The paper was submitted in 2014 and accepted in 2015 in IEEETr Pattern Analysis and Machine Intelligence - DOI 10.1109/TPAMI.2015.2404835 - IF=5.694; IQ - engineering, electrical & electronic ; 5 ou of 248 Mixture models are often used to compactly represent samples from heterogeneous sources. However, in real world, the samples generally contain an unknown fraction of outliers and the sources generate different or unbalanced numbers of observations. Such unbalanced and contaminated samples may, for instance, be obtained by high density data sensors such as imaging devices. Estimation of unbalanced mixture models from samples with outliers requires robust estimation methods. In this paper, we propose a novel robust mixture estimator incorporating trimming of the outliers based on component-wise confidence level ordering of observations. The proposed method is validated and compared to the state-of-the-art FAST-TLE method on two datasets, one consisting of synthetic samples with a varying fraction of outliers and a varying balance between mixture weights, while the other dataset contained structural magnetic resonance images of the brain with tumors of varying volumes. The results on both datasets clearly indicate that the proposed method is capable to robustly estimate unbalanced mixtures over a broad range of outlier fractions. As such, it is applicable to real-world samples, in which the outlier fraction cannot be estimated in advance.
COBISS.SI-ID: 9795156
Methods for automated segmentation of brain MR images are routinely used in large-scale neurological studies. Automated segmentation is usually performed by unsupervised methods, since these can be used even if different MR sequences or different pathologies are studied. The unsupervised methods model intensity distribution of major brain structures using mixture models, the parameters of which need to be robustly estimated from MR data and in presence of outliers. In this paper, we propose a robust mixture-parameter estimation that detects outliers as samples with low significance level of the corresponding mixture component and iteratively re-estimates the fraction of outliers. Results on synthetic and real brain image datasets demonstrate superior robustness of the proposed method as compared to the popular FAST-TLE method over a broad range of trimming fraction values. The latter is important for segmenting brain structures with pathology, the extent of which is hard to predict in large-scale imaging studies.
COBISS.SI-ID: 10527316
Detection of longitudinal changes in brain structures is a common clinical task when assessing the progress of cerebrovascular and neurodegenerative diseases, which manifest in appearing and disappearing white matter lesions (WMLs). Changes of WMLs are usually quantified by their manual outlines and compared across longitudinal, serial magnetic resonance (MR) brain images. Since manual outlining in 3D MR images is subjective and inaccurate, several automated methods were proposed so as to enhance the sensitivity, reliability and repeatability of change detection of WMLs. However, the absence of publicly available synthetic or clinical MR image databases with corresponding ground truth of changes renders the validation and comparison of any new and existing automated methods highly subjective. In this paper, we focus on the validation and comparison of three state-of-the-art intensity based methods for detection of longitudinal changes of WMLs. To objectively assess the three methods we created several synthetic MR image databases using a generative lesion model, which was trained on manually outlined patches of WMLs in a clinical MR image database of 22 patients. Validation was also performed on clinical MR image database of MS patients. Performances of the three change detection methods were evaluated by computing the similarity index and sensitivity between the obtained and the ground truth binary change map. The obtained similarity indices were in the range of 0.40-0.77, which should be improved for clinical use, while the comparison of methods revealed that the intensity subtraction method achieved similar performance as the change vector analysis method, which employed two MR sequences for change detection. Third method was based on local steering kernels and exhibited stable performance both on synthetic and clinical MR image databases.
COBISS.SI-ID: 10811220