6.1.1.

Photon scattering and absorption

Photon penetration depth is determined by both scattering and absorption within media. The intensity of light of wavelength $\lambda$ at a depth $z$ within the tissue is defined as

Fig. 6

Scattering and absorption data for rat heart tissue. (a) Absorption and reduced scattering coefficients for visible wavelengths in the rat heart (adapted from Ref. 122); (b) effective attenuation coefficients for visible wavelengths in rat heart tissue (adapted from Ref. 122) and diluted blood at 100% ${\mathrm{O}}_2$ saturation (adapted from Ref. 123).

Because of the large attenuation of visible wavelengths within the cardiac tissue, imaging at the endocardial surface and subsurface and optical access to the conduction system of the heart cannot be performed from outside the heart. Access to the epicardial surface is impractical for routine clinical procedures, as it requires invasive, open-chest surgery to expose the heart. Thus, endoscopic approaches have become standard for clinical cardiac imaging procedures.

6.1.2.

Vasculature and size constraints for endoscopy

To optically actuate and record from the endocardial surface, the imaging system must be small enough to be used in cardiac catheterization procedures. As discussed previously, the conduction system is most easily accessed from the right atrium, where the catheter enters through either the SVC or IVC. Although the data above are for humans, allometric (power-law) scaling provides a convenient way to relate cardiovascular parameters in various animals. In mammals, the major blood vessels and their branches (but not capillaries) obey allometric laws. The relationship between some physiological dimension $y$ and body mass $M$ in kg is given by

Table 3

Allometric (power-law) scaling of vasculature124 and calculated dimensions for relevant vasculature (in centimeter) for the mouse, rat, canine, goat, and human. AA—ascending aorta; CA—coronary artery; BA—brachiocephalic artery; SVC—superior vena cava; IVC—inferior vena cava.

6.1.3.

Motion artifacts

Mechanical contractions of the heart pose a particular challenge to in vivo cardiac imaging by creating motion artifacts in the optically measured signal. In neuroscience, optical systems can be physically mounted to the subject’s skull, allowing the subject to freely move while keeping the detection system relatively stationary with respect to brain structures. However, no such rigid structure surrounds the heart, and the heart itself would still be constantly moving. For in vitro optical mapping, motion uncouplers, such as 2,3-butanedione monoxime^9,13,14 and Blebbistatin,^46–48 are used to reduce motion artifacts; however, their effect on the electrical properties of the tissue makes them unsuitable for use in vivo. Ratiometric techniques and dual-wavelength subtraction provide a drug-free alternative to filtering motion artifacts, with the latter of the two having already been used for fiber-optic cardiac imaging.^15,45,49 However, both these options require the system to accommodate an additional (fourth) wavelength.

A practical approach involves looking at clinical techniques used to avoid complications due to heart wall motion for chronically implanted pacing leads as inspiration for an endoscopic imaging system. For example, long-term treatment of bradycardia requires flexible pacing electrodes at the endocardial surface within the right atrium. To prevent the dislodgment, the electrode is physically fixed to the endocardial surface via passive or active fixation methods.¹²⁶ Passive fixation is the most commonly used, where flexible tines or fins entrap the lead within the trabeculae; however, these leads are susceptible to fibrotic encapsulation, making them almost impossible to remove.¹²⁶ Active fixation provides a more suitable alternative for endoscopic imaging, where a corkscrew tip is used to directly screw the lead into the endocardial wall.¹²⁶ This provides much more flexibility in lead placement within the right atrium and ventricle and allows for the device to be removed if necessary.¹²⁶ Additional stabilization techniques can involve the combination of the microendoscopic guide with a deployable stent-like structure to keep in place.

6.1.4.

Combined optical recording and actuation: probes and spectral considerations

Cardiac optical mapping achieves superior spatial resolution compared with electrodes by employing fluorescent probes. Probe selection is dependent on both response speed and signal strength to perform the high-spatiotemporal resolution imaging on the millisecond timescale. The two key classes of fast response indicators are voltage (potentiometric) and calcium probes; examples of these indicators can be seen in Table 1. For in vitro voltage sensing, the styryl dyes di-4-ANEPPS, di-8-ANEPPS, and RH-237 undergoing spectral shifts that vary linearly with voltage changes have been used extensively in optical mapping^{5–7,30,31,35,38,40,127,128} and other cardiac imaging systems.^{12–15,44,46–48} However, these potentiometric dyes are inherently inefficient. For example, di-4-ANEPPS has a quantum efficiency estimated to be on the order of 0.3^129,130 and is excited at irradiances as low as 0.1 up to ${5\mathrm{mW}/\mathrm{mm}}^2$ before light-induced tissue heating occurs¹²⁸ resulting in, at best, a 10% relative change in fluorescence per 100 mV.⁵ More recently, new near-infrared dyes, such as di-4-ANBDQPQ and di-4-ANBDQBS, have been used in blood-perfused hearts as their absorption peak is $>70\mathrm{nm}$ higher than that of hemoglobin.^5,131

The cycling of intracellular calcium is also of interest for understanding cardiac electrophysiology and is an essential part of excitation-contraction coupling within cardiomyocytes.^5,6 Unlike potentiometric dyes, commonly used calcium dyes, such as Fluo-4, Fluo-3, and Fura-2, have much stronger optical signals, thus imaging is typically not limited by the efficiency of these dyes; however, these probes do not capture the action potential itself, and certain probes, such as Fluo-4 and Fura-2, are known to prolong calcium transients.^5–7 Fluo-3, Fluo-4, Rhod-2, and Rhod-4 are commonly used dyes excited by blue-green wavelengths that can be employed in conjunction with voltage-sensitive dyes RH-237 and the red-shifted Di-4-ANBDQPQ. The UV-excited Fura-2 has also been used, including in conjunction with Di-4-ANBDQPQ.³²

For in vivo use, genetically encoded indicators with excitation and absorption spectra suitable for imaging through blood will be necessary. For in vivo calcium imaging, genetically encoded GCaMP variants, which use GFP as the fluorescent reporter, are commonly used. The most recent variant, GCaMP6, shows great promise for speed and high SNR.¹³² The creation of genetically encoded, fast-voltage sensors has been challenging, yet some successful probes have recently been generated such as voltage-sensitive fluorescent proteins (VSFPs). The VSFP variant VSFP3 (based on a phosphatase voltage-sensing domain) includes a promising indicator based on mKate as the reporter, though speed remains an issue.¹³² Recently, genetically encoded opsins, such as mutants of Arch, have been adapted as voltage sensors.¹³³ The Arch-derived sensor Arch(D95H) has been shown to have the fluorescence response variations in both amplitude and time course due to voltage change.^134,135 In addition, the Arch variants Arch-EEN and Arch-EEQ have also been demonstrated as voltage sensors with faster kinetics and larger signal intensities compared with Arch(D95H).¹³⁶

The key challenge to combine all actuation and sensing is cleanly separating wavelengths for optogenetic stimulation and fluorescent excitation and emission with sufficient resolution. Ideally, the actuators should be chosen such that their absorption spectrum lies either above or below the excitation and emission spectra of the probe. Spectral separation can be further augmented by careful selection of excitation and emission filters and dichroic mirrors. In cases where the absorption spectrum of the actuator does not lie below both spectra of the sensor, a notch or multiedge rather than a long-pass dichroic mirror can be used to separate the emitted fluorescence from the combined light sources, without requiring major modifications to the optical system. Other spectral issues, not as easily addressed by simple filtering, may arise. For example, as seen in Table 1, ChR2 and Rhod-4 are compatible for use in the same optical system and have been used in combined actuation and sensing in cardiac tissue.^85,137 However, ChR2 has an intermediate deactivating green state at 520 nm after activation with 480 nm,¹³⁸ and, therefore, this combination may not be ideal. Some actuators are fundamentally incompatible with commonly used sensors, where their absorption spectra either overlap or lie in between the excitation and emission spectra of the sensor. For example, the spectra of hyperpolarizing opsins, such as NpHR, lie between the excitation and emission spectra of commonly used sensors such as Di-8 and Rhod-4. Thus, shorter wavelength sensors, such as Fura-2, have been used to monitor the modulation of neurons in neonatal mouse cortical slices by both hyperpolarizing NpHR and depolarizing ChR2.⁵⁴ From the more relevant for in vivo use genetically encoded sensors, the calcium sensor G-CaMP has been used to map functional connections in ChR2-expressing neurons of Caenorhabditis elegans, as a demonstration of an all-optical approach.¹³⁹ Very recent work also shows the possibility to use genetically encoded voltage sensors in vivo.

In addition to spectral overlap of optical actuators and sensors, spectral overlap of the probes and the absorption peaks of both tissue and blood must also be considered to optimize the light delivery and detection. As can be seen in Fig. 6(b), although scattering increases for shorter wavelengths, the effective attenuation coefficient of light in tissue mimics that of diluted blood across visible wavelengths, peaking at 550 nm. Thus, for optimal penetration depth, red-shifted sensors must be used. Two-photon illumination, which allows for red-shifted light to be used for excitation, can also be used. However, this method requires higher irradiances, which can lead to tissue heating.⁷⁷

6.2.1.

Optical fibers

Optical fibers are commonly used in both microendoscopy and optogenetics in neuroscience to remotely deliver and collect light deep within the tissue. Fiber selection is dependent on the choice of wavelengths, the diameter of the optical system, and the choice of collection optics; the NA of the fiber must be matched to that of the optics in order to maximize the collection efficiency of the system. Step-index fibers are the most commonly used, where light is guided using total internal reflection through a core, surrounded by a lower index of refraction cladding. The diameter of the core determines the wavelength ranges that can be efficiently coupled, the number of spatial modes that can be transmitted, and the NA. For visible wavelengths, step-index single-mode fibers have core diameters between 3 and $6\mu \mathrm{m}$ and NAs up to 0.13 NA, while multimode fibers have larger core diameters ($10\mu \mathrm{m}$ to 1 mm) giving them larger NAs (0.1 to 0.48 NA). Step-index fibers, however, are only able to transmit the intensity information and are subjected to intermodal dispersion due to different modes propagating at different velocities through the core.

In order to accommodate three wavelengths (excitation, stimulation, and emission), maximize the light collection efficiency and perform the multipoint imaging, a bundle of multimode optical fibers would be the ideal choice for an all-optical actuation system. FOIBs are commonly used in multipoint imaging microendoscopic systems in neuroscience including for in vivo optical mapping. The resolution of the FOIB is limited by both the core-to-core spacing of the individual bundles as well as the NA of the fiber bundle. The IGN-08/30 Sumitomo FOIB has been used in multiple microendoscopic systems^{97,98,102,140} including for ${\mathrm{Ca}}^{2+}$ calcium imaging in freely behaving mice.¹⁰² The 0.35-NA jacketed bundle has an outer diameter of $<1\mathrm{mm}$ and consists of 30,000 individual fibers with a core-to-core spacing of $\sim 4\mu \mathrm{m}$, core diameter of $\sim 2.4\mu \mathrm{m}$, and bending radius of 40 mm,¹⁰² making it a potentially suitable option for in vivo cardiac approaches. The bending radius may be too large for rodent subjects, but standard optical fibers have bending radii ranging from 10 to 40 mm.

6.2.2.

Optics

Due to photon scattering and absorption and low-quantum efficiencies of currently available fluorescent reporters, the optical system must collect as much light as possible. The effective NA of the microendoscopic objective dictates both the spatial resolution and the light-gathering ability—the latter directly affecting the limits of the temporal resolution of the optical system. Although traditional microscope objectives have been used for fiber-based systems,^{88–92,140–142} even small diameter, custom-made objectives have been too bulky for most in vivo applications and only provide a modest light-gathering ability ($<0.35\mathrm{NA}$).

The collection efficiency of an optical system is dependent on the solid angle $\mathrm{\Omega }$ subtended by the collection optics and is related to the NA of the system by

Optical fibers alone typically have small collection efficiencies; for example, multimode optical fibers with NAs between 0.1 and 0.48 only have collection efficiencies of 0.0025 to 0.061. However, using fiber-coupling or collimation systems designed to have NAs between 0.5 and 0.8, collection efficiencies between 0.07 and 0.2 can be achieved.

As discussed above, small diameter microendoscopic imaging has benefited from the recent development of GRIN optics, which range from 0.35 to 1 mm in diameter and have NAs as high as 0.65. GRIN lenses are used as singlet, doublet, or triplet systems and can be combined with plano-convex lenses.⁹³ Typically, doublet systems are used for higher-NA systems, whereas triplet systems employ an extra relay lens to lengthen the objective for deeper imaging. GRIN lenses used in conjunction with plano-convex microlenses are best suited for high-resolution systems, where the GRIN objectives can be designed to compensate for aberrations from the plano-convex lens.⁹³

Selection of a GRIN objective is determined by the FOV of the optical system, which is usually limited by the diameter of the optical system. Wider GRIN objectives have larger FOVs and NAs, but this is commonly accompanied by longer WDs. For the single-photon microendoscopic systems used in neuroscience, the WD is limited to approximately 0.1 mm from the endoscopy tip, as at longer distances photon scattering and absorption greatly decrease the SNR.⁹³ In GRIN doublet objectives, the relay lens is selected based on the NA of the collection optics; the NA must be equal to or smaller than that of the optical fibers or FOIBs to ensure an efficient coupling. GRIN doublet systems have been used with FOIBs for microendoscopic imaging in vivo^98,102 including performing video-rate ${\mathrm{Ca}}^{2+}$ recordings using an EMCCD camera in freely moving mice.¹⁰² However, a large portion of the imaging system was mounted outside the animal’s skull, allowing larger diameter optics to be employed: a 2 mm diameter coupling lens, a 2 mm diameter focusing lens, and a 1 mm objective with 0.47 NA.¹⁰² A similar scanning microendoscope system consisting of 1 mm diameter optics, a relay lens of 0.2 NA and an objective lens of 0.5 NA, to couple into a 0.35 NA FOIB will be a more suitable model for cardiac use.⁹⁸

6.2.3.

Illumination

High quality light sources for both excitation of fluorescent dyes and optical actuation are necessary to produce an effective cardiac optogenetic imaging and actuation system. When selecting a light source, the key features that must be considered include the source’s spectrum, power, stability and controllability (maximum frequency the source can be pulsed for optical actuation). For optical actuation of cardiac tissue, the source must be capable of variable intensity control that can be pulsed or switched at speeds relevant for cardiac pacing, typically 0.1 to $5\mathrm{mW}/{\mathrm{mm}}^2$ irradiances at the site of actuation and 0.5–12 Hz pacing frequencies. The required optical power for imaging must be sufficient to excite the fluorescent probe but without causing tissue heating. For example, di-4-ANEPPS and di-4-ANBDQBS (excited by 535 nm and 660 nm, respectively) show little difference in percent of fluorescence change during an action potential for irradiances varied from 0.1 to ${5\mathrm{mW}/\mathrm{mm}}^2$; however, at higher irradiances, tissue heating can occur thereby altering the measured action potentials.¹²⁸ Other practical considerations include the power consumption of the light source, whether or not cooling units are needed, the size of the device, and how easily the light source can be coupled into the system.

Advances in solid-state technology have led to the availability of cost-effective LED-based light sources that are a more practical alternative to traditionally used broad-spectrum arc lamps, as they can be pulsed using frequencies well over 1 kHz using digital pulses (e.g., TTL pulses, which act as a binary on/off switch) or analog modulation. Although they can be purchased as fiber-coupled units at specific wavelengths, they are incoherent sources (typically having bandwidths of 10 nm) causing considerable intensity loss when coupled into optical fibers. However, new LED technologies show promise for miniaturized systems that do not require the light sources to be fiber coupled. For in vivo use, new semiconductor fabrication techniques have allowed for the fabrication of $\mu$LEDs ($1000\times 600\times 200\mu {\mathrm{m}}^3$) for miniaturized probes that can be directly inserted into tissue.⁷³ Also, recently flexible, stretchable $\mu$LED arrays have been manufactured and used in balloon-based catheters; these sheets could also be placed on the heart in order to provide on-site illumination.¹⁴³

For applications requiring coherent, high-intensity light, lasers are preferred, as they can be tightly focused and efficiently coupled into optical fibers. They are more expensive than LED-based options and not all laser types can be as reliably pulsed using digital or analog modulation. Of particular interest are solid state lasers, including direct diode lasers, diode pumped solid state (DPSS) lasers, and optically pumped semiconductor lasers, available in a range of wavelengths with narrow bandwidths ($<1\mathrm{nm}$). Certain laser types, such as DPSS lasers, are known to have unpredictable power stabilities when modulated with TTL pulses, complicating optogenetic stimulation, but they are still attractive for fluorescence excitation which requires continuous illumination.

6.2.4.

Detectors

Small diameter optical systems have limited collection efficiencies, thus special consideration must be made when choosing a detector, including the detector’s quantum efficiency at the wavelengths of the fluorescent reporter being used, active area compared with the system’s aperture size, and SNR at the needed temporal resolution. For optical mapping, acquisition rates must be at least 200 Hz,⁶ and as red-shifted fluorescent probes are more desirable due to less scattering and absorption at longer wavelengths, detectors with quantum efficiency peaks in the red are preferred.

Single-point detectors are commonly used in both single-point imaging as well as multipoint imaging employing scanning systems or detector arrays. PMTs are among the most commonly used detectors in fluorescence detection and microendoscopy, as they have high gain (typically on the order of ${10}^6$) and relatively low noise. Although they can be purchased with relatively large detector areas, they require high voltages to stably operate and their quantum efficiency is poor in the near-IR range ($<20\%$). Solid state detectors, such as avalanche photodiodes (APDs) have higher quantum efficiencies in the longer wavelengths (50 to 90%). APDs perform like a solid-state version of a PMT, where gains on the order of 1000 can be achieved; however, they have much smaller active areas compared with PMTs. Single-photon counting modules are a specific subset of APDs that can be used for detecting extremely small signals, including single-photon detection. Although these detectors are able to measure low-intensity signals with high-quantum efficiency in the near-IR ($>60\%$), they have very small active diameters, typically less than $200\mu \mathrm{m}$.

High-resolution CCD cameras and EMCCD cameras have been popular for use in optical mapping^{5–7,30–34,37–40,127,128,144} and microendoscopy due to their high-spatiotemporal resolution. EMCCD cameras take advantage of avalanche processes to perform the on-chip electron multiplication to enhance SNRs, have higher frame rates, and improved quantum efficiency.⁵ Current EMCCD models, such as the Andor iXon Ultra 897, have $512\times 512\text{pixels}$ and can achieve full-frame rates of 56 frames per second; this rate be increased to above 100 frames per second by employing crop modes. EMCCD cameras have been used in optical mapping systems and in microendoscopic systems, including for recording video rate ${\mathrm{Ca}}^{2+}$ imaging in freely moving mice using a FOIB and a GRIN objectives.¹⁰² In addition to using fiber-based systems, microendoscope probes have also incorporated miniaturized cameras mounted directly on to the subject. Recently, miniaturized CMOS cameras have been incorporated into miniature, head-mounted microendoscopes for imaging calcium transients in behaving mice.⁹⁹ Since spatial mapping is of interest in cardiac applications, EMCCD and newer CMOS detectors with suitable characteristics are most desirable.