Selection of Dataframes Presenting Glioma from Magnetic Resonance Images: a Deep Learning Approach

Due to its complexity and time-consuming nature, identifying gliomas at the Magnetic Resonance Imaging (MRI) slice-level before segmentation could assist clinicians in minimizing the time required for this procedure. In the literature, many studies proposed machine learning and deep learning-based algorithms for glioma identification at MRI slice-level. However, all these methods classify the slices using the previously extracted MRI features. So, the aim of this work is to propose a deep learning-based algorithm for the automatic detection of glioma from F LAIR MRI slices. Performance were assessed with a 5-fold cross validation and quantified by SENsitivity (SEN), SPEcificity (SPE), ACCuracy (ACC) and Area Under the Curve (AUC). Finally, a PDF report is generated reporting the MRI slice indexes where the tumor is. Results confirm the goodness of the proposed model with SEN=100%, SPE=99.4%, ACC=99.8% and AUC=99.9%. The model outperforms all studies found in the literature. In conclusion, the proposed model is the first feature-independent automated approach for glioma identification at the MRI slice-level. Additionally, the generation of a PDF report makes the model ready to use for clinicians.