Development of the Self Optimising Kohonen Index Network (SKiNET) for Raman Spectroscopy Based Detection of Anatomical Eye Tissue

Abstract

Raman spectroscopy shows promise as a tool for timely diagnostics via in-vivo spectroscopy of the eye, for a number of ophthalmic diseases. By measuring the inelastic scattering of light, Raman spectroscopy is able to reveal detailed chemical characteristics, but is an inherently weak effect resulting in noisy complex signal, which is often difficult to analyse. Here, we embraced that noise to develop the self-optimising Kohonen index network (SKiNET), and provide a generic framework for multivariate analysis that simultaneously provides dimensionality reduction, feature extraction and multi-class classification as part of a seamless interface. The method was tested by classification of anatomical ex-vivo eye tissue segments from porcine eyes, yielding an accuracy >93% across 5 tissue types. Unlike traditional packages, the method performs data analysis directly in the web browser through modern web and cloud technologies as an open source extendable web app. The unprecedented accuracy and clarity of the SKiNET methodology has the potential to revolutionise the use of Raman spectroscopy for in-vivo applications.

Spectroscopic detection of traumatic brain injury severity and biochemistry from the retina

“…Fortunately, multivariate techniques allow us to capture all of this information and extract the most important spectral features which characterize a group of data, such as an injury state. Recently, we highlighted the value of SOMs in the analysis of Raman spectra from biological samples [ 14 ]. The 400 spectra measured across each tissue sample were grouped according to injury state from both eyes.…”

“…Recently, we developed a machine learning technique based on self organizing maps (SOM)s, the self optimizing Kohonen index network (SKiNET) for simultaneously providing rich information and classification from biological samples, even with noisy or poor quality spectra that would result from a lower laser power and short acquisition times [ 14 ]. SOMs provide visually intuitive 2D clustering (e.g.…”

Spectroscopic profiling variations in extracellular vesicle biochemistry in a model of myogenesis

“…This has provided no additional insight into the data and therefore was omitted for the final analyses ( Supplemental Figure S1 ). The EVs data was further analysed through a self-organising map (SOM) classification model implementing learn vector quantisation (LVQ), 35 (1000 learning step and a learning rate of 0.5). The SOM data supports the PCA findings in terms of being able to distinguish between the two EV populations at Days 2, 10 and 13 but not at day 6 ( Supplemental Figure S2 ).…”

Unsupervised Functional Link Artificial Neural Networks for Cluster Analysis

“…Firstly, we use the standardized accuracy evaluation metric because the synthetic/real data we generated/ scraped had predefined classes that could be used to treat it as a standard classification problem, and penalizing the metric for wrongly generated predictions. Thus, the first metric is unsupervised clustering accuracy (ACC): (27) where, yi is the ground-truth label, ci is the cluster assignment generated by the algorithm, and the map is a mapping function that ranges over all possible one-to-one mappings between assignments and labels.…”

“…On the other hand, if the output neurons of a competitive learning network are arranged geometrically (such as in a one-dimensional array or two-dimensional arrays), then we can update the weight vectors of the winners as well as the neighbouring losers. Such a capability corresponds to the notion of Kohonen feature maps [27,28]. The above unsupervised neural networks have suffered from many difficulties like slow learning rate, trapping in local optimality, not scale well for patterns with a large number of elements, etc.…”