Search

Mass spectrometry imaging can produce large amounts of complex spectral and spatial data. Such data sets are often analyzed with unsupervised machine learning approaches, which aim at reducing their complexity and facilitating their interpretation. However, choices made during data processing can impact the overall interpretation of these analyses. This work investigates the impact of the choices made at the peak selection step, which often occurs early in the data processing pipeline. The discussion is done in terms of visualization and interpretation of the results of two commonly used unsupervised approaches: t-distributed stochastic neighbor embedding and k-means clustering, which differ in nature and complexity. Criteria considered for peak selection include those based on hypotheses (exemplified herein in the analysis of metabolic alterations in genetically engineered mouse models of human colorectal cancer), particular molecular classes, and ion intensity. The results suggest that the choices made at the peak selection step have a significant impact in the visual interpretation of the results of either dimensionality reduction or clustering techniques and consequently in any downstream analysis that relies on these. Of particular significance, the results of this work show that while using the most abundant ions can result in interesting structure-related segmentation patterns that correlate well with histological features, using a smaller number of ions specifically selected based on prior knowledge about the biochemistry of the tissues under investigation can result in an easier-to-interpret, potentially more valuable, hypothesis-confirming result. Findings presented will help researchers understand and better utilize unsupervised machine learning approaches to mine high-dimensionality data.

High dimensionality omics and hyperspectral imaging datasets present difficult challenges for feature extraction and data mining due to huge numbers of features that cannot be simultaneously examined. The sample numbers and variables of these methods are constantly growing as new technologies are developed, and computational analysis needs to evolve to keep up with growing demand. Current state of the art algorithms can handle some routine datasets but struggle when datasets grow above a certain size. We present a training deep learning via neural networks on non-linear dimensionality reduction, in particular t-distributed stochastic neighbour embedding (t-SNE), to overcome prior limitations of these methods.One Sentence SummaryAnalysis of prohibitively large datasets by combining deep learning via neural networks with non-linear dimensionality reduction.

The EEG acquired simultaneously with fMRI is distorted by a number of artefacts related to the presence of strong magnetic fields, which must be reduced in order to allow for a useful interpretation and quantification of the EEG data. For the two most prominent artefacts, associated with magnetic field gradient switching and the heart beat, reduction methods have been developed and applied successfully. However, a number of artefacts related to the MR-environment can be found to distort the EEG data acquired even without ongoing fMRI acquisition. In this paper, we investigate the most prominent of those artefacts, caused by the Helium cooling pump, and propose a method for its reduction and respective validation in data collected from epilepsy patients. Since the Helium cooling pump artefact was found to be repetitive, an average template subtraction method was developed for its reduction with appropriate adjustments for minimizing the degradation of the physiological part of the signal. The new methodology was validated in a group of 15 EEG–fMRI datasets collected from six consecutive epilepsy patients, where it successfully reduced the amplitude of the artefact spectral peaks by 95 ± 2 % while the background spectral amplitude within those peaks was reduced by only −5 ± 4 %. Although the Helium cooling pump should ideally be switched off during simultaneous EEG–fMRI acquisitions, we have shown here that in cases where this is not possible the associated artefact can be effectively reduced in post processing.

A study of the epileptic seizure dynamics using simultaneous recording of electroencephalography correlated functional magnetic resonance imaging (EEG-fMRI) data from 5 focal epilepsy patients undergoing pre-surgical evaluation was realised with the aim of identifying the focus and propagation network seizure-involved. A method based on the General Linear Model (GLM) at different neurophysiology regressor lags (LasgM), a connectivity model-based method, Dynamic Causal Modelling (DCM), and a data-driven method, Granger Causality (GC) were investigated. DCM analysis provided meaningful and significant results when a sufficient number of seizure events was recorded, but suffered from the generally poor data signal-to-noise ratio (SNR). The concordance between the LagsM results and the clinical expectation suggests that LagsM can be useful as a complementary approach. Intending to establish the validity of a GC approach for the problem addressed, a simulation study was performed and the results showed that it is not appropriated.