Multiview Deconvolution Two-photon Laser Scanning Microscopy

Imaging in three dimensions is necessary for thick tissues and small organisms. This is possible with tomographic optical microscopy techniques such as confocal, two-photon and light sheet microscopy. All these techniques suffer from anisotropic resolution and limited penetration depth. In the past, Multiview microscopy - imaging the sample from different angles followed by 3D image reconstruction - was developed to address this issue for light sheet microscopy based on uorescence signal. In this study we applied this methodology to accomplish Multiview imaging with two-photon microscopy based on uorescence and additionally second harmonic signal from myosin and collagen. It was shown that isotropic resolution was achieved, the entirety of the sample was visualized, and interference artifacts were suppressed allowing clear visualization of collagen brils and myobrils. This method can be applied to any scanning microscopy technique without microscope modications. It can be used for imaging tissue and whole mount small organisms such as heart tissue, and zebrash larva in 3D, label-free or stained, with at least 3-fold axial resolution improvement which can be signicant for the accurate quantication of small 3D structures.


Introduction
In the last two decades, numerous novel optical microscopy imaging methods have been developed.
They centre around increasing resolution below the diffraction limit (Super resolution methods), increasing penetration depth in tissue (Multiphoton microscopy and tissue clearing), and increasing speed and reducing photodamage for high-throughput imaging of large elds of view (Light sheet microscopy). Whatever their speci c goal, all optical microscopy methods suffer from anisotropic resolution. Consequently, the details visible in the axial plane (XZ or YZ plane) are of inferior quality to those visualized in the imaging plane (XY-plane). When imaging at samples such as cultured cells, the manifestation of this anisotropic resolution may not be as signi cant, however when imaging thick samples, such as tissues or small organisms, it will result in blurry 3D objects. This is more pronounced for two-photon laser scanning microscopy (TPLSM), where typically water immersion objectives between 0.6-1.0 NA are used. Based on the theoretical resolution of TPLSM 1 , for an 1.0 NA objective at 820 nm excitation, anisotropy (the ratio of axial to lateral resolution) is 0.27, whereas for a 0.6 NA objective it is 0.13. Consequently, objects with size in the range of the excitation wavelength, will appear very elongated in the axial dimension and this distortion is exacerbated with lower NA objectives.
Anisotropic resolution in microscopy has been addressed before. It has been shown that the use of two opposing objectives creates an interference excitation eld, which signi cantly reduces the axial size of the point spread function (PSF) and therefore axial resolution. This was implemented in scanning mode through 4Pi microscopy 2 and in wide eld mode in I 5 M 3 . Both techniques required deconvolution to remove shadow effects and were able to produce images with 100 nm axial resolution. However there were limitations in the sample thickness (100 µm), e cient removal of artifacts, and alignment of the opposing beams 4 , which limited their applicability and wide spreading to the biomedical community.
Another method proposed was the tilted view imaging 5 of the sample and subsequent reconstruction of the 3D image. Although proposed almost three decades ago it did not nd many applications due to di culties in implementation and image processing.
The idea of tilted view 5 or multiview imaging was later adapted in Light Sheet Fluorescence Microscopy 6 (LSFM) (or Selective Plane Illumination microscopy -SPIM), to decrease anisotropy in resolution.
Implementation of this so-called Multiview Imaging (MVI) in LSFM, meaning rotating and imaging the sample from different angles, allowed achievement of isotropic resolution (anisotropy equals 1) 7,8 . Since detection is camera based on a wide eld con guration, LSFM is a fast technique with high signal-tonoise ratio (SNR). Using LSFM has enabled imaging and tracking of single cells, in vivo imaging of the development of whole embryos of drosophila and zebra sh, and imaging of large xed and cleared specimens such as whole brain of rodents at high speed [9][10][11][12] . MVI has become a standard imaging technique in LSFM, and it can be either applied sequentially or simultaneously in order to reduce acquisition time 11 . Further decrease of resolution in MVI can be accomplished through deconvolution. By applying MVI deconvolution (MVD) the artifacts caused by the elongated PSF of each view can be minimized and this can improve further the contrast and decrease the resolution of the image. This is particularly useful when the imaged structures are in the range of the excitation wavelength 8 .
Despite computational power being readily available, MVI, has not found many applications in scanning microscopy, such as TPLSM and confocal microscopy. Indeed, some applications in TPLSM have been demonstrated, based on the signal of Second Harmonic Generation (SHG) from collagen bres 13,14 .
However, MVD has not been demonstrated. This additional step could offer improved contrast and decreased resolution.
TPLSM has the advantages of deeper tissue penetration, due to the use of near infrared excitation, lower photobleaching, due to the exclusive focal excitation, excitation of multiple uorophores with a single excitation wavelength, due to increase excitation pathways 15 . Moreover it has the unique advantage of inducing Second Harmonic Generation (SHG) signal in the proteins collagen, myosin, and tubulin 16 .
These advantages allow imaging thick tissue Sect. 17 or small organisms 18 , detection of multiple signals simultaneously 19 , imaging of stained structures but also detection of inherent auto uorescence signal and SHG 20 , with reduced overall photobleaching.
In this study the possibilities and capabilities of applying MVD in TPLSM, based on both uorescence and SHG contrast were explored. For this purpose, a custom rotation chamber was constructed and was used on a commercial TPLSM. Image processing was performed with processing tools already developed for LSFM. MVD was performed on uorescence and SHG signal. It was demonstrated that isotropic resolution could be achieved with MVD and that this method could be applied to any scanning microscope without microscope modi cations.

Samples
Zebra sh larvae at 3 days post fertilization (dpf) were used. The transgenic line sensory:GFP was used 21 to visualize all sensory neurons in the developing larvae in a nacre background, lacking melanophores 22 , thus eliminating interference from pigmented cells. Larvae were xated for 15 min in 4% formaldehyde and stored in PBS at 4 °C until further use. Subsequently, nuclei were stained using SYTOX orange 2 nM (Invitrogen, LOT: 1933342) for 10 min. For heart tissue imaging, the heart from a zebra sh larva was dissected after xation. Rat-tail tendon was extracted from a healthy adult rat, xated in 4% formaldehyde for 15 min, and stored in PBS. For imaging, a small section was cut and used. All samples were stored at 4 °C. The experiments in this study were performed according to Dutch regulations and approved by the Dutch Central Committee of Animal Use (CCD) and the Maastricht University Committee for Animal Welfare. A speci c animal study protocol for the zebra sh was not required since the samples used were younger than 5dpf. Rat tail tendon was acquired based on the Three Rs principle from an approved CCD protocol (2016-004).

PSF generation
Theoretical PSF were generated on Huygens (SVI, Hilversum, the Netherlands). Parameters used were 1.00 NA objective, water immersion, 820 nm excitation, 520 nm emission.

Imaging
For imaging, a Leica TCS SP5 (Leica Microsystems GmbH, Wetzlar, Germany) TPLSM was used, with a Ti-Sapphire Chameleon Ultra II (Coherent Inc, Santa Clara, CA, USA) laser. Excitation was at 820 nm. A Leica objective, HCX APO L 20x/1.00 was used. Fluorescence detection was performed using the descanned detectors set according to the emission spectra of the dyes used in each sample. Image acquisition was performed simultaneously for all channels. SYTOX was detected at 560-600 nm, GFP at 500-540 nm, and SHG signal from collagen and myosin were detected with a forward detector with a bandpass lter (380-420 nm). Fluorescence beads (FluoSpheres, Life Technologies Inc, Eugene, OR, USA, LOT:1835801, excitation/emission 540/560 nm) were used as ducial marker and detected in the same channel as SYTOX. Auto uorescence in the zebra sh sample was detected in the same channel as GFP.

Rotation chamber
A custom chamber made of acrylic glass, was constructed (Fig. 1A). A motor, controlled by an Arduino board, was attached to the rotation shaft of the chamber for the rotation of the sample. The chamber was mounted on the microscope stage between the objective and the condenser lens (NA:0.9) (Fig. 1B). The chamber was lled with d-water. Experiments were performed at room temperature (~ 21 °C).

Sample preparation
The samples were immersed in 1% low melting point agarose gel (Sigma, LOT:SLBW8410) containing uorescent beads. A stock solution of beads was prepared (2 µl dispersed in 0.5 ml water) and 16 µM of the stock solution was dissolved in 0.2 ml, 1% low melting point agarose containing the sample. The mixture, while still liquid, was suctioned in a haematocrit capillary. The capillary was subsequently mounted on the rotation chamber. After the mixture solidi ed (typically 1-2 minutes), it was extruded to the opposing capillary. The agarose column containing the sample was exposed to the objective (Fig. 1C). A drop of agarose on each side of the capillary was added to ensure the agarose column, containing the sample, was rmly attached to the rotating capillary.
Performing MVD includes several steps and caution must be taken to avoid pitfalls. Regarding sample mounting it is important to achieve good refractive index matching. For this reason, low melting point agarose was used, which has a refractive index similar to that of water. The low agarose content makes the exposed agarose column fragile and caution must be taken on the length of this column so that it does not deform during rotation. In practice, exposed agarose columns of 3-4 µm where rigid enough ( Fig. 1C).

Sample rotation
The sample was rotated at discrete angles with a stepper motor controlled by an Arduino board. Sample could be rotated with 1° angle resolution and full 360° was possible. Rotation steps between 45°, and 90°a ngles, depending on the experiment were selected.

Multiview image reconstruction and deconvolution
After acquisition, images were analysed with FIJI 23 . Registration, fusion, and deconvolution were performed with the FIJI plugin Multiview reconstruction 8 . Registration was based on the beads, which served as ducial markers, embedded on the agarose column containing the sample. MVD was performed with the same plugin using the PSF extracted from the beads, with deconvolution setting "independent", on 10 iterations.

Multiview imaging resolution
In this section, the PSF of single view (SV) and MVI were calculated. The focus of this study was on imaging thicker samples were all details are not equally visualized in all views, therefore multiple views are necessary to capture the whole object. We calculated the PSF for the ideal case where the PSF is not deformed due to aberrations at different depths and all details are equally visible in all views. The purpose of this calculation is to estimate the degree of resolution improvement and isotropy it can be achieved. Therefore, the PSF of SV, 4-view MVI, and 8-view MVI were calculated and compared. The theoretical PSF of an SV was generated with the Huygens software. It was calculated based on the parameters of a typical TPLSM objective of NA:1.0, water immersion, at 820 nm excitation. The MVI PSF was calculated by rotating the PSF of the SV with steps of 90° for the 4-view MVI and 45° for the 8-view MVI. The individual rotated PSFs of each view were added to create the MVI PSF for 4-and 8-view MVI. The resulting PSFs can be seen in Fig. 2.
The PSF of the SV (Fig. 2a) appears as an ellipsoid. The minimum resolution of a SV is found in the lateral (XY) axis and is 320 ± 13 nm, while the maximum in the Z axis (XZ or YZ) is 1183 ± 49 nm, in good agreement with the theoretical estimation 1 . The yellow dotted line in Fig. 2A indicates the Full width at half maximum (FWHM) in all directions around the centre. The average resolution was calculated as the effective diameter of the area formed by the FWHM, and it was found to be 624 ± 6 nm. Similar calculations were made for the 4-view MVI PSF in Fig. 2b. The PSF is maximum on the same axis as in the SV and was found to be 545 ± 23 nm, which is signi cantly smaller compared to that in SV. The axis where the PSF becomes minimum is located at 45°, in reference to the axis of maximum diameter of the PSF, and it was 440 ± 18 nm; slightly bigger compared to that of the SV. The average PSF diameter based on the FWHM was 485 ± 5 nm which is smaller compared to that observed in the SV. For 8-view MVI (Fig. 2c) the maximum PSF diameter was 480 ± 20 nm, minimum 480 ± 20 nm, and average 480 ± 5 nm.
Anisotropy in resolution, de ned as the ratio of R min over R max is 0.27 for SV, 0.80 for 4-view, and 1.00 for 8-view, thus in 8-view MVI con guration, isotropic resolution is achieved.
Comparing the average MVI resolution of 8-, and 4-views to the SV, it becomes evident that MVI produces PSFs with smaller average dimensions, therefore offering better overall resolution. This is accomplished because the resolution is more isotropic compared to that in SV. The maximum and minimum PSF diameter tend to equalize with higher view number. While the absolute minimum in resolution can be achieved with SV in the XY plane, MVI will improve the average resolution. The 8-view MVI does offer the best average resolution (480 nm), however the improvement compared to 4-view MVI is very small.
Considering that the 8-view requires twice the acquisition time compared with 4-view, it was concluded that often 4-view MVI can offer a good compromise between speed and average resolution. Whether that is enough depends on the sample and how visible are all the details of the sample in all views.

Multiview deconvolution with 4 views
A nacre zebra sh, expressing GFP on the pain sensory neurons, was used for 4-view MVD. The entire eld of view (738 × 738 µm 2 ) of the 20X, NA: 1.0 objective with a voxel size of 721 × 721 × 1482 nm 3 was used. To decrease acquisition time a 1024 × 1024-pixel format was used, although formats up to 8192 pixels were possible. Four views at 90° rotation step were acquired with a total acquisition time of 25 minutes. Excitation was at 820 nm and all three channels were acquired simultaneously. In the blue channel, SHG from muscle bres (from the body), and collagen (mainly from the ns) were recorded. In the green channel, the signal of GFP was detected, as also a signi cant amount of auto uorescence. In the red channel, nuclei stained with SYTOX orange and the uorescent beads, which were used as ducial markers, were visualized. The nuclear stain did not penetrate in the body so only cells on the skin were stained, which gave a clear view of the larva's external boundaries.
In Fig. 3 the cross sections of the individual registered views, the fusion of the registered views (MVI), and the MVD result are presented. First step in the analysis is the registration of each individual view. This was accomplished by using as ducial markers the uorescent beads embedded in the sample. Using the Multiview Reconstruction plugin of FIJI 8 these ducial markers were identi ed in each view. Subsequently, each view was rotated to t the reference frame of the rst view. In Fig. 3 all views were registered to View 1 and from here on it is referred to as Single View (SV) and is used as comparison with the MVI and MVD images. In each view the direction of the z-stack is indicated with the z axis sign. The registered views were fused into one image (Fig. 3E). While in each individual view only a part of the sample could be seen, in the fused image clearly the entirety of the sample became visible with all the details from all the views. The fused image constitutes the MVI image. However, because of the elongated PSF on the z axis of each view, the MVI image appeared blurrier than each individual view. To remove this effect, deconvolution was applied. In this process, the PSF of each view, extracted from the uorescent beads of each view, were used to perform the deconvolution using the Multiview Reconstruction plugin of FIJI. The result is presented in Fig. 3F. Clearly, contrast in the MVD image is higher compared to the MVI image, while the entirety of the sample remains visible.
In Fig. 4 the 3D reconstruction of the SV and MVD images are presented. In Fig. 4A the SV is seen from the XY perspective. The outer structure of the sh is visible. Nuclei on the skin are seen in red, auto uorescence from the yolk sac and myotomes are seen in green, and muscles in blue. In the corresponding MVD image (Fig. 4D) the same structures are visible, however they appear crisper. Nuclei appear smaller because they are less blurry with more well-de ned borders. Also, lipids in the yolk sac and myotomes appear crisper because resolution is better, and noise is reduced. What becomes also better visible are cells near the skin surface, that were faintly visible on the SV (arrows). This is because structures with dim signal are enhanced in MVD. The nature of these cells is not known as there is no speci c stain, but since they have a dendritic like structure, they could be cells of the immune system.
Muscle bres also look clearer as also collagen in the ns (diamond arrow) which is faintly visible in the SV.
In Figs. 4B and 4E the 3D images are rotated 90° to the right and the sagittal view (XZ) is visualized. In the SV image only part of the sample is visible since penetration depth was restricted (~ 120 µm). In the MVD image all parts of the sample are visible with good contrast. From this viewing angle the GFP signal can be discerned. Spectrally, it is di cult to separate the GFP signal from auto uorescence but based on Conclusively in the MVD image the entirety of the sample is visible with better resolution, and higher contrast. The improvement in resolution is further demonstrated in the supplementary gures (Figures S1, S2, and S3) where it is shown that nuclei are better resolved and, muscle bres can be discerned in the XZ and YZ axis while this is not possible in the SV image. Noise is reduced which allows imaging of faint signal of auto uorescence structures to be visualized clearly. Moreover, the bene ts of MVD are demonstrated for both uorescence and SHG signal, which can be a signi cant advantage when imaging unstained samples based only on the inherent signal. The improvement in contrast and resolution is demonstrated in 3D images in the Supplementary Video 1.

Multiview deconvolution of collagen SHG with 4 views and 8 views
In the previous section, the bene ts of 4-view MVD on axial resolution improvement and visualization of the entirety of the sample in uorescence and SHG contrast were demonstrated. In this section, further improvement in resolution in SHG contrast by increasing the number of views was investigated. For this purpose, a rat-tail tendon containing mainly parallel organized collagen bres was used.
In Fig. 5 the SV (Fig. 5A, 5B), the 4-view MVD − 90° rotation step - (Figs. 5C, 5D), and the 8-view MVD − 45°r otation step - (Fig. 5E,5F) of a rat-tail tendon based on SHG signal are presented. To exploit the full resolution capabilities of the microscope, images were acquired with high sampling density -pixel size smaller than microscope resolution. Acquisition time was in average 3 minutes for each view with a FOV of 105 × 105 µm 2 and voxel size of 103 × 103 × 494 nm. The individual registered views are presented in Fig. 5G.
In the lateral XY plane of the SV image (Fig. 5A), collagen bres appear well de ned and continuous. In the 4-view (Fig. 5C), and the 8-view (Fig. 5E) MVD images, collagen brils do not appear as continuous as in the SV. This impression is created because brils are three-dimensional, and their axis does not exactly coincide with the XY imaging plane. In reality, the brils cross the imaging plane, but because of the poor axial resolution in the SV they appear to be along the XY imaging level. On the other hand, in the MVD images, brils appear to come in and out of the imaging plane, therefore they do not appear as continuous as in the single view.
In the sagittal (YZ plane) sections of the SV (Fig. 5B), the SHG signal from brils overlap creating blurred elongated structures where individual brils are not clearly resolvable. In the MVD sections (Figs. 5D, and 5F) individual bres become visible. In the 4-view MVD, a striping effect is visible. This striping has the direction of the z-axis of each individual view and is due to the incomplete removal of the PSF elongation after deconvolution caused by strong local constructive interference effects observed in such tissues 25,26 . The addition of more views minimizes this effect, and individual brils become better visible (Fig. 5F). On the magni ed inset of Figs. 5B, 5D, and 5E a single bre which is isolated from other bres is analysed to demonstrate resolution improvement. While in the SV image contrast is low, in the MVD images contrast is much higher. In the SV image the bre appears elongated in the direction of the Z-axis, in the MVD images the bre appears rounder. The corresponding plot pro le is presented in Fig. 5H. The apparent diameter of the bre in the Z-axis in the SV is 2112 ± 152 nm, in the 4-view MVD is 520 ± 42, and 461 ± 35 in the 8-view MVD.
In Fig. 5I comparison of the normalized intensity pro le of the dotted line in Figs. 5B, 5D, and 5F are presented. In the SV only a single structure is visible, while in the MVD images more peaks are present indicating the presence of more brils.
To better demonstrate the advantages of MVD the 3D images are presented in Fig. 6. The SV image appears very blurry, especially in the Z-axis. In the 8-view MVD image many more brils can be resolved and gives a clearer visualization of the tendon. Quanti cation of the number of brils based on the 8-view MVD can yield more accurate results. The animation of the 3D reconstruction of Fig. 5 are provided in Supplementary Video 2.

MVD on Zebra sh heart
In the previous section the strong SHG signal from collagen was imaged. In this section the advantages of using the MVD method to image the faint SHG signal from myosin of a 3-dpf zebra sh heart was explored. The signal from myosin brils of the developing heart was low, which resulted to very low SNR and noisy images (Fig. 7C). For this reason, 8-view MVD setting to capture most of the detail of this faint signal was used. To avoid sample photodamage a relatively large voxel size was used (206 × 206 × 1280 nm). To increase the detected SHG, laser power was increased compared to normal uorescence imaging. This caused some auto uorescence cross talk in the SHG channel, however this signal was well below the SHG levels. The results are given in Fig. 7. In the SV image, in the XY imaging plane, (Fig. 7A) some myo brils are clearly visible; however, the image has low contrast and structures inside the heart are not well de ned. In the corresponding MVD image (Fig. 7B) myo brils appear well de ned and with high contrast. In the sagittal (YZ) plane, the SV image (Fig. 7C) signal in the upper layers of the image is high, while it is reduced at deeper layers (arrow indicates the direction of the z-stack). Also, the structures visible in the upper layers are blurry because of the low resolution in the axial dimension. On the corresponding MVD image (Fig. 7D) noise is reduced and structures are visible with comparable contrast in all depths of the sample. Individual brils are clearly visible throughout the sample with high contrast, therefore MVD provided a detailed image of a sample with faint signal, that was not attainable with other means.
The bene t of performing MVD is illustrated even better by the 3D reconstruction images (Fig. 7E-7H). In the MVD images (Figs. 7F, and 7H) the entirety of the heart is visible, as are also individual cardiac myo brils. In contrast, the SV images (Figs. 7E and 7G) are very blurry and contrast at the deeper layers is low. The Atrium, Ventricle, and Bulbus arteriosus of the heart are clearly visible in Fig. 7H. The animation of the 3D reconstructions of Fig. 7 are provided in Supplementary Video 3. Interestingly, the structure of the developing heart is very different to that of a mature heart. Cardiac muscle bres are not so dense, and appear to be crossing in between, but do not have a precise directional order as is seen in mature hearts. In the developed heart the entire heart wall is expected to be covered by muscle bres. Imaging of the intrinsic signal of the heart is particularly interesting when investigating cardiomyocyte morphology in cardiomyopathies and genetic diseases such as laminopathies, especially in developing model organisms such as the zebra sh embryo.

Discussion
In this study the feasibility and bene ts of applying MVD on conventional TPLSM was demonstrated. This was accomplished by developing a custom rotation chamber where the sample could be mounted and rotated on a horizontal con guration. Registration, fusion, and deconvolution of the MVD image was performed based on the general public licence plugin Multiview Reconstruction 8 of FIJI. MVD imaging, clearly demonstrated improvement in resolution as well as imaging the entirety of the sample. Additionally, low signal structures became better visible in the MVD image ( Figures S1, S2, and 7). Improvement in resolution was demonstrated in both uorescence and SHG signal. This method could be applicable to small organs or tissue sections, bigger cleared organs or tissue sections, and small organisms such as zebra sh embryos.
The PSF of SHG depends strongly on the local sample structure making its de nitions challenging, and for this reason deconvolution applications have been limited 27 . SHG is a coherent phenomenon, emission is not isotropic as uorescence, and is mainly in the direction of the excitation beam. SHG signal depends on the organization of the emitters (the harmonophores), the intensity and polarization of the excitation light, and the orientation of the harmonophores relative to the excitation beam 16,28 . As a consequence, resolution in SHG imaging cannot be described accurately by a single PSF. However, since the PSF of an imaging system is the product of the excitation PSF and the emission PSF, it was chosen to use the same PSF for SHG as for uorescence, since they have common excitation PSF. Based in this approximation a signi cant improvement in resolution was demonstrated (Fig. 5, SV: 2112 nm, MVD: 461 nm).
In the tissue level where multiple brils in collagen, and multiple myo brils in muscle, bundle together in bres, it is di cult to separate them in SV. In MVD, these structures were clearly discerned in the axial dimension. In the SV, constructive interference between emitters can enhance the signal between brils making their visual separation impossible. An example of such interference phenomena is the Vernier like artifacts reported for muscle bres 29,30 and the elongated collagen brils seen in the Z-axis (Figs. 5, 6). These interference phenomena are view speci c, and usually are present in one view but not the others. When the MVD image is formed the view speci c interference artifacts tend to be outweighed by the inherent signal, which is present in every view, and therefore lead to a clearer image. In MVD such artifacts are greatly suppressed and allow for clear visualization of the myo brils ( Figures S1 and S2).
Such interference phenomena are also present in the 4-view MVD (striping effect), therefore increasing the number of views signi cantly supressed them, as was shown in the 8-views MVD (Fig. 5).
Interferometric second harmonic generation microscopy 25,26 has been proposed to eliminate interference artifacts with very promising results. Indeed, such artifacts are completely suppressed offering a clear view of the inherent SHG signal, although since this technique is very sensitive it is limited to thin tissue sections.
Regarding the number of views in MVD, in this study only 4 and 8 views were explored, but any other combination is possible. The number of views depends on the resolution improvement required but also on the sample optical properties. In theory close to isotropic resolution can be achieved with only 4 views but for thicker samples not all details are captured equally in every view. In a study based on SHG signal MVI, it was shown that a intertwining collagen bre was visualized more clearly using 10 views 14 , whereas for LSFM MVI up to 24 views might be necessary to capture all the details of a large sample and to overcome local distortion due to scattering and refraction within the sample 31 11,36 . With a similar con guration implemented for TPLSM, MVI acquisition would require the same time as a single view. However, this would also add limitations to sample size and positioning of the objectives. Another possibility to increase the acquisition speed would be to use multi point excitation 37,38 . In this case, a single beam can be split up to 64 beams that simultaneously scan different parts of the eld of view, resulting to up to 64 times faster scanning of a single image. This would signi cantly increase the acquisition time of each individual view and rotation of the sample and size would not be restricted. However, it should be taken into consideration that axial resolution and penetration depth of multibeam excitation can be inferior to that of the single beam 39 . Nonetheless, the speed increase could be a signi cant advantage for samples up to 200 µm thick, such as zebra sh larvae, and could be a promising alternative. Therefore, even though TPLSM MVD is still relatively slow as demonstrated in this study, it can potentially become much faster through microscope modi cations, maybe even fast enough to image dynamic phenomena with temporal resolution of some seconds.