Researchers from the Biomedical Data Science Lab (BDSLab) at the ITACA Institute of the Universitat Politècnica de València have led an international study that developed a methodology to identify, using conventional magnetic resonance imaging (MRI), areas around the tumor with features compatible with possible tumor infiltration.

The study, published in the scientific journal Neuro-Oncology, was validated using data from 940 patients across four multicentre cohorts.

“Our aim was to improve the analysis of the areas surrounding the tumour that do not show clear enhancement on contrast-enhanced scans. These regions lie beyond the visible contrast-enhancing tumour core and may play an important role in disease progression and recurrence”, says María Gómez-Mahiques, lead author of the study and a researcher at BDSLab.

Glioblastoma is one of the most aggressive brain tumours and conventional MRI scans cannot accurately distinguish between vasogenic edema and areas of tumor infiltration

Glioblastoma is one of the most aggressive brain tumours. Although surgery aims to remove as much tumor tissue as possible, conventional MRI scans cannot accurately distinguish between vasogenic edema -the accumulation of fluid around the tumor- and areas of tumor infiltration.

Artificial intelligence to analyse the tumour environment

The methodology combines a deep-learning model, which automatically delineates the tumour, necrosis and oedema, with an unsupervised algorithm that analyses differences observed in two commonly used MRI sequences: T2 and FLAIR.

The tool divides the area initially appearing as edema into two regions: one compatible with vasogenic edema and another with imaging features associated with a possible non-contrast-enhancing tumour

This allows the tool to divide the area initially appearing as edema into two regions: one compatible with vasogenic edema and another with imaging features associated with a possible non-contrast-enhancing tumour.

“The main advance is that it enables objective and reproducible analysis of a region that had previously been very difficult to delineate, using MRI scans routinely available in hospitals,” says the ITACA researcher.

The researchers have developed the T2/FLAIR Heterogeneity Index (TFHI), a new biomarker that quantifies the proportion of edema

Based on this analysis, the researchers have developed the T2/FLAIR Heterogeneity Index (TFHI), a new biomarker that quantifies the proportion of edema exhibiting features compatible with possible tumor infiltration.

“The method detected significant differences between the two regions in 97.5% of the patients analysed. In one of the cohorts, exceeding the established threshold was associated with approximately twice the risk of death and a 98-day difference in median survival,” says Elies Fuster-García, a BDSLab-ITACA researcher and study supervisor.

Towards more personalised neuro-oncology

From a clinical perspective, the methodology provides a quantitative and reproducible framework to support planning for supramaximal resection of glioblastoma. Guidelines from the RANO group recommend this type of surgery — which extends beyond the contrast-enhancing tumour — but do not define in a standardised manner which regions should be included.

The long-term aim is to move towards more personalised neuro-oncology by tailoring treatment strategies to the specific characteristics of each patient’s tumour

This tool makes it possible to objectively map areas with features compatible with tumour infiltration, providing neurosurgeons with additional information to define the resection margin and balance oncological benefit against preserving the patient’s function.

“The long-term aim is to move towards more personalised neuro-oncology by tailoring treatment strategies to the specific characteristics of each patient’s tumour,” says Juan M. García-Gómez, principal investigator of BDSLab at the ITACA Institute-UPV.

The tool has been integrated into the ONCOhabitats platform and is openly available for research purposes. At present, its results do not yet translate into direct clinical application, but they provide a new quantitative framework for studying the extent and heterogeneity of glioblastoma.

Reference

Maria Gómez-Mahiques, Carles Lopez-Mateu, F Javier Gil-Terrón, Victor Montosa-I-Micó, Siri Fløgstad Svensson, Eduardo Erasmo Mendoza Mireles, Einar Osland Vik-Mo, Kyrre E Emblem, Carme Balañà-Quintero, Josep Puig, Cristina Alenda, Elena Martinez-Saez, Fran Martínez-Ricarte, Marta Quirós-Martí, Vicent Quilis-Quesada, Juan M García-Gómez, Elies Fuster-Garcia. Automated delineation of putative non-contrast-enhancing tumor in glioblastoma: Prognostic insights. Neuro-Oncology.

DOI: https://academic.oup.com/neuro-oncology/article/28/6/1571/8497911?login=true