A novel AI-based, non-invasive diagnostic tool has been developed that enables accurate brain tumour diagnosis.
The tool, named DISCERN, uses deep learning that leverages information of magnetic resonance imaging and facilitates brain tumour classification.
The Diagnosis in Susceptibility Contrast Enhancing Regions for Neuroncology (DISCERN) tool has been trained to differentiate between the three most common brain malignancies, and has demonstrated 78% accuracy in the classification of these tumours.
Made to be open access, DISCERN is based on the training of patterns using AI models from information of standard magnetic resonance imaging (MRI).
Developed by the Vall d’Hebron Institute of Oncology’s (VHIO) Radiomics Group in collaboration with researchers of the Neuroradiology Unit at the Bellvitge University Hospital (HUB), results of the DISCERN study have been published in Cell Reports Medicine.
Results of a the study demonstrate the feasibility and accuracy of DISCERN as an enabler of accurate brain tumor diagnosis from perfusion MRI, outperforming conventional methods.
Diagnosing brain tumours
A definitive diagnosis of brain tumours requires neurosurgical interventions that compromise the quality of life of patients.
Raquel Perez-Lopez, Head of VHIO’s Radiomics Group and corresponding author of the study, stated: “The non-invasive differential diagnosis of brain tumors is currently based on the assessment of magnetic resonance imaging before and after contrast administration.
“However, a definitive diagnosis often requires neurosurgical interventions that compromise the quality of life of patients.”
“This work is the result of more than five years’ research focused on the identification of innovative magnetic resonance perfusion imaging biomarkers to enable the differential diagnosis of brain tumors,” added Albert Pons-Escoda, a Clinical Neuroradiologist and Investigator of the Neuroradiology Unit at the Bellvitge University Hospital, study co-author.
“This present study integrates insights generated by other previous research projects on artificial intelligence, resulting in the development of software that automates presurgical diagnostic classification with very good precision, while facilitating its clinical applicability with a user-friendly interface for clinicians.”
This novel deep learning tool leverages the full spatial and temporal information of conventional MRI to identify behavioral patterns on imaging specific to each tumor.
“Deep learning teaches the machine the characteristics of each tumor detected by magnetic resonance imaging of already diagnosed patients. For example, if we show the machine thousands of images of dogs and cats, it will learn the distinct and defining characteristics of both species, and when it sees a new image, it can differentiate between the two,” explained Alonso García-Ruiz, Ph.D. Student of VHIO’s Radiomics Group and first author.
In this case, the learning units are voxels which represent the minimum measurement of volume to study MRI scans and are equivalent to pixels, but three-dimensional display elements. The investigators trained DISCERN to learn the characteristics of three of the most common brain malignancies on 50,000 voxels from 40 diagnosed patients.
According to the researchers, the results of the study demonstrated 78% accuracy in the classification of the tumours – outperforming conventional diagnostic methods. The team say this helps guide medical decision making regarding the need for and type of surgery required to confirm diagnosis.
To enhance study reproducibility and accelerate its adoption in clinical studies, the proposed method has been implemented on the user-friendly, open access DISCERN application. For demonstrative purposes, the DISCERN app can be accessed at http://84.88.64.102:5000/discern-app for research purposes.
