Radiomics and machine learning based non-invasive biomarkers for prediction of peritumoral invasion in brain metastases
- Reza Forghani, McGill University
Membres de l'équipe :
- Marie-Christine Guiot, Montreal Neurological Hospital and Institute
- Stephanie Lam, Montreal Neurological Institute
- Laurent Létourneau-Guillon, Centre hospitalier de l'université de Montréal(CHUM)
- Kevin Petrecca, McGill University
- Peter Siegel, McGill University
- Canadian Cancer Society
- Canadian Institutes of Health Research
Aperçu du projet
Need for project: Treatment of brain metastases (BrM) includes a combination of surgery and radiation therapy (RT). Recurrence after treatment is a problem that limits survival and quality of life. Our team has discovered that highly invasive growth of BrM is associated with recurrence and sensitivity to particular systemic treatments. However, the only way to identify the invasion pattern in BrM is with histopathology, highlighting a need for non-invasive approaches to characterize peritumoral invasion in BrM.
Goal of project: We will conduct advanced image analysis and machine learning (ML), with histopathological assessment, to identify quantitative features from images as non-invasive biomarkers of peritumoral invasion in BrM. Our goal is to develop a methodology to non-invasively assess invasion in BrM, improving planning of surgical resections and RT. This tool may also lead to stratification approaches for BrM patients treated with systemic therapies based on invasion status.
Project description: We will use pre-operative MRI images with their corresponding histopathological invasion status, to extract radiomics features from MRI images. Then, machine learning models will be used to select features that are the most discriminative biomarkers for non-invasively predicting peritumoral invasion in brain metastases, which we have shown is associated with pertinent patient outcomes such as local recurrence and shortened overall survival.
Future impact: Our data suggests that invasive cancer cells remain overlooked in BrM, driving local treatment failure and shortened overall survival. This project will allow us to develop non-invasive methods to identify BrM lesions with peritumoral invasion. Such capabilities can result in improved local control in patients with surgically resected BrM via tailored surgical and adjuvant RT treatment plans, better outcomes for non-surgical BrM patients treated with stereotactic radiosurgery (SRS), and the ability to define patient subgroups that can benefit from targeted systemic treatments based upon invasion status.