27 nov 2020
08:45 Defesa de Doutorado University of Groningen
Unsupervised Brain Anomaly Detection in MR Images
Samuel Botter Martins
Orientador / Docente
Alexandre Xavier Falcão e Alexandru-Christian Telea
Breve resumo
Brain disorders are characterized by morphological deformations in shape and size of (sub)cortical structures in one or both hemispheres. These deformations cause deviations from the normal pattern of brain asymmetries, resulting in asymmetric lesions that directly affect the patient’s condition. It is hence clinically crucial to define normal brain asymmetries for the identification and detection of these deformations (brain anomalies) early for proper diagnosis and treatment. Most automatic computational methods in the literature rely on supervised machine learning to detect or segment anomalies in brain images. However, these methods require a large number of high-quality annotated training images, which is absent for most medical image analysis problems. Besides, they are only designed for the lesions found in the training set, and some methods still require weight fine-tuning (retraining) when used for a new set of images. In contrast, unsupervised methods aim to learn a model from unlabeled healthy images, so that an unseen image that breaks priors of this model, i.e., an outlier, is considered an anomaly. As these methods do not use labeled images, they are less effective in detecting lesions from a specific disease when compared to supervised approaches trained from labeled images for the same disease. For the same reason, however, unsupervised methods are generic in detecting any lesions, e.g., coming from multiple diseases, as long as these notably differ from healthy training images. This thesis addresses the development of solutions to leverage unsupervised machine learning for the detection/analysis of abnormal brain asymmetries related to anomalies in magnetic resonance (MR) images. First, we propose an automatic probabilistic-atlas-based approach for anomalous brain image segmentation. Its goal is to define our target macro-regions of interest — i.e., right and left hemispheres, cerebellum, and brainstem — to improve the preprocessing, restrict the analysis, and compute hemispheric asymmetries in some cases. Second, we explore an automatic method for the detection of abnormal hippocampi from abnormal asymmetries. Our solution uses deep generative networks and a one-class classifier to model normal hippocampal asymmetries inside pairs of 3D patches from healthy subjects and detect abnormal hippocampi. Third, we present a more generic framework to detect abnormal asymmetries in the entire brain hemispheres. Our approach extracts pairs of symmetric regions — called supervoxels — in both hemispheres of a test image under study. One-class classifiers then analyze the asymmetries present in each pair. This method is limited to detect asymmetric lesions only in the hemispheres. Finally, we generalize the previous solution for the detection of (a)symmetric lesions based on registration errors. Experimental results on 3D MR-T1 images from healthy subjects and patients with a variety of lesions show the effectiveness and robustness of the proposed unsupervised approaches for brain anomaly detection.
Banca examinadora
Alexandre Xavier Falcão IC/UNICAMP
Alexandru-Christian Telea University of Groningen
Nicolai Petkov University of Groningen
Michael Biehl University of Groningen
Roberto Marcondes Cesar Junior IME/USP
Ricardo da Silva Torres IC/UNICAMP