Open Access
Add to BibSonomy
Abstract
In biological research, rapid wide-field fluorescence lifetime imaging has become an important imaging tool. However, the biological samples with weak fluorescence signals and lower sensitivity often suffer from very low precision in lifetime determinations which restricts its widespread utilization in many bioimaging applications. To address this issue, a method is presented in this paper to substantially enhance the precision of rapid lifetime determination (RLD). It expedites the discrimination of fluorescence lifetimes, even for the weak signals coming from the cells, stained with long-lived biocompatible AIS/ZnS QDs. The proposed method works in two phases. The first phase deals with the systematic noise analysis based on the signal and contrast of the images in a time-gated imaging system, wherein acquiring the high-quality imaging data through optimization of hardware parameters improves the overall system performance. In the second phase, the chosen images are treated using total variation denoising method combined with the Max/Min filtering method for extracting the region of interest to reconstruct the intensity images for RLD. We performed several experiments on live cells to demonstrate the improvements in imaging performance by the systematic optimizations and data treatment. Obtained results demonstrated a great enhancement in signal-to-noise and contrast-to-noise ratios beside witnessing an obvious improvement in RLD for weak signals. This approach can be used not only to improve the quality of time-gated imaging data but also for efficient fluorescence lifetime imaging of live biological samples without compromising imaging speed and light exposure.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Fluorescence imaging has become an indispensable tool in modern biomedical research due to its highly sensitive real-time versatile imaging credentials. Essentially, in common fluorescence imaging experiments the spatial variation of collected emission signals within specific wavelength bands represents the characteristics of fluorescently labeled targets [1]. Virtually, the fluorescence imaging technique can be implemented using any fluorescent molecule of interest with known spectral profiles and organelle-specific targeting aptitude, without having much interference from the molecular function. However, the possibility of achieving the best possible spatial resolution is still constrained by the diffraction limit [2,3]. Nevertheless, it is a powerful imaging tool for diverse important biological applications, such as DNA sequencing, tissues/cells imaging, and in vivo imaging studies intended for pre-clinical studies of diseases and monitoring therapeutic advancements [4–6]. In practical bioimaging applications, in spite of having tremendous potential, the fluorescence imaging techniques encounter a major limitation in terms of the reasonable interference from background signals, arisen out of the strong autofluorescence of biological samples in complex biological environments [7]. In this context, fluorescence lifetime-based imaging has been coined for various interference-free imaging practices.
Fluorescence lifetime-based imaging is usually obtained using the time correlated single photon counting [8], or frequency-domain method [9]. Yet, it is not suitable for real-time analysis as more sophisticated highly sensitive instrumentations are required to distinguish the tiny differences in lifetimes. Besides standard techniques, time-gated technique [1], can be employed to obtain the fluorescence lifetime imaging wherein a pulsed laser source is used to excite the samples, followed by capturing consecutive images of the fluorescence decay at several delays by a gated camera. Time-gated lifetime imaging based on intensifier charge-coupled device (ICCD) sensor can achieve a high frame rate and capture all parallel pixels of the image to enable rapid lifetime determination (RLD) of the regions of interest with a wide field-of-view. However, the inability of RLD to accurately resolve the spectrally overlapping multiple lifetimes within a single pixel and the obtained low-accuracy for the pixels with much shorter or longer lifetime values than the mean value, are the major concerns for utilizing the wide-field RLD technique.
To overcome these limitations, several strategies have been implemented including the use of two-photon method, development of near and short infrared emitting probes and upconversion and down-conversion nanoparticles [10,11]. In fact, the time-gated acquisition of fluorescence emission combined with long-lived fluorescent probes might be helpful to curb the contribution of autofluorescence to reduce the noise [12,13]. The use of time-gated technique in bioimaging for temporal assessment has been widely reported in literature. For instances, time-resolved molecular imaging [14] and time-gated fluorescence lifetime imaging [15] have been frequently employed. The RLD is convenient for in vivo lifetime mapping, especially for a large field of view (FOV) [16].
Semiconductor quantum dots, such as AgInS2 (AIS) & CuInS2 & CdSe, can exhibit some outstanding optical properties, including large absorption coefficient, tunable color emission, and long-lived photoluminescence [17–20]. In particular, silver-indium-sulfide quantum dots (AIS) as well as its ZnS coated core-shell structure (AIS@ZnS) have demonstrated great potential in fluorescent bioimaging and bio-sensing applications [21–24]. Particularly, in biomedical applications, AIS QDs have emerged as an attractive alternative compared to the other QDs, which containing toxic heavy metals such as cadmium, mercury, lead, tellurium, selenium and arsenide [25]. It noteworthy that compared to AIS QDs, the AIS@ZnS core-shell structure significantly increases the recombination sites, photoluminescence quantum yield and stability of the QDs [26], which facilitates its efficient utilization in widespread in-vitro and in-vivo bioimaging studies [23,26,27].
In biological imaging, the ‘noise’ is intrinsically associated with the image quality pertaining many useful information. To evaluate the noise, it is also important to contemplate the signal and contrast of the images under consideration. Improving the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) are the essential criteria for achieving good quality images in fluorescence imaging of biological samples. Improved SNR and CNR have been achieved via different means, such as developing advanced imaging systems, new fluorescent materials or advanced denoising and filtering algorithms [28–34]. However, the collected signal can be enhanced by increasing optical configurations efficiency, the light path length through the sample, or using a broader acquisition time. In the contrast, the background can be reduced by blocking reflected and refracted excitation light and scattered light. These ways can enhance the quality of the images in fluorescence imaging, but they are not useful for samples that suffer from strong autofluorescence and emit weak signals. Therefore, there is an urgent need to improve the detection conditions and enhance new methods to overcome the limitations of the fluorescence imaging method, aiming at improving the detection sensitivity and reducing the noise.
In this context, the present study not only deals with the systematic analysis of the noise sources but also investigates various main parameters of the system to optimize the image data processing quality of the Wide-Field TGI system-based ICCD sensor. Additionally, this study includes the implementation of total variation (TV) denoising as well as Max/Min filtering models on the optimal intensity images towards minimizing the noise distribution. Importantly, the present study reveals that combining the optimal gated parameters with TV denoising and Max/Min filtering models result in the enhanced precision of RLD in time-gated technique. Therefore, these methods can be applied either individually or in combination for separate dimensions to achieve improved RLD. Using our approach, it is demonstrated that SNR, CNR and lifetime accuracy can be improved for weak signals from live cells in fluorescence lifetime studies even if the Poisson and Gaussian distributed noise, frames accumulation value and gate overlapping are rightly taken into consideration for lifetime determination procedures.
2. Materials and methods
2.1 Experimental setup
The home-built experimental set-up of TGI is shown in Fig. 1(a). The system was designed based on the fundamental principles of fluorescence microscopy. As illustrated in Fig. 1(a), a typical wide-field TGI system consists of three main parts: a pulsed source, optical components, and ICCD camera. The main instruments including, the laser source and the camera, were used in our previous work for the time-resolved analysis of induced plasma [35]. A Q-switched pulsed laser operating at 532 nm with pulse repetition of 10 Hz and pulse width of 10 ns was used for excitation. The variable metallic neutral density filter was adopted to adjust the excitation light power. The laser power was adjusted to 0.8 mW at the back of the objective. The laser beam was coupled into multimodal fiber-optics. The expanded beam with 4x-objective lens was used to collimate the light coming out of the fiber. Using a dichroic mirror, the collimated laser beam was reflected to the back aperture of the objective, focusing on the sample surface. Fluorescence was collected in the same axis of the excitation beam which was separated by a dichroic mirror from the excitation light. Two convex lenses were set in front of ICCD camera to collimate the focus of collected signal to the sensor. A long-pass filter was setup in front of the camera to reflect the excitation light. The collected fluorescence was recorded by an iSTAR ICCD, DH312T-18U-03, Andor Technology Ltd instrument. Finally, a high-performance 3D-axis linear translation stage was used with different micrometer screws to fix the focal point.
Fig. 1. (a) Experimental setup used for TGI system (a). ICCD camera diagram (b). The operating principle for time-gated method (c). Overlapping gates operation for lifetime imaging (d). Note that, M is Mirror, DM is Dichroic Mirror, L is Lens, LPF is Long-Pass Filter, NDF is Nature Density Filter, MMF is Multi-Modal Fiber, and MCP is Micro-Channel Plate.
Originally, the ICCD consists of two main parts: the image intensifier and CCD-chip. The intensifier is the main part of ICCD camera for the TGI technique which comprises three main components: a photocathode, microchannel plate (MCP), and phosphor screen as shown in Fig. 1(b). As depicted in Fig. 1(b), the main role of photocathode is to capture the incident light that comes from the input window and convert it into photoelectrons. The photoelectrons move toward the MCP channels where each electron is amplified to create secondary electrons and a high-voltage across the MCP is applied to accelerate down the process. Subsequently, the intensified photoelectrons are captured by the phosphor coating screen and convert the photoelectrons into photons. Finally, the output of the image intensifier is coupled to the CCD-chip by a fiber optic coupler. The image intensifier functions as a camera shutter by applying a high voltage between the photocathode and MCP. The system is electronically gated in such a way that it allows the flow of photoelectrons from the photocathode towards the MCP when the voltage is negative whereas the movement of photoelectrons towards the MCP is prevented if the voltage is positive. In this fashion the image intensifier effectively acts as an electro-optical gate with a very fast optical shutter “subnanosecond” time resolution.
The operating principle of the TGI method of our rationally devised system has been illustrated in Fig. 1(c) wherein the pulsed laser is used to illuminate the samples and the emission signals are recorded after disappearance of the pulse. Figure 1(d) illustrates the “steps” highlighting the positions of overlapping gate operations for RLD in four observation- gates wherein each measurement requires equal gate width.
2.2 Materials and synthesis
Reagents: AgNO3, In(CH3CO2)3, Octadecene (ODE), Toluene, Chloroform were purchased from Aladdin (China). 1-Dodecanethiol (DDT), Ethanol, 3-Mercaptopropionic acid (MPA), Zn(CH3CO2)3 and Hexane were purchased from the commercial supplier Mackun (China). Oleic acid (OA), Oleamine (OAm), Sulfur powder (S), Bovine Serum Albumin (BSA) and Tri-n-octylphosphine (TOP) were purchased from Sigma-Aldrich (America)). All reagents are of reagent grade and used without further purification.
2.2.1 Synthesis of (AIS) cores
The typical synthesis piratical was reported in [36]. At first, 0.1 mMol AgNO3 and 0.4 mMol In (CH3CO2)3 were taken in a 50 mL three-neck flask. To this mixture, 1 mL OA, 2 mL DDT and 5 mL ODE were sequentially added and the resultant mixture was heated at 80 °C under nitrogen atmosphere with occasional vacuuming. Once bubbles completely disappeared, the mixture was allowed to be heated at 130 °C and the S precursor (0.65 mMol S dissolved in 2 mL OAm under 60 °C) was injected to the mixture under that condition. AIS core solution was obtained after reaction lasted for 20 minutes. AIS cores were then purified via serval repeated precipitation/dispersion cycles using hexane and ethanol. Ultimately, AIS cores were dispersed in 4 mL toluene for further use.
2.2. Synthesis of AIS@ZnS QDs
The obtained AIS core in toluene solution was loaded in a three-neck flask. 1 mL OA, 1 mL OAm, 1 mL DDT and 3 mL ODE were sequentially added to it. Next, we repeated the process of heating and vacuuming was repeated as described above. Afterwards, a mixture of 0.4 mMol Zn(CH3CO2)3 and 0.4 mMol S precursors were dissolved in 2 mL TOP and subsequently injected into the flask within 2 minutes. The reaction lasted for 20 minutes in order to give enough time for ZnS shell to grow. The purified AIS@ZnS was obtained by doing several precipitation/dispersion cycles using hexane and ethanol. Finally, AIS@ZnS were dispersed in 8 mL chloroform for further use.
2.2.3 Modification of the quantum dots (QDs)
At first, to a solution of MPA (5 µL) and 4 mL chloroform, 1 mL of AIS@ZnS QDs solution was added with stirring for 5 minutes. Then 1 mL of ammonia was added to it. After stirring for 4 hours, the mixture turned into an aqueous phase. The obtained aqueous QDs were collected and purified with ethanol. The AIS@ZnS-MPA QDs obtained by centrifugation were dispersed in 1 mL pure water. Finally, 2 mL BSA solution (10 mg/mL) was added to the AIS@ZnS-MPA QDs solution and stirred for 30 minutes. The collected AIS@ZnS-MPA@BSA QDs materials were centrifuged for 10 minutes with 50 kDa ultrafiltration tube at the speed of 5000 rpm for 10 minutes for purification.
The typical synthesis and optical properties of these QDs include the absorption and emission lines were reported in literature [36]. Moreover, the photon absorption of these QDs is low in the green region, which emits a weak fluorescence.
2.3 Cell culture
OVCAR-3 cells were cultured in RPMI-1640 with 10% fetal bovine serum (FBS, Invitrogen) and 1% penicillin/streptomycin (Invitrogen). The culture condition was maintained in a humidified atmosphere containing 5% CO2 under 37 °C.
2.4 SNR and CNR analysis
For the time-gated technique a pulsed light source was used to excite the samples and the gated-detector was employed to record the emitted signals. The proportion intensity of the decaying fluorophores from the molecules, collected after the disappearance of laser pulse with detection gate can be simply expressed by [37–39]:
where A represents the amplitude of excited state, tD represents the delay between the excitation laser pulse and detection gate, TG is the width of time-gate and τ is fluorescence lifetime of excited state. For the time gate with gate width ${T_G} \gg \tau ,{e ^{ - ({T_G}/\tau )}}$ can be ignored.The intensity collected by time-gated technique comes from different substances, where the total intensity for single image can be represented by [40,41]:
where n is the number of fluorescence substances, IEm is the sample fluorescence, IEN and ISN represent the environment noise (Poisson noise), and the system noise (Gaussian noise), respectively.SNR and CNR are frequently used for image quality assessment, especially, to describe the performance and sensitivity of imaging systems. In a TGI system, the stability of laser pulses, the environmental noise caused by ambient light and the performance of the ICCD is important aspects which determine the SNR and CNR for the collected data.
The SNR for single pixel is defined as the mean ratio of the intensity and its standard deviation, which can be defined by the following equation:
where, Ii and std(Ii) represent the mean of intensity and the standard deviation of signal respectively, within the pixel of interest for a specified image. Usually, the mean intensity (Ii) acquired by the ICCD is given by the result (kmPt) of the multiplication of a gain function, km, by the number of photons, P, reaching the photocathode at exposure time, t.Generally, in an ICCD, the noise emanates from two sources: the intensifier and CCD-chip. One of the main noise components in image intensifier is Photocathode Noise (Npc), which consists of the photon shot noise correlated with the random influx of photons into the CCD, and Equivalent Background Illumination (EBI) which basically measures the dark current of the photocathode. Another noise component, Micro-channel Plate Noise (Fm), is an additional noise element that is generated in the MCP as a result of loss mechanisms and photoelectron multiplication statistics. The last noise component, Phosphor Screen Noise (Nc) arises from the times decay of the materials and the uncertainty in quantum efficiency of phosphor screen. However, in CCD-chip, the main components of noise are: (i) Dark noise (Nd) and (ii) Readout noise (Nr). The earlier arises from statistical variation in the number of thermally generated electrons which is highly dependent on the CCD chip temperature whereas the latter is generated within the on-chip amplifier during the readout process.
In the ICCD camera, the image intensifier amplifies the signal so that the CCD-chip sections do no high dominates the noise of the camera, especially, in shot noise and dark noise. Considering the referred main noise sources the total SNR of gated ICCD can be express as [42–44]:
In contrast to the SNR described above, the CNR can be defined as:
where Ii1 and Ii2 are the average gray levels at different regions of interest in the image, std(Is1) and std(Ii2) are the standard deviation of Ii1 and Ii2, respectively.2.4 Rapid fluorescence lifetime mapping
Time-gated enhancement of final rapid fluorescence lifetime images are processed in six steps, as illustrated in the Fig. 2. At first, the system parameters, investigated in the previous section, are optimally tuned. Next, background signal images at incident light under the same experimental conditions with that of the fluorescence images are captured. The background image signals essentially signify the instrumental and environmental background. Thereafter, time-gated intensity images are captured at different time-delay with fixed gate-width. The background is then subtracted from the intensity images to obtain the background free new intensity images Inew as given below.
where IBG is the background signal and IRN is the residual interference noise that cannot be removed by background subtraction.Fig. 2. The process diagram of the steps following for lifetime determination. The left side shows the mapping non-optimal gates and the right side shows optimal gates, with two approaches of steps for denoising and flirting treatments.
Next, denoising and filtering are applied to generate new intensity images wherein the TV denoising [45,46] is combined with Max/Min filtering [47] for acquiring new time-gated intensity images. As shown in Fig. 2, two different processes can be used to compare the enhancement in precision of lifetime mapping. Eventually, the rapid fluorescence lifetime imaging maps are basically created using four gate acquisitions by determining the lifetime of each pixel. This method is more precise to implement than the two-gate acquisition [48–51]:
Finally, completing the previous steps, a pseudo-color coding is used to show the visibility of the enhanced image, and a fluorescence lifetime image is finally obtained using two different approaches.
3. Results and discussion
3.1 Optimal gating parameters
Intensifier gain is an analogues signal amplification feature that amplifies the photoelectron by applying high voltage to the micro-channel plate across several CCD registers before readout [52]. Adjustment of the optimal intensifier gain of the ICCD camera is aimed at improving the sensitivity of the imaging system. The detection sensitivity of an ICCD can be increased to obtain a higher gain value, but it will result in the increased noise. According to Eq. (4), the SNR of the emission signal will be clearly influenced by the image intensifier gain which will reach a plateau at higher values. For bright illumination signals, the signal is much greater than the read noise wherein the photocathode noise is dominated in the total noise. Therefore, when the signal is high enough to subdue the read noise Nr, it is unnecessary to apply a higher gain. However, for dark and super-weak signals, where the read noise is dominated in total noise, Eq. (4) implies that the SNR will be augmented from a greater gain, by effectively lowering the read noise. The CNR will also be benefited from a higher gain value for dark and super-weak signals because of its similar noise dependence.
It is important to experimentally investigate the effect of gain value on TGI, especially for determining its impact on SNR and CNR. This investigation is intended to assess whether the intensifier gain can consistently improve the SNR and CNR or not. The range of investigation gain was set between 0 to 4000 ×. The gates width of 500 ns, delay 50 ns, and accumulation of 10 frames was used. The gates width of 500 ns, delay 50 ns, and accumulation of 10 frames was used. We randomly selected small uniform regions from images for calculating the ratio of the average signal and standard deviation. Figure 3(a) shows the average of SNR and CNR results plotted versus the different gain values. It is found that the SNR increases rapidly upon increasing gain value from 0 to 2000 ×. Further increase in gain value actually leads to a very slow increase in SNR and eventually, it reaches to plateau at greater gain value in range of 2500–4000 ×. On the other hand, the CNR increases slowly at low-gain (between 100–750 ×) while it increases steeply between (750–2500 ×). Further increase in gain reveals a slow increase in CNR similar to the case of SNR which reaches to plateau in the gain value above 3000 ×. The SNR and CNR show that the ICCD can be operated in medium and high gain mode for weak-signals and have enough sensitivity to acquire adequate signal from low-emission intensity samples over the short-width gate. Moreover, these results indicated that there is an optimal range for the gain in order to achieve an optimal SNR and CNR. Hence, the optimal range of gain can be experimentally investigated using the above method. Essentially, the results were compiled from the average of five samples and ten measurements.
Fig. 3. Plots of SNR and CNR under different parameters from ICCD. (a) The SNR and CNR for various ICCD gain between 0 and 4000 ×. (b) The SNR and CNR for different frames accumulate between 1 and 50 frames. (c) The stability of average intensity at different frame numbers. (d) The CNR and CNR are investigated at various delays between −10 and 50 ns.
Integrating time relies heavily on the ICCD’s dynamic range. The integrating time can be adjusted by regulating the exposure time of CCD-chip and gate-width of intensifier. The intensifier can generate a very short gate-width compared to the exposure time, generated by CCD-chip. Thus, almost a full dynamic range of the ICCD and a maximum SNR and CNR can be achieved by setting these parameters suitably. According to Eq. (4), higher exposure time would result in a better SNR. However, the higher integration time adversely affects the overall imaging speed, which is not suitable for being implemented in TGI, demanding very high temporal resolution. In the time-gated method, the exposure time of the camera shall be synchronized with the gate width of the intensifier according to measuring parameters. Time integration is an important parameter in a time-gated method for covering the dynamic range of ICCD and to avoid the signal overlapping of the successive pulse train. The frame accumulation method with reducing of gates-width can be used to cover the limitations of integration time.
In order to demonstrate the benefit of frame accumulation, for each measurement, the data were collected under different frame numbers to investigate the SNR, CNR and stability of intensity value. Each frame value had one image where five frames are overlapped to the five frames to the first one (between 1 - 50 units). The gain of 1500 ×, gates width of 500 ns, delay 50 ns was used.
The measurements in Fig. 3(b) indicate that the SNR and CNR are increased with the higher accumulated frame values, wherein for fewer than 20 accumulated frames the improvements in SNR and CNR are faster with increasing frame members. However, once it crosses more than 20 frames, SNR and CNR improve very slowly upon increasing the frames. The stability of collected data at different frame accumulations are shown in Fig. 3(c). The results indicate that, the stability of the collected signal is improved by increasing the accumulated frame numbers. Generally, the stability of the intensity images increases at higher frames.
Thus, to achieve better performance for ICCD in the TGI systems, the integration time of the collected signal should be set to the lower edge of the exposure time for a given illumination photon flux. The accumulation of a larger number of frames (up to a threshold) is the decent mean of improving the imaging sensitivity, SNR and CNR besides integration time.
Time-gated detection requires the labelling of biological targets with long-lived-luminescent probes, typically having a much longer fluorescence lifetime than autofluorescence lifetimes. The investigation parameters was sit with gain of 1500 ×, gates width of 500 ns, and accumulation of 10 frames. The CNR and SNR at different delays were investigated to demonstrate the effect of delay on noise. Figure 3(d) shows the effect of time-delay on the SNR and CNR of detection gates between −10 and 50 ns. The highest SNR and CNR appear clearly between 10 ns and 20 ns. The early detection is impacted by influences of excitation and autofluorescence which lead to low SNR whereas the late delay detection is also affected by low emission signals to reveal the lower SNR and CNR. Time delay of detection gates not only influence the sensitivity of sensor positively but also can be instrumental in curtailing several limitations by lowering the noise, and eliminating short-term signal of autofluorescence contribution, especially for biological samples.
3.2 Estimation of the performance of fluorescence lifetime determination
The decay curves of AIS/ZnS QDs were generated based on time-gated method to assess the mean value of fluorescence lifetime as shown in Fig. 4(a). The decay curve was generated from the mean ratio intensity of images at the region of interest using multi gates and four gate acquisition to examine our time-gated system’s capability to measure the fluorescence lifetime. The data was collected with contiguous scheme, gain of 1000 ×, time-delay of 20 ns of the first gate, gate-widths of 50 ns and accumulations of 20 frames for multi-gates. While the four gates were collected contiguous scheme with gate-widths of 250 ns. The results of time-gated PL decays in comparison with the results obtained from standard photoluminescence are shown in Fig. 4(a,b,c) which reveal the acceptable decay value for TGI compared to PL lifetime values: TGS, 278 ns; RLD, 243.2 ns; PL, 327 ns. The slight difference in mean lifetime values is perhaps due to the instrument response of the system during the signal noise correction. In addition, the numbers of the gates, noise ratio, and the complex chemical environment may also affect the accuracy of the mean lifetime. These data show that our newly developed system is capable of acquiring fluorescence lifetime data with enough accuracy to show the expected results without much alternation of the system parameters.
Fig. 4. Comparison of PL decays curves. (a) Time-gated method. (b) HORIBA photoluminescence system. (c) Four gates model.
3.3 Comparison of two gates and four gates acquisitions
Generally, it was reported in literature that the accuracy of fluorescence lifetime based on time-gated imaging increases with many gates [48,53–55]. However, only two gates acquisition can achieve the maximum speed to generate the fluorescence decay profile for RLD. But, RLD based on two gates scheme is more sensitive to noise than the multi gates scheme [49,50,56,57], practically, in the sample with a significant lifetime difference within the FOV, multiple gates are required to obtain a reliable measurement, especially for the lifetime close to the instrument responses [55].
To compare the performance RLD based on two gates and four gates schemes, the edge of dry drop on a glass slide for AIS/ZnS QDs with low consecration were imaged. Fig. (5) shows the comparison of two gates and four gate intensity images and their lifetime mapping. The images were recorded using contiguous gates with gate-widths of 300 ns and delay time of 20 ns of the first gate with an accumulation of 10 frames for the two gated scheme, and gates-width of 150 ns for the four gates scheme. The efficiencies of a number of gates were compared by calculating SNR and CNR of intensity images and RSD of lifetime maps. Table 1 the mean intensity, SNR, and CNR of gated images (Fig. 5(a)) for both schemes at different time delays are shown. Despite the SNR and CNR are higher of images acquired by two gates acquisition, the four gates scheme reveals better enhancement in the precision of fluorescence lifetime map as shown in Fig. 5(b). Not only the standard deviation was enhancing, but also many pixels were not appearing a lifetime value in the two gated scheme. This difference is because of the sensitivity of the two gates to the noise. These results demonstrate that the four gate scheme is capable providing better accuracy of RLD with acceptable acquisition time. Hence, the four gates scheme can be an appropriate option in RLD because it combines the characteristics of two gates and multi gates schemes while being fast enough to work in real-time imaging compared to multi-gates schemes. Finally, the four gate scheme is not much affected by noise compared with the two gate scheme.
Fig. 5. Comparison of two gates and four gates schemes. (a) Two gates and four gates acquisitions. (c) Two gates Lifetime map four gates intensity images, scale bars: 100 µm.
Table 1. The mean intensities, SNR and CNR value for two gates and four gates intensity images.
3.4 Compression denoising models
The performance of the TV denoising and Max/Min filter model is compared with other denoising models, which are used for time-resolved imaging denoising. Specifically, we compared our denoising method with Low-rank denoising [58], and TV-Poisson denoising [59]. The evaluation of TV and Mix/Min filtering model for denoising raw data shows in Fig. 6(d). This model was compared with raw intensity image (Fig. 6(a), Low-rank denoising (Fig. 6(b)), and TV-Poisson denoising (Fig. 6(c)) to evaluate their quality by using SNR, and CNR. To ensure a fair comparison, we used the same predetermined acquisition parameters with a gate width of 500 ns. Comparisons of the three methods show reasonable results with a higher improvement of our proposed in RSD, SNR, and CNR than the other denoising methods (Table 2).
Fig. 6. Three denoising models. (a) Raw intensity image. (b) Low-rank denoising. (c) Poisson TV denoising. (d) TV denoising and Max/Min filtering.
Table 2. The mean intensities, SNR and CNR value varying denoising models.
3.5 Acquisition procedure
Based on the above analyses, the suitable operational procedures can be developed for the acquisition of fluorescence lifetime images using time-gated detection method with upgraded imaging parameters in terms of signal, contrast, and noise. It is important to optimize the acquisition procedure so that the parameters are set accurately to achieve enhanced CNR and SNR as well as reduced background. First, the gain of the ICCD can be set to less than 2500 × to reduce the read noise so that it can maximize the dynamic range of the ICCD to obtain higher SNR and CNR with enough sensitivity, as shown in Fig. 3(a). A gate-width can be set with the period as high as possible, taking into account the strength of collection intensity, the time decay of the sample, and overlapping of the repetition rate of laser pulses. Sequential triggering of the intensifier to the accumulated frame cycles might be proposed to eliminate the limitation of extended exposure time. For the accumulation of frames, more than 15 frames per acquisition should be adjusted to achieve good SNR and CNR values, as shown in Fig. 3(b) and Fig. 3(c). In our TGS, the initial gain is set at 2500 ×, exposure time at 10 ms, time delay at 20 ns, and the gate width is 500 ns for intensity. Lifetime range of the application can be work works for the samples have lifetime longer than 10 ns.
For rapid time-gated lifetime imaging, the read-noise can be possibly eliminated by collecting more photons by means of accumulating sequential acquisitions combined with the overlapping gates which will eventually increase the gate-width to enhance the precision of lifetime imaging.
3.6 Cells imaging
Fluorescence lifetime imaging in biological applications provides useful information about the unique molecular variations of fluorophores that are not apparent with common spectral techniques alone [60]. The frame accumulation is applied to evaluate the improvements of SNR and CNR for intensity gated imaging and the precision of rapid lifetime mapping. The fluorescence was imaged for ovarian cancer cells injected with long-lived AIS@ZnS QDs. For each measurement, the investigation encompassed four intensity images at different delays and different frame accumulation values. To validate this impact, intensity images of the same field-of-view were imaged at different frame accumulation values (from 2 to 25 frames), see in Fig. 7. The time-gated intensity images were acquired with different frame values ((Fig. 7(a)) which clearly revealed that the SNR and CNR were improved with increasing frames values as shows in Table 3. This improvement in the CNR and CNR is due to increased photon counts and collected intensity stability over all pixels. Four intensity images were considered for lifetime mapping wherein the gates were continuously detected. For the first gate operation, gate-width of 150 ns, delay time of 20 ns and overlapping delay of 150 ns were applied. The lifetime standard deviation, obtained from the post-image processing of lifetime images clearly showed the differences in the standard deviation values of different frames of the lifetime mapping (Fig. 7(b)). The standard deviation significantly decreases with high frame values. It is worth noting that for same FOV the random noise, shown in low frame values disappears in large-frame values due to the effect of readout noise and low stability of recorded fluorescence.
Fig. 7. Effect of frame accumulations on fluorescence intensity imaging and lifetime mapping. Time-gated imaging (a) and Rapid lifetime mapping (b): From left to right are accumulation of different frames between 2 to 25 frames for each measurement, scale bars: 20 µm.
Table 3. The mean intensities, SNR and CNR value different frame accumulations.
We also performed paired t-test to determine the fluorescence lifetime distribution of the data collected at 25 frame accumulations, which was significantly lower (p < 0.01) compared to the data collected at two frames as shown in Fig. 7.
Frame accumulation is an important parameter that can significantly improve SNR and CNR of photoluminescence intensity images and enhance the precision of lifetime imaging; especially for low-level signals by reducing the power of reading noise and random noise. It also offers a larger constant for the intensity component in the decay curve compared to the single frames.
The gate-width and its position for each PL intensity images affect the accuracy of lifetime determination computed using the RDL algorithm. The gate-width should be set as wide as possible to maximize the number of collected photons with optimal delay for each gate [61]. However, for gates in long-overlapping, the highest possible gate width can be achieved without much variation in delay between the gates. However, the overlapping with uneven gating can be employed to increase the collected signals and then increase the SNR in later gates to enhance the precision of the RLD without changes in the delay between detection gates [39,62,63]. The overlapping process of the gates is schematically shown in Fig. 1(d). Herein, the impact of overlapping value with the increasing gate-width on the precision of rapid lifetime imaging is investigated. On average, 15 frames were used to collect the data wherein five gate values; separate gate (25%), continue gate (0%), and overlapping gate (75%, 50%, and 25%); with increasing gate widths of 100 ns, 150 ns, 200 ns, 250 ns, and 300 ns, respectively were employed as shown in Fig. 8. The lifetime mapping clearly revealed the conspicuous differences in the standard deviation values for different overlapping and gate widths. The results indicate that the lifetime standard deviation is a minimum at 50% overlapping, while the standard deviation is higher at 75% gate overlapping.
Fig. 8. Overlapping gates combined with gate-widths. Rapid lifetime mapping, separate gates 25% and gate width 100 ns (a), continue gates and gate width 150 ns (b), overlapping gates 25% and gate width 200 ns (c), overlapping gate 50% and gate width 250 ns (d),overlapping gate 75% and gates width 300 ns (e), for each measurement, scale bar: 20 µm.
For low-intensity samples, decreasing the gate step translates into the use of more overlapping which leads to the overall improvement in precision since more photons are being captured.
Optimal gating, denoising, and filtering of intensity images are all responsible for improving lifetime precision. Fluorescence lifetime mapping of QDs, showing the noise distribution in cells for non-optimal and optimal gates are illustrated in Fig. 9. In the direct determination of lifetime, the threshold showed as red points (Fig. 9(a)) are set at zero in treated images (Fig. 9(c)). The random fluctuations of weak-signals caused high random noise in some pixels which was evident from the τ value, exceeding the threshold. Those pixels clearly appeared as holes in lifetime images for direct mapping of intensity images as shown in Fig. 9(c). On the other hand, Figs. 9(c,b) showed much distorted images with high RSD values, while RSD values declined for Figs. 9(b,d). In addition, the denoising and filtering of lifetime images (Figs. 9(c,d)) showed close RSD values of direct lifetime maps (Figs. 9(a,b)).
Fig. 9. Fluorescence lifetime images of ovarian cells dyed with fluorescent AIS@ZnS QDs. Four intensity images were acquired. Direct lifetime mapping without treatment of images, (a) non-optimal gates, and (b) optimal gates. Treatment of lifetime images by denoising and Max/Min filtering, (c) non-optimal gates and (d) optimal gate. Lifetime mapping of treated intensity images with denoising and Max/Min filtering, (e) non-optimal gates and (f) optimal gates, scale bars: 20 µm.
The TV denoising and Max/Min filtering were used to process the intensity images for non-optimal and optimal gates, which resulted in the improved lifetime images as shown in Figs. 9(e,f), respectively. Essentially, the image quality was enhanced substantially and smoother lifetime images with lesser RSD value (15%) and improved τ values were obtained using optimal gates, combined with denoising and filtering.
The correction of data using denoising and filtering for additional pixels resulted in the observed better-quality spatial pattern as poor data sets were effectively “rejected”. Further improvement was then achieved by again applying the denoising/filtering approach on the optimally gated images. Therefore, it can be anticipated that the optimal gate operations either can be employed independently or can be coupled with the TV denoising and Max/Min filtering approaches to improve the precision in low-light time-gated lifetime imaging.
Finally, proposed lifetime precision enhancement approach was also applied to test the feasibility of the method in live cell imaging. Using this method in live cells, the background-free intensity images (Fig. 10(a)), and the RLD (Fig. 10(b)) of the same FOV were obtained. It is important to note that, the autofluorescence did not overlap with photoluminescence emission in this time-gated method as the long-lived QDs were used for fluorescent labelling. Eventually, the lifetime precision enhancement method provided much clearer background-free images in live cells (Fig. 10(a)) which is also reflected in the observed lifetime precision as confirmed by RSD value (Fig. 10(b)). By applying our method, the background was eliminated and the SNR & CNR were improved for the fluorescence intensity images while maintaining the minimum RSD for lifetime images. Notably, the rapid lifetime imaging was obtained using 15 frames, wherein the measurement time to obtain single intensity image was 75 ms. Total acquisition time for rapid lifetime mapping was about 0.8 - 6 seconds, which is comparable to that of the standard fluorescence cellular imaging techniques but shorter than the time frame of various biological events.
Fig. 10. Optimal time gated imaging for ovarian cancer cells. (a) Background-free imaging. (b) Rapid lifetime mapping, scale bar: 20 µm.
4. Conclusion
In this paper, we present a method that can overcomes the challenges of noise distributions of weak signals for rapid lifetime determination using the optimal gates, combined with TV denoising and Max/Min filtering. Optimized gates can improve the quality of imaging data and maximize the SNR and CNR of the images. While the TV denoising and Max/Min filtering can remove uncertainties of removing unwanted noise and contrast of the images. Results indicate that there are optimal ranges for systematic parameters (such as sensor gain, frame numbers, and delay) to achieve an optimal SNR and CNR, that can be experimentally determined. Notably, improvements in lifetime precision can be easily observed for weak-signal images using our method. Even though, increased frame accumulation and gate-overlapping lead to an increase in the photon counts, which contributed significantly to reducing the standard deviation while acceptable acquisition times. Hence, increase in the frame accumulations as well as the overlapping with large-gate widths can improve the sensitivity of weak signals, coming from biological specimens to enhance the image quality. Furthermore, the results indicated that the TV denoising and Max/Min filtering models could improve the lifetime determination, reliably. The approach presented the scope to improve time-gated data acquisition by increasing imaging speed and minimizing sample light exposure to avoid biological sample damage, photobleaching, and unwanted sample movement detection in fluorescence lifetime imaging applications.
Funding
National Natural Science Foundation of China (61835009, 62127819, 61935012, 62175163, 61961136005); Shenzhen Key Projects (JCYJ20200109105404067); Shenzhen Talent Innovation Project (RCJC20210706091949022); Shenzhen International Cooperation Project (GJHZ20190822095420249).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. W. Yang and S. L. Chen, “Time-gated fluorescence imaging: advances in technology and biological applications,” J. Innov. Opt. Health Sci. 13(03), 2030006 (2020). [CrossRef]
2. A. Ulku, A. Ardelean, M. Antolovic, S. Weiss, E. Charbon, C. Bruschini, and X. Michalet, “Wide-field time-gated SPAD imager for phasor-based FLIM applications,” Methods Appl. Fluoresc. 8(2), 024002 (2020). [CrossRef]
3. C. Schiattarella, R. Moretta, T. Defforge, G. Gautier, B. D. Ventura, M. Terracciano, C. Tortiglione, F. Fardella, P. Maddalena, L. D. Stefano, R. Velotta, and I. Rea, “Time-gated luminescence imaging of positively charged poly-l-lysine-coated highly microporous silicon nanoparticles in living Hydra polyp %J Journal of Biophotonics,” J. Biophotonics. 13, e202000272 (2020). [CrossRef]
4. M. Gaudin and C. Desnues, “Hybrid capture-based next generation sequencing and its application to human infectious diseases,” Front. Microbiol. 9, 2924 (2018). [CrossRef]
5. S. Andersson-Engels, C. A. Klinteberg, K. Svanberg, and S. Svanberg, “In Vivo Fluorescence Imaging for Tissue Diagnostics,” Phys. Med. Biol. 42(5), 815–824 (1997). [CrossRef]
6. F. Leblond, S. C. Davis, P. A. Valdés, and B. W. Pogue, “Pre-clinical whole-body fluorescence imaging: Review of instruments, methods and applications,” J. Photochem. Photobiol. B 98(1), 77–94 (2010). [CrossRef]
7. J. Jin and J. A. Piper, “Time-gated luminescence microscopy allowing direct visual inspection of lanthanide-stained microorganisms in background-free condition,” Anal. Chem. 83(6), 2294–2300 (2011). [CrossRef]
8. D. S. Talaga, “Information-theoretical analysis of time-correlated single-photon counting measurements of single molecules,” J. Phys. Chem. 113(17), 5251–5263 (2009). [CrossRef]
9. W. Becker and A. Bergmann, “Lifetime imaging techniques for optical microscopy,” Technical report, Becker & Hickl GmbH (2003).
10. Y. W. Jun, H. R. Kim, Y. J. Reo, M. Dai, and K. H. Ahn, “Addressing the autofluorescence issue in deep tissue imaging by two-photon microscopy: the significance of far-red emitting dyes,” Chem. Sci. 8(11), 7696–7704 (2017). [CrossRef]
11. Y. Zhong and H. Dai, “A mini-review on rare-earth down-conversion nanoparticles for NIR-II imaging of biological systems,” Nano Res. 13(5), 1281–1294 (2020). [CrossRef]
12. X. Liu, D. Lin, W. Becker, J. Niu, B. Yu, L. Liu, and J. Qu, “Fast fluorescence lifetime imaging techniques: A review on challenge and development,” J. Innov. Opt. Health Sci. 12(05), 1902003 (2019). [CrossRef]
13. X. Cao, C. Yao, S. Jiang, J. Gunn, A. C. V. Namen, P. Bruza, and B. W. Pogue, “Time-gated luminescence imaging for background free in vivo tracking of single circulating tumor cells,” Opt. Lett. 45(13), 3761–3764 (2020). [CrossRef]
14. X. Intes, “Wide-field Time-Resolved Molecular Optical Tomography,” in Optical Trapping Applications. Optical Society of America, p. JW4A. 2 (2013).
15. K. Dowling, M. J. Dayel, M. J. Lever, P. M. W. French, J. D. Hares, and A. K. L. Dymoke-Bradshaw, “Fluorescence lifetime imaging with picosecond resolution for biomedical applications,” Opt. Lett. 23(10), 810–812 (1998). [CrossRef]
16. Y. Liu, Y. Zhou, and Y. Liu, “A rapid fluorescence lifetime image acquistion method based on time-gated fluorescence lifetime imaging microscopy,” in The 2014 2nd International Conference on Systems and Informatics (ICSAI 2014) (IEEE2014), pp. 808–812.
17. P. Alivisatos, “The use of nanocrystals in biological detection,” Nat Biotechnol 22(1), 47–52 (2004). [CrossRef]
18. A. Armaselu, A. Popescu, I. Apostol, I. Ardelean, V. Damian, I. Iordache, I. Sarchizian, and D. Apostol, “Passive nonspecific labeling of cyanobacteria in natural samples using quantum dots,” Optoelectron. Adv. Mater. Rapid Commun. 5, 1084–1090 (2011).
19. J. Wang, S. Han, D. Ke, and R. Wang, “Semiconductor quantum dots surface modification for potential cancer diagnostic and therapeutic applications,” J. Nanomater. 2012, 1–8 (2012). [CrossRef]
20. A. M. Smith, H. Duan, A. M. Mohs, and S. Nie, “Bioconjugated quantum dots for in vivo molecular and cellular imaging,” Adv. Drug Deliv. Rev. 60(11), 1226–1240 (2008). [CrossRef]
21. K.-T. Yong, J. Qian, I. Roy, H. H. Lee, E. J. Bergey, K. M. Tramposch, S. He, M. T. Swihart, A. Maitra, and P. N. Prasad, “Quantum rod bioconjugates as targeted probes for confocal and two-photon fluorescence imaging of cancer cells,” Nano Lett. 7(3), 761–765 (2007). [CrossRef]
22. H. Kobayashi, Y. Hama, Y. Koyama, T. Barrett, C. A. Regino, Y. Urano, and P. L. Choyke, “Simultaneous multicolor imaging of five different lymphatic basins using quantum dots,” Nano Lett. 7(6), 1711–1716 (2007). [CrossRef]
23. S. Chen, M. Ahmadiantehrani, N. G. Publicover, K. W. Hunter, and X. Zhu, “Thermal decomposition based synthesis of Ag-In-S/ZnS quantum dots and their chlorotoxin-modified micelles for brain tumor cell targeting,” RSC Adv. 5(74), 60612–60620 (2015). [CrossRef]
24. B. Chen, Q. Liu, Y. Zhang, L. Xu, and X. Fang, “Transmembrane delivery of the cell-penetrating peptide conjugated semiconductor quantum dots,” Langmuir 24(20), 11866–11871 (2008). [CrossRef]
25. L. Liu, R. Hu, I. Roy, G. Lin, L. Ye, J. L. Reynolds, J. Liu, J. Liu, S. A. Schwartz, and X. Zhang, “Synthesis of luminescent near-infrared AgInS2 nanocrystals as optical probes for in vivo applications,” Theranostics 3(2), 109–115 (2013). [CrossRef]
26. I. A. Mir, V. Radhakrishanan, K. Rawat, T. Prasad, and H. B. Bohidar, “Bandgap tunable AgInS based quantum dots for high contrast cell imaging with enhanced photodynamic and antifungal applications,” Sci. Rep. 8(1), 9322 (2018). [CrossRef]
27. L. Liu, R. Hu, W.-C. Law, I. Roy, J. Zhu, L. Ye, S. Hu, X. Zhang, and K. T. Yong, “Optimizing the synthesis of red-and near-infrared CuInS 2 and AgInS 2 semiconductor nanocrystals for bioimaging,” Analyst 138(20), 6144–6153 (2013). [CrossRef]
28. F. Yang, M. S. Ozturk, R. Yao, and X. Intes, “Improving mesoscopic fluorescence molecular tomography through data reduction,” Biomed. Opt. Express 8(8), 3868–3881 (2017). [CrossRef]
29. K. M. Tichauer, K. S. Samkoe, K. J. Sexton, J. R. Gunn, T. Hasan, and B. W. Pogue, “Improved tumor contrast achieved by single time point dual-reporter fluorescence imaging,” J. Biomed. Opt. 17(6), 066001 (2012). [CrossRef]
30. R. Bhavane, Z. Starosolski, I. Stupin, K. B. Ghaghada, and A. Annapragada, “NIR-II fluorescence imaging using indocyanine green nanoparticles,” Sci. Rep. 8(1), 14455 (2018). [CrossRef]
31. S. Gioux, A. M. Degrand, D. S. Lee, S. Yazdanfar, J. D. Idoine, S. J. Lomnes, and J. V. Frangioni, “Improved optical sub-systems for intraoperative near-infrared fluorescence imaging,” in Optical Methods in Drug Discovery and Development (International Society for Optics and Photonics2005), p. 60090C.
32. Y. Jung, S. Jeong, W. Nayoun, B. Ahn, J. Kwag, S. G. Kim, and S. J. Kim, “Quantum dot imaging in the second near-infrared optical window: studies on reflectance fluorescence imaging depths by effective fluence rate and multiple image acquisition,” J. Biomed. Opt. 20(4), 046012 (2015). [CrossRef]
33. C. W. Chang and M. A. Mycek, “Enhancing precision in time-domain fluorescence lifetime imaging,” J. Biomed. Opt. 15(5), 056013 (2010). [CrossRef]
34. G. Gong, H. Zhang, and M. Yao, “Speckle noise reduction algorithm with total variation regularization in optical coherence tomography,” Opt. Express 23(19), 24699–24712 (2015). [CrossRef]
35. M. Al-Salihi, R. Yi, S. Wang, Q. Wu, F. Lin, J. Qu, and L. J. O. E. Liu, “Quantitative laser-induced breakdown spectroscopy for discriminating neoplastic tissues from non-neoplastic ones,” Opt. Express 29(3), 4159–4173 (2021). [CrossRef]
36. J. Wei, F. Li, C. Chang, and Q. Zhang, “Synthesis of emission tunable AgInS2/ZnS quantum dots and application for light emitting diodes,” J. Phys. Commun. 4(4), 045016 (2020). [CrossRef]
37. S. Zhu, Y. Xu, and D. Li, “Three-observation-window time-gated algorithm for fluorescence lifetime detection,” Appl. Opt. 59(9), 2739–2745 (2020). [CrossRef]
38. X.-D. Chen, Y. Zheng, B. Du, D.-F. Li, S. Li, Y. Dong, G.-C. Guo, and F.-W. Sun, “High-contrast quantum imaging with time-gated fluorescence detection,” Phys. Rev. Appl. 11(6), 064024 (2019). [CrossRef]
39. Y. Li, H. Jia, S. Chen, J. Tian, L. Liang, F. Yuan, H. Yu, and D. D.-U. Li, “Single-shot time-gated fluorescence lifetime imaging using three-frame images,” Opt. Express 26(14), 17936–17947 (2018). [CrossRef]
40. D. Ma, J. Liu, J. Qi, and L. Marcu, “A novel method for fast and robust estimation of fluorescence decay dynamics using constrained least-square deconvolution with Laguerre expansion’,” Phys. Med. Biol. 62(4), 1637–1641 (2017). [CrossRef]
41. K. B. Lee, J. Siegel, S. Webb, S. Leveque-Fort, M. Cole, R. Jones, K. Dowling, M. J. Lever, and P. M. W. French, “Application of the stretched exponential function to fluorescence lifetime imaging,” Biophys. J. 81(3), 1265–1274 (2001). [CrossRef]
42. W. Zhang and Q. Chen, “Signal-to-noise ratio performance comparison of electron multiplying CCD and intensified CCD detectors,” in 2009 International Conference on Image Analysis and Signal Processing (IEEE2009), pp. 337–341.
43. D. Dussault and P. Hoess, “Noise performance comparison of ICCD with CCD and EMCCD cameras,” In Infrared Systems and Photoelectronic Technology, 5563, 195–204 (2004).
44. J. McGinty, J. Requejo-Isidro, I. Munro, C. Talbot, P. Kellett, J. Hares, C. Dunsby, M. Neil, and P. P. M. W. French, “Signal-to-noise characterization of time-gated intensifiers used for wide-field time-domain FLIM,” J. Phys. D: Appl. Phys. 42(13), 135103 (2009). [CrossRef]
45. C. W. Chang and M. Mycek, “Total variation versus wavelet-based methods for image denoising in fluorescence lifetime imaging microscopy,” J. Biophotonics 5(5-6), 449–457 (2012). [CrossRef]
46. F. Deger, A. Mansouri, M. Pedersen, J. Y. Hardeberg, and Y. Voisin, “A sensor-data-based denoising framework for hyperspectral images,” Opt. Express 23(3), 1938–1950 (2015). [CrossRef]
47. X. Zhong, X. Wang, L. Sun, and Y. Zhou, “Enhancement of rapid lifetime determination for time-resolved fluorescence imaging in forensic examination,” Chin. Opt. Lett. 19(4), 041101 (2021). [CrossRef]
48. H. C. Gerritsen, M. Asselbergs, A. Agronskaia, and W. G. J. H. M. V. Sark, “Fluorescence lifetime imaging in scanning microscopes: acquisition speed, photon economy and lifetime resolution,” J. Microsc. 206(3), 218–224 (2002). [CrossRef]
49. R. Cubeddu, D. Comelli, C. D’Andrea, P. Taroni, and G. Valentini, “Time-resolved fluorescence imaging in biology and medicine,” J. Phys. D: Appl. Phys. 35(9), 201R61 (2002). [CrossRef]
50. J. Jo, Q. Fang, T. Papaioannou, and L. Marcu, “Novel ultra-fast deconvolution method for fluorescence lifetime imaging microscopy based on the Laguerre expansion technique,” in The 26th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (IEEE2004), pp. 1271–1274.
51. D. Sud, W. Zhong, D. G. Beer, and M.-A. Mycek, “Time-resolved optical imaging provides a molecular snapshot of altered metabolic function in living human cancer cell models,” Opt. Express 14(10), 4412–4426 (2006). [CrossRef]
52. T. J. S. Junk, “S. Methods in Physics Research Section A: Accelerators, Detectors, and A. Equipment,” Confidence level computation for combining searches with small statistics. 434, 435–443 (1999).
53. L. Marcu, P. M. French, and D. S. Elson, Fluorescence lifetime spectroscopy and imaging: principles and applications in biomedical diagnostics (CRC Press, 2014).
54. C. J. D. Grauw and H. C. Gerritsen, “Multiple time-gate module for fluorescence lifetime imaging,” Appl. Spectrosc. 55(6), 670–678 (2001). [CrossRef]
55. D. M. Grant, J. McGinty, E. J. McGhee, T. D. Bunney, D. M. Owen, C. B. Talbot, W. Zhang, S. Kumar, I. Munro, P. M. P. Lanigan, G. T. Kennedy, C. Dunsby, A. I. Magee, P. Courtney, M. Katan, M. A. A. Neil, and P. M. W. French, “High speed optically sectioned fluorescence lifetime imaging permits study of live cell signaling events,” Opt. Express 15(24), 15656–15673 (2007). [CrossRef]
56. R. M. Ballew and J. N. Demas, “An error analysis of the rapid lifetime determination method for the evaluation of single exponential decays,” Anal. Chem. 61(1), 30–33 (1989). [CrossRef]
57. C. Moore, S. P. Chan, J. N. Demas, and B. J. A. DeGraff, “Comparison of methods for rapid evaluation of lifetimes of exponential decays,” Appl. Spectrosc. 58(5), 603–607 (2004). [CrossRef]
58. T. Furnival, R. K. Leary, and P. A. Midgley, “Denoising time-resolved microscopy image sequences with singular value thresholding,” Ultramicroscopy 178, 112–124 (2017). [CrossRef]
59. T. Le, R. Chartrand, and T. J. Asaki, “A variational approach to reconstructing images corrupted by Poisson noise,” J. Math. Imaging Vis. 27(3), 257–263 (2007). [CrossRef]
60. R. Datta, T. M. Heaster, J. T. Sharick, A. A. Gillette, and M. C. Skala, “Fluorescence lifetime imaging microscopy: fundamentals and advances in instrumentation, analysis, and applications,” J. Biomed. Opt. 25(07), 1 (2020). [CrossRef]
61. S.-J. Chen, N. Sinsuebphon, and X. Intes, “Assessment of gate width size on lifetime-based Förster Resonance Energy Transfer parameter estimation,” Photonics 2(4), 1027–1042 (2015). [CrossRef]
62. S. P. Chan, Z. J. Fuller, J. N. Demas, and B. A. Degraff, “Optimized gating scheme for rapid lifetime determinations of single-exponential luminescence lifetimes,” Anal. Chem. 73(18), 4486–4490 (2001). [CrossRef]
63. D. D. U. Li, S. Ameer-Beg, J. Arlt, D. Tyndall, R. Walker, D. R. Matthews, V. Visitkul, J. Richardson, and R. K. Henderson, “Time-Domain Fluorescence Lifetime Imaging Techniques Suitable for Solid-State Imaging Sensor Arrays,” Sensors 12(5), 5650–5669 (2012). [CrossRef]
