Evaluating severity of white matter lesions from computed tomography images with convolutional neural network

Johanna Pitkänen,Philip Scheltens,Leonardo Pantoni,Daniel Rueckert,Tuomas Nieminen,Frederik Barkhof,Hanna Jokinen,Gerli Sibolt,Ivan Marinkovic,Juha Koikkalainen,Lars-Olof Wahlund,Sami Curtze,Antti Korvenoja,Susanna Melkas,Jyrki Lötjönen,Reinhold Schmidt,Timo Erkinjuntti

doi:10.1007/s00234-020-02410-2

Abstract

PurposeSeverity of white matter lesion (WML) is typically evaluated on magnetic resonance images (MRI), yet the more accessible, faster, and less expensive method is computed tomography (CT). Our objective was to study whether WML can be automatically segmented from CT images using a convolutional neural network (CNN). The second aim was to compare CT segmentation with MRI segmentation.MethodsThe brain images from the Helsinki University Hospital clinical image archive were systematically screened to make CT-MRI image pairs. Selection criteria for the study were that both CT and MRI images were acquired within 6 weeks. In total, 147 image pairs were included. We used CNN to segment WML from CT images. Training and testing of CNN for CT was performed using 10-fold cross-validation, and the segmentation results were compared with the corresponding segmentations from MRI.ResultsA Pearson correlation of 0.94 was obtained between the automatic WML volumes of MRI and CT segmentations. The average Dice similarity index validating the overlap between CT and FLAIR segmentations was 0.68 for the Fazekas 3 group.ConclusionCNN-based segmentation of CT images may provide a means to evaluate the severity of WML and establish a link between CT WML patterns and the current standard MRI-based visual rating scale.

Highlights

White matter lesions (WML) are a surrogate for cerebral small vessel disease (SVD), which is the major cause of accumulating vascular burden in aging populations
The Dice overlap measures that the ratio of voxels segmented as WML in both images and the voxels segmented as WML in computed tomography (CT) and in FLAIR: Dice 1⁄4 j2XjXjþ∩jYYjj, where |X| and |Y| are the WML volumes of the CT and FLAIR segmentations, and |X ∩ Y| is the volume of voxels segmented as WML in both CT and FLAIR
The index values are low for small WML volumes: the average Dice similarity index was 0.43 for the whole dataset

Summary

Introduction

White matter lesions (WML) are a surrogate for cerebral small vessel disease (SVD), which is the major cause of accumulating vascular burden in aging populations. The most common method for grading WML extent has been the Fazekas visual rating scale developed for MRI [5, 6]. It was preceded by several proposals for CT-based visual rating scales by the authors Gorter [7], Blennow et al [8], van Swieten et al [9], and Wahlund et al [10] which have not been widely adopted in clinical practice [6, 11]. Computer-aided image analysis and machine learning methods are increasingly used in medicine. They enable automated and quantitative analyses of large image databases and help to develop tools that complement the manual visual assessment. Especially in the field of deep learning, have improved the ability to identify, quantify, and classify patterns in medical images [11]

Objectives

Methods

Results

Conclusion