Early and more sensitive detection of small aneurysms in 3D cerebral angiograms is required to prevent potentially fatal rupture events. We propose a novel method that entails structure enhancement filtering to highlight potential aneurysm locations, intra-vascular distance mapping for regional vascular shape encoding and dimensionality reduction and a convolutional neural network to automatically determine optimal features and classification rules for aneurysm detection. Evaluation on 15 3D digital subtraction angiograms showed better performance of the proposed method compared to a state-of-the-art method based on enhancement filtering and random forest classification, as it achieved a 100% detection sensitivity at a low number of false positives (2.4 per dataset). The benefit of the proposed method is that it is also applicable to other angiographic modalities, which further increases its practical value.
COBISS.SI-ID: 11774036
A certain quantity can usually be measured by several different measurement methods. For instance, in medical image analysis, there may be different image segmentation methods to delineate and measure the volume of a certain structure, but the true value may still be unknown or accessible using a more costly or even destructive method. We proposed a novel computational framework that automatically determined the most accurate and precise method of measurement of a certain quantity, when there was no access to the true value of the measurand. The accuracy of each measurement method was characterized by systematic error (bias), which was modeled with polynomials in true values of measurand, and the precision as random error modeled with a Gaussian random variable. In contrast to previous works the random errors were modeled jointly across all methods, thereby enabling the framework to analyze measurement methods based on similar principles, which may have correlated random errors. Furthermore, the posterior distribution of the error model parameters is estimated from samples obtained by Markov chain Monte-Carlo and analyzed to estimate the parameter values and the unknown true values of the measurand. The framework was validated on a dataset containing measurements of total lesion load, a biomarker of neurodegenerative diseases, which was obtained with four automatic methods by analyzing brain magnetic resonance images. The obtained estimates of bias and random error, and true value estimates, were in a good agreement with the corresponding least squares regression estimates against a reference consensus-based measurements.
COBISS.SI-ID: 11948116