306
M. J. Huttunen and A. Kiviniemi
scattering of light is considerably decreased when longer wavelengths are used to
excite objects. However, combining super-resolution techniques with good imaging
depths have turned out to be very challenging. Only in a very recent work a spatial resolution of 106 nm at imaging depth of around 100 µm using 3PEF has been
demonstrated [28]. The demonstrated technique is called interferometric temporal
focusing (ITF) microscopy, and combines SIM microscopy with the temporal focusing scheme. Despite the impressive progress, additional work is necessary to further
improve the imaging depths of multiphoton microscopies.
A promising route to image objects even deeper is to perform multiphoton
endoscopy, which can provide over 1 cm imaging depths facilitating, for example,
studies of intact animals [97–102]. Multiphoton endoscopes are minimally invasive and could thus be suitable for clinical applications, such as for label-free in
situ histopathology diagnosis. In order to facilitate excitation of objects using short
femtosecond pulses, gradient-index (GRIN) lenses are often used in multiphoton
endoscopes due to their minimal pulse dispersion [97, 98]. Despite the advantages
of multiphoton endoscopes a major drawback is their somewhat limited lateral and
axial resolutions, which commonly are around 1 µm and 6 µm, respectively [99, 101].
This is mostly due to moderate NAs (∼0.8) of available GRIN lenses. In addition, it
seems very challenging to implement a diffraction-limited GRIN lens-based imaging
system, in practice degrading the practical resolution further from the ideal one [97].
Therefore, major advances in the existing technology and capabilities seem entirely
feasible in the near future. In addition, interesting possibilities might arise by developing super-resolution endoscopy [103].
Another upcoming trend is the implementation of multimodal multiphoton systems [89, 104–107]. The motivation for multimodal multiphoton imaging is the possibility to gather more detailed information of the object under study, as is seen in
Fig. 12.10, demonstrating combined SHG, 2PEF, and CARS imaging of atherosclerotic lesions. In addition, modification of a multiphoton system into a multimodal
one is often relatively straightforward, and can in some cases be achieved just by
adding a few strategically placed dichroic mirrors, filters and detectors to the detection arms. This is especially the case for CARS microscopes, where several laser
beams at different wavelengths are already used.
For future applications, there is also a need to develop automated image analysis,
for example, to classify between healthy and cancerous ovarian tissues [78, 108].
Machine learning, convolutional neural networks, and deep learning are especially
seen as key technologies, which can be used to transform massive amounts of data,
easily recorded using a modern optical microscope, into as-it-is useful information.
These technologies could be especially valuable in clinical in situ applications benefiting from real-time image analysis, such as detection of residual disease at the
time of surgery. Because the amount (and quality) of data plays a key role in these
machine learning approaches, we expect that multimodal imaging is becoming more
and more important in the future.
M. J. Huttunen and A. Kiviniemi
scattering of light is considerably decreased when longer wavelengths are used to
excite objects. However, combining super-resolution techniques with good imaging
depths have turned out to be very challenging. Only in a very recent work a spatial resolution of 106 nm at imaging depth of around 100 µm using 3PEF has been
demonstrated [28]. The demonstrated technique is called interferometric temporal
focusing (ITF) microscopy, and combines SIM microscopy with the temporal focusing scheme. Despite the impressive progress, additional work is necessary to further
improve the imaging depths of multiphoton microscopies.
A promising route to image objects even deeper is to perform multiphoton
endoscopy, which can provide over 1 cm imaging depths facilitating, for example,
studies of intact animals [97–102]. Multiphoton endoscopes are minimally invasive and could thus be suitable for clinical applications, such as for label-free in
situ histopathology diagnosis. In order to facilitate excitation of objects using short
femtosecond pulses, gradient-index (GRIN) lenses are often used in multiphoton
endoscopes due to their minimal pulse dispersion [97, 98]. Despite the advantages
of multiphoton endoscopes a major drawback is their somewhat limited lateral and
axial resolutions, which commonly are around 1 µm and 6 µm, respectively [99, 101].
This is mostly due to moderate NAs (∼0.8) of available GRIN lenses. In addition, it
seems very challenging to implement a diffraction-limited GRIN lens-based imaging
system, in practice degrading the practical resolution further from the ideal one [97].
Therefore, major advances in the existing technology and capabilities seem entirely
feasible in the near future. In addition, interesting possibilities might arise by developing super-resolution endoscopy [103].
Another upcoming trend is the implementation of multimodal multiphoton systems [89, 104–107]. The motivation for multimodal multiphoton imaging is the possibility to gather more detailed information of the object under study, as is seen in
Fig. 12.10, demonstrating combined SHG, 2PEF, and CARS imaging of atherosclerotic lesions. In addition, modification of a multiphoton system into a multimodal
one is often relatively straightforward, and can in some cases be achieved just by
adding a few strategically placed dichroic mirrors, filters and detectors to the detection arms. This is especially the case for CARS microscopes, where several laser
beams at different wavelengths are already used.
For future applications, there is also a need to develop automated image analysis,
for example, to classify between healthy and cancerous ovarian tissues [78, 108].
Machine learning, convolutional neural networks, and deep learning are especially
seen as key technologies, which can be used to transform massive amounts of data,
easily recorded using a modern optical microscope, into as-it-is useful information.
These technologies could be especially valuable in clinical in situ applications benefiting from real-time image analysis, such as detection of residual disease at the
time of surgery. Because the amount (and quality) of data plays a key role in these
machine learning approaches, we expect that multimodal imaging is becoming more
and more important in the future.