381
13 Vibrational Spectroscopy: Disease Diagnostics and Beyond
no well defined shapes can be observed under these conditions. Another observation is the presence of 2 different organelles in the cytoplasm of the A549 cell. this
observation is not surprising as the cytoplasm of the cells has a complex organisation and different structures can be found such as the actin or microtubules [135,
136], endoplasmic reticulum [136, 137], the mitochondria [138] or the golgi apparatus [139, 140].
Raman spectral maps were performed in immersion using a horiba Jobin Yvon
LabRam hR800 using a 785 nm laser as source. the step between two successive
measurements was set to 1.5 µm and the backscattered Raman signal was integrated
for 10 s over the spectral ranges from 400 to 1,800 cm
−1
and accumulated twice to
improve the signal to noise ratio. In order to visualise the subcellular structures in
the spectral maps recorded, K-means clustering analysis was employed.
K-means clustering analysis is one of the simplest unsupervised learning algorithms used for spectral image analysis. It groups the spectra according to their
similarity, forming clusters, each one representing regions of the image with identical molecular properties [141]. the distribution of chemical similarity can then be
visualised across the sample image. the number of clusters (k) has to be determined
a priori by the operator before initiation of the classification of the data set. K centroids are defined, ideally as far as possible from each other, and then each point
belonging to a data set is associated to the nearest centroid. When all the points have
been associated with a centroid, the initial grouping is done. the second step consists of the calculation of new centroids as barycentres of the clusters resulting from
the previous step. A new grouping is implemented between the same data points
and the new centroids. these operations are repeated until convergence is reached
and there is no further movement of the centroids. Finally k clusters are determined,
each containing the most similar spectra from the image. From here, colours can be
attributed to each cluster and false colour maps can be constructed to visualise the
organisation of the clusters in the original image.
K-means clustering analysis is currently commonly used for the analysis of tissue sections by infrared and Raman spectroscopy, especially for the detection of the
different structures present in the samples studied and the identification of tumoral
regions [142–144]. K-means clustering can also be employed for the study of individual cells by Raman spectroscopy but only a few studies have elucidated the
subcellular organisation and they are usually based on fixed cells grown on a non
optimal substrate such a quartz window or equivalent [127, 145]. the effects of cell
fixation on the cellular composition has been already investigated [129, 130] and
thus it is desirable to be able to obtain the Raman signature from live cells, avoiding
any structural and molecular modification which may occur as a result of the chemical processing. moreover, the use of collagen gels enables extended measurement
periods and thus more detailed mapping of live cells. In contrast to quartz substrates
which have a toxic effect necessitating fixation of the cells for prolonged measurement times, collagen gels offer a better adhesion of the cells to the substrate, enhancing the cell viability.
Within the cells, it may be predicted that distinct biochemical regions can be
identified, corresponding to, for example, nucleus, nucleoli, cytoplasm and cell
13 Vibrational Spectroscopy: Disease Diagnostics and Beyond
no well defined shapes can be observed under these conditions. Another observation is the presence of 2 different organelles in the cytoplasm of the A549 cell. this
observation is not surprising as the cytoplasm of the cells has a complex organisation and different structures can be found such as the actin or microtubules [135,
136], endoplasmic reticulum [136, 137], the mitochondria [138] or the golgi apparatus [139, 140].
Raman spectral maps were performed in immersion using a horiba Jobin Yvon
LabRam hR800 using a 785 nm laser as source. the step between two successive
measurements was set to 1.5 µm and the backscattered Raman signal was integrated
for 10 s over the spectral ranges from 400 to 1,800 cm
−1
and accumulated twice to
improve the signal to noise ratio. In order to visualise the subcellular structures in
the spectral maps recorded, K-means clustering analysis was employed.
K-means clustering analysis is one of the simplest unsupervised learning algorithms used for spectral image analysis. It groups the spectra according to their
similarity, forming clusters, each one representing regions of the image with identical molecular properties [141]. the distribution of chemical similarity can then be
visualised across the sample image. the number of clusters (k) has to be determined
a priori by the operator before initiation of the classification of the data set. K centroids are defined, ideally as far as possible from each other, and then each point
belonging to a data set is associated to the nearest centroid. When all the points have
been associated with a centroid, the initial grouping is done. the second step consists of the calculation of new centroids as barycentres of the clusters resulting from
the previous step. A new grouping is implemented between the same data points
and the new centroids. these operations are repeated until convergence is reached
and there is no further movement of the centroids. Finally k clusters are determined,
each containing the most similar spectra from the image. From here, colours can be
attributed to each cluster and false colour maps can be constructed to visualise the
organisation of the clusters in the original image.
K-means clustering analysis is currently commonly used for the analysis of tissue sections by infrared and Raman spectroscopy, especially for the detection of the
different structures present in the samples studied and the identification of tumoral
regions [142–144]. K-means clustering can also be employed for the study of individual cells by Raman spectroscopy but only a few studies have elucidated the
subcellular organisation and they are usually based on fixed cells grown on a non
optimal substrate such a quartz window or equivalent [127, 145]. the effects of cell
fixation on the cellular composition has been already investigated [129, 130] and
thus it is desirable to be able to obtain the Raman signature from live cells, avoiding
any structural and molecular modification which may occur as a result of the chemical processing. moreover, the use of collagen gels enables extended measurement
periods and thus more detailed mapping of live cells. In contrast to quartz substrates
which have a toxic effect necessitating fixation of the cells for prolonged measurement times, collagen gels offer a better adhesion of the cells to the substrate, enhancing the cell viability.
Within the cells, it may be predicted that distinct biochemical regions can be
identified, corresponding to, for example, nucleus, nucleoli, cytoplasm and cell