Poster Presentation 36th TROG Cancer Research Annual Scientific Meeting 2024

Computational analysis and artificial intelligence approaches for the FET PET in Glioblastoma (FIG) Trial - TROG 18.06 (#101)

Nathaniel Barry 1 2 , Eng-Siew Koh 3 4 , Pejman Rowshanfarzad 1 2 , Martin A Ebert 1 2 5 , Andrew M Scott 6 7 8 9
  1. University of Western Australia, Crawley, WA, Australia
  2. Centre for Advanced Technologies in Cancer Research (CATCR), Perth, WA, Australia
  3. Department of Radiation Oncology, Liverpool and Macarthur Cancer Therapy Centres, Liverpool, NSW, Australia
  4. South Western Clinical School, University of New South Wales, Sydney, NSW, Australia
  5. Department of Radiation Oncology, Sir Charles Gairdner Hospital, Nedlands, WA, Australia
  6. Olivia Newton-John Cancer Research Intitute, Heidelberg, VIC, Australia
  7. Department of School of Cancer Medicine, La Trobe University, Heidelberg, VIC, Australia
  8. Department of Molecular Imaging and Therapy, Austin Health, Heidelberg, VIC, Australia
  9. Faculty of Medicine, University of Melbourne, Parkville, VIC, Australia

Background

The [18F]-fluoroethyl-L-tyrosine (FET) PET in Glioblastoma (FIG) Study is an Australian, prospective, multicenter trial evaluating the impact of FET PET on the treatment management of glioblastoma patients. Up to 210 adult glioblastoma patients will undergo FET PET pre-radiochemotherapy, one-month post-radiochemotherapy, and at suspected progression. Detailed analysis of static and dynamic FET PET images and comparison to clinical outcomes is a key objective of the trial, and computational modelling approaches are part of this approach.

Aims

To develop an automatic FET PET segmentation network to enhance computational analysis of the FIG study image datasets.

Methods

Although we have experience in quantitative analysis of FET PET, to understand the current framework of artificial intelligence approaches, we have also conducted a review of the literature to gauge its applicability to the FIG study. The methodologies employed in these studies will be incorporated into network and model design, highlighting what tools may be developed, to create efficient workflows to analyse the resulting trial data.

Results

Development of a multilabel automatic segmentation solution for FET PET images will be presented, taking advantage of state-of-the-art, intuitive networks. Our approach will conform with published guidelines when contouring a biological tumour volume, noting the emerging relevance of neural networks to final workflows. The automatically generated structures may then be used to assess treatment response by change in quantitative metrics. In addition, radiomics and AI models can follow once a network is established, which we will explore in the FIG study, using static and dynamic image data.

Conclusion

The data obtained from the ongoing FIG trial will present a unique opportunity to use computational analysis and artificial intelligence to develop automatic segmentation networks and prognostic models for FET PET studies.