Early detection and novel stratification of pediatric low-grade gliomas by MRI-based artificial intelligence
- Farzad Khalvati, The Hospital for Sick Children
Membres de l'équipe :
- Eric Bouffet, The Hospital for Sick Children
- Cynthia Hawkins, The Hospital For Sick Children
- Birgit Ertl-Wagner, The Hospital for Sick Children
- Uri Tabori, The Hospital For Sick Children
- Canadian Cancer Society
- Canadian Institutes of Health Research
Aperçu du projet
Need for project: Pediatric low-grade glioma (pLGG) is the most common pediatric brain tumour and affected children suffer from multiple recurrences. Recently, genetic markers of pLGG have shown potential for improved diagnosis and prognosis. The genetic profile of pLGG is assessed through biopsy. This invasive procedure reflects only a small part of tumour and it takes a long time to perform. There is an urgent need for noninvasive and fast methods to assess the genetic markers of pLGG in its entirety.
Goal of project: We will develop novel Artificial Intelligence (AI) and deep learning (DL) solutions based on magnetic resonance imaging (MRI) to determine genetic markers of pLGG with high accuracy compared to the analysis of biopsy-obtained tissue. We will use several routinely acquired MRI sequences (T1w, T2w, DWI, FLAIR, and contrast-enhanced T1w) of patients with pLGG treated at SickKids Hospital between 2000 and 2018 to train and validate the proposed DL models for different pLGG genetic markers.
Project description: We will develop MRI-based AI solutions to determine molecular markers of pLGG with high accuracy and no need for manual annotations of the tumours in MRI. We will design a method to provide the necessary confidence for AI results to be used in clinical decision-making. We will test the potential of our AI solutions to be deployed in clinical settings by evaluating their efficacy in predicting pLGG molecular markers based on initial MRIs (2 years before surgery) and predicting therapy response.
Future impact: MRI-based AI solutions that accurately predict genetic markers of pLGG will pave the way toward reducing and eventually eliminating biopsies, which are both invasive and spatially limited. MRI-based AI solutions could assist in early diagnosis, prognostication and determining individualized treatment strategies and may also provide insight into pLGG tumour clonal and mutational evolution. In addition, providing quantitative metrics for the reliability and confidence of the AI results will be an important step for deploying AI models in the clinical workflow allowing the usage of AI results with confidence in real-world settings.