  • S. Förster

    1   Klinikum Bayreuth, Klinik und Institut für Nuklearmedizin, Bayreuth
Further Information
Also available at
 
 

Ziel/Aim Considering the increasing role of KRAS mutation status in the treatment of NSCLC and the limited knowledge about PET radiomic features in predicting KRAS mutation we aimed to build a prediction model by combining F18-FDG-PET radiomics, molecular pathology and machine learning.

Methodik/Methods 119 patients with therapy-naive lung adenocarcinoma and PET/CT were retrospectively selected. Datasets were randomly divided into three subgroups, a training set, a validation set and a testing set. Two open-source softwares, 3D Slicer and Python, were used to segment tumours and extract radiomic features from PET images. Feature selection was performed by Mann-Whitney U test, spearman’s rank correlation coefficient and RFE. Logistic regression was used to build prediction models. Then models were compared by ROC curves and evaluated by calibration plots and DCA curves (1).

Ergebnisse/Results In total, 1781 PET radiomic features were extracted. 163 predictive models were established according to each original feature group and their combinations. After model comparisons the model with the highest predictive abilities including wHLH_fo_IR, wHLH_glrlm_SRHGLE, wHLH_glszm_SAHGLE and smoking history was identified. The model obtained AUCs 0.731 (95% CI: 0.619~0.843), 0.750 (95% CI: 0.248~1.000), and 0.750 (95% CI: 0.448~1.000) in the training-, validation- and testing set, respectively. Calibration indicated no difference between observed and predictive values in the two datasets (P=0.377 and 0.861). DCA curves of our model showed higher performance depending on KRAS mutation rate.[1]

Schlussfolgerungen/Conclusions Our model combining PET radiomics and machine learning showed promising ability to predict KRAS mutation status in therapy-naive lung adenocarcinoma. It might be a helpful clinical tool to non-invasively screen KRAS mutation status in addition to biopsy sampling.



Publication History

Article published online:
14 April 2022

© 2022. Thieme. All rights reserved.

Georg Thieme Verlag KG
Rüdigerstraße 14, 70469 Stuttgart, Germany