DEEP-LESIONBRAIN: software to automatically segment white matter lesions from MRI data and to provide a report with the volumes of the lesions and their locations

Challenges

DeeplesionBrain (DLB) is a deep leanring based segmentation tool that is able to automatically segment the hyperintensity white matter lesions using a T1w MRI and a FLAIR MRI. DLB classified lesions into several categories according to their position. DLB is integrated within the web-service volBrain. First, the user provides a 3D T1 MRI and a FLAIR MRI to the system. The system will first preprocess the image (denoising, inhomogeneity correction, registration) and then will segment lesion. After few minutes, the results are sent back to the user by email. This email contains a report as pdf and csv files with volumes considered lesions. Moreover, the pdf file provides screenshots to enable a fast and efficient quality control by the user. DLB can be used for multiple sclerosis (MS), Alzheimer’s disease, aging, lupus, etc,

Innovative solution

DeeplesionBrain is a pipeline to automatically segment white matter lesions from MRI data (T1 + FLAIR). It gets anonymized MRI brain volumes in NIFTI format and produces a pdf report with the volumes of the lesions and their locations.

APPLICATIONS

• Clinical trials

• Companion diagnostics

• Clinical routine

COMPETITIVES ADVANTAGES

• No need of learning complex software packages

• No need having expensive computational infrastructures in your local site

• Faster and more accurate than other solutions

• LesionBrain works in a fully automatic manner and is able to provide brain structure volumes without any human interaction

• Two reports (CSV and PDF) providing all the volumetry values calculated from the segmentations as well as asymmetry indexes

DEVELOPMENT STATUS: 

• DeepLesionBrain is part of the volBrain platform wich has already processed more than 400.000 MRI.

How it works

White matter lesions were manually outlined by an expert radiologist using multimodal MRI data (T1w + FLAIR) to create a library of 43 cases. After the segmentation process, lesions are classified based on their location as periventricular, deep white, juxtacortical and infratentorial. In the figure a brief outline of the process is displayed.

Once the process is finished, you will be notified by e-mail so you will be able to download a package including some image files and two reports (CSV and PDF) offering all the volumes estimated from the segmentations. As you can see in the figure below, the PDF includes patient information, lesion classes, volumes and their locations in MNI space. It also includes several snapshots from the different labeling steps as a quality control.