Pim de Graaf

Amsterdam UMC, location VUMC

In short, Pim de Graaf's research entails the following: 

'MRI-based deep learning segmentation and quantitative radiomics in retinoblastoma: a next step towards personalized interventions

' In general, considerable progress is made in personalized cancer therapy. However, in Retinoblastoma, a rare cancer that develops in the eyes of young children, individualization of treatment is complicated, since intraocular tumor-biopsy is contra-indicated. The key to personalized treatment is identifying tumor-subtypes that require a tailored approach with the help of detailed tumor characterization and unravelling tumor-heterogeneity. The Dutch Retinoblastoma Center will initiate multicenter studies between specialized centers for Retinoblastoma treatment within Europe to assess the value of MRI for the detection of genetic subtypes of this rare form of paediatric eye cancer and to improve the detection of risk factors for developing distant metastasis (in particular tumor invasion into the optical nerve). To reach these goals deep learning algorithms for automated tumour delineation will be developed. These delineations will subsequently be used for radiomics analyses to characterize and quantify the tumour phenotype.'

Publications

• Automatic segmentation and quantification of the optic nerve on MRI using a 3D U-Net
  
  • Pubmed, May 15, 2023
  • Click here to view the publication
    
    
• MRI Features for Identifying MYCN-amplified RB1 Wild-type Retinoblastoma
  • Pubmed, May 16, 2023
  • Click here to view the publication
    
    
• Optic nerve thickening on high-spatial-resolution MRI predicts early-stage postlaminar optic nerve invasion in retinoblastoma
  • Pubmed, December 13, 2023
  • Click here to view the publication

Brief summary of progress / results December 2023

We have shown that it is possible for both the radiologist and the computer to distinguish between the rarer (and more aggressive form) of retinal cancer and the most common form of retinal cancer. This makes it possible to tailoring patient's treatment even more specifically in the future. In addition, we have shown that the thickness of the optic nerve, measured by a radiologist or computer, provides a good prediction of whether there is ingrowth of the tumor into the optic nerve.

In another study we provided an overview of possible side effects that may occur after selective intra-arterial chemotherapy and that are visible on MRI. This allows radiologists to put findings after such a type of treatment into a better context. In the same study it was shown that eyes treated with selective chemotherapy are smaller than healthy eyes based on automatic volume measurements made by the computer. In addition, work has been done on the implementation of a computer algorithm that can automatically detect and delineate the tumor and optic nerves on MRI images from different hospitals. A method has also been developed for automatically measuring the cross-sectional dimensions of these nerves. Initial analysis of this automatically measured optic nerve cross-section shows that it can be used to detect retinal cancer invasion of the optic nerve. This automatically measured cross-section of the optic nerve could eventually also be applied to other tumors of the optic nerve, such as optic nerve glioma, which can be used, for example, to determine more accurately whether a treatment is effective.

Brief summary of progress / results November 2022

After a start-up period, during which hard work was done on the data collection and the obtaining approval for the studies, the first results are coming in. Some of these have already been presented at a major international conference. This one results showed that, using computer-computed features on MRI images, it is possible to distinguish between the classical picture of retinal cancer in children and the more aggressive growing type retinal cancer. In addition, work has started on improving and renewing an automatic segmentation network, whereby different structures of the eye are automatically recognized on MRI images from different hospitals. This network will eventually be able to independently operate certain parts of the eye recognize and tell more about the underlying genetics of the retinal tumors.