Extra\c{c}\~ao e Classifica\c{c}\~ao de Caracter\'isticas Radi\^omicas em Gliomas de Baixo Grau para An\'alise da Codele\c{c}\~ao 1p/19q

TL;DR

This study uses radiomics and machine learning to predict 1p/19q chromosomal deletion in low-grade gliomas from MRI images, potentially reducing the need for invasive biopsies.

Contribution

It introduces a radiomics-based approach combined with neural networks for non-invasive prediction of chromosomal deletion in gliomas.

Findings

High accuracy in predicting 1p/19q deletion

Effective dimensionality reduction with PCA

Potential to replace surgical biopsies

Abstract

Radiomics is an emerging area, which presents a large set of computational methods and techniques to extract quantitative characteristics from magnetic resonance images. In the feature extraction stage, its outputs must be well defined and carefully evaluated, to provide imaging diagnostics, prognoses and responses to treatment therapies. In this study, we present the extraction of quantitative characteristics from magnetic resonance images in low-grade gliomas using the Pyradiomics library and, using a multilayer perceptron neural network, we will show the prediction of the deletion of the 1p / 19q chromosomes in these gliomas. Several studies show that 1p / 19q chromosomal codelection is a positive prognostic factor in low-grade gliomas, as they are more sensitive to chemotherapy. Due to the large number of extracted characteristics, it was necessary to use a dimensionality reduction technique, the analysis principal components, which proved to be efficient in this study. After training and testing the characteristics performed by the multilayer perceptron neural network, the results showed to be very promising in detecting the deletion status of chromosomes 1p / 19q, mainly taking into account the possibility of avoiding surgical biopsies for this diagnosis.