Automated Segmentation of Hyperintense Regions in FLAIR MRI Using Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Accurate segmentation of brain tumors in MRI is crucial for diagnosis and treatment planning.
• Manual segmentation is time-consuming and subject to interobserver variability.
• Deep learning offers potential for automated and consistent image analysis.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) based on autoencoders for segmenting signal-increased regions in fluid-attenuated inversion recovery (FLAIR) magnetic resonance imaging (MRI).
• To assess the performance of the proposed method against expert segmentations and determine if it falls within interobserver variability.

Main Methods:

• A deep convolutional neural network utilizing autoencoders was designed for image segmentation.
• The model was trained using the Brain Tumor Image Segmentation Benchmark (BRATS) dataset.
• Performance evaluation involved comparing automated segmentations against three expert segmentations using metrics like Dice coefficient, Jaccard coefficient, true positive fraction, and false negative fraction.

Main Results:

• The developed CNN autoencoder achieved segmentations within the interobserver variability for Dice coefficient, Jaccard coefficient, and true positive fraction.
• The method demonstrated high accuracy in identifying and segmenting tumor-related signal-increased regions.

Conclusions:

• The proposed deep learning approach provides an effective and automated method for segmenting signal-increased regions in FLAIR MRI.
• This technique shows potential to reduce manual segmentation efforts and improve consistency in brain tumor analysis.