Unsupervised representation learning for anomaly detection on neuroimaging. Application to epilepsy lesion detection on brain MRI

Z Alaverdyan 1
1 Images et Modèles
CREATIS - Centre de Recherche en Acquisition et Traitement de l'Image pour la Santé
Abstract : Epilepsy affects around 50 million people worldwide, a third of those diagnosed with medically refractory epilepsy where seizures cannot be controlled by pharmacotherapy. For such patients, surgical resection of the epileptogenic zone may offer a seizure-free life. The success of such surgeries largely depends on the accuracy of the epileptogenic zone localization. Neuroimaging, including magnetic resonance imaging (MRI) and positron emission tomography (PET), has been increasingly considered in the pre-surgical examination routine. This work represents one attempt to develop a computer aided diagnosis system for epileptogenic lesion detection based on neuroimaging data, in particular T1-weighted and FLAIR MR sequences. Given the complexity of the task and the lack of a representative voxel-level labeled data set, the adopted approach, first introduced in [El Azami et al., 2016], consists in casting the lesion detection task as a per-voxel outlier detection problem. The system is based on training a one-class SVM model for each voxel in the brain on a set of healthy controls, so as to model the normality of the voxel. For an unseen patient, each voxel is assessed against the corresponding one-class SVM model which yields a signed score of its anomalousness. Anomalous lesions can hence be found as local neighborhoods of voxels with low scores. The main focus of this work is to design representation learning mechanisms, capturing the most discriminant information from multimodality imaging. Manual features, designed to mimic the characteristics of certain epilepsy lesions, such as focal cortical dysplasia (FCD), on neuroimaging data, are tailored to individual pathologies and cannot discriminate a large range of epilepsy lesions. Such features reflect the known characteristics of lesion appearance; however, they might not be the most optimal ones for the task at hand. Our first contribution consists in proposing various unsupervised neural architectures as potential feature extracting mechanisms and, eventually, introducing a novel configuration of siamese networks, to be plugged into the outlier detection context. The proposed system, evaluated on a set of T1-weighted MRI of epilepsy patients, showed a promising performance but a room for improvement as well. To this end, we considered extending the CAD system so as to accommodate multimodality data which offers complementary information on the problem at hand. Our second contribution, therefore, consists in proposing strategies to combine representations of different imaging modalities into a single framework for anomaly detection. The extended system showed a significant improvement on the task of epilepsy lesion detection on T1-weighted and FLAIR images. Our last contribution focuses on the integration of PET data into the system. An obstacle encountered often in medical applications is the small number of subjects with the full set of imaging modalities. This limits the performance of a system when the subjects with missing data are discarded. We therefore delve into strategies of synthesizing PET data from the corresponding MRI acquisitions and show an improved performance of the system when synthesized images are used in addition to the real ones.
Complete list of metadatas

Cited literature [335 references]  Display  Hide  Download

Contributor : Carole Lartizien <>
Submitted on : Friday, March 8, 2019 - 4:17:00 PM
Last modification on : Thursday, July 18, 2019 - 1:25:21 AM
Long-term archiving on : Monday, June 10, 2019 - 3:04:21 PM


Files produced by the author(s)


  • HAL Id : tel-02062210, version 1



Z Alaverdyan. Unsupervised representation learning for anomaly detection on neuroimaging. Application to epilepsy lesion detection on brain MRI. Medical Imaging. Université de Lyon, 2019. English. ⟨tel-02062210v1⟩



Record views


Files downloads