PhD Scientific Days 2018

Budapest, April 19–20, 2018

Single patients as outliers in neuroimaging? Introducing the Mahalanobis-distance for epileptic lesion detection using DTI data

Gyebnár, Gyula

Gyula Gyebnár 1, Zoltán Klimaj 1, László Entz 2, Dániel Fabó 2, Gábor Rudas 1, Péter Barsi 1, Lajos R Kozák 1
1. Magnetic Resonance Research Centre, Semmelweis University, Budapest, Hungary
2. National Institute of Clinical Neurosciences, Budapest, Hungary

Language of the presentation


Text of the abstract

Identifying a possible malformation of cortical development (MCD) behind a drug resistant epilepsy case often poses a serious challenge on conventional MR images. Recently, quantitative MRI methods gained a lot of interest and are currently seeing substantial developments (for example ‘MR-fingerprinting’, ‘Radiomics’, etc.); however, their application in such problems is often limited by the statistical difficulties of single patient vs control group comparisons.
The current study was aimed at developing a new voxel-wise evaluation method based on the Mahalanobis-distance, by combining quantitative MRI-maps into a multidimensional parameter space and detecting lesion voxels as outliers compared to the distribution of values corresponding to healthy individuals.
Maps of the three DTI eigenvalues were used in extensive simulations and in the examination of 18 patients with MCDs, and 45 controls. DWI processing, including thorough corrections and robust tensor fitting, was performed with ExploreDTI. Spatial coregistration was achieved with the DARTEL tools of SPM12. Optimal critical values and cluster size thresholds were determined by simulations with multivariate Gaussian data and resampled eigenvalue maps of healthy individuals with added ‘lesions’ of controlled effect strength; clusters emanating from registration artefacts were classified by a semi-automatic method, fine-tuned using leave-one-out approach on the control data.
Clusters of outlying diffusion profiles, concordant with expert neuroradiological evaluation and independent automated analysis using the MAP07 toolbox, were identified in 17 cases.
The proposed multidimensional statistical method proved sufficiently sensitive in pinpointing regions of abnormal tissue microstructure; however, inherent registration artefacts and the systematic nature of the underlying pathologies impeded the generalization of findings. Nevertheless, this approach may aid the everyday clinical workup, ever so more upon extending the framework with quantitative information from other MRI modalities, such as susceptibility mapping, relaxometry, or perfusion.

Data of the presenter

Doctoral School: János Szentágothai Doctoral School of Neurosciences
Program: Clinical Neurological Research
Supervisor: Lajos R Kozák
E-mail address: