Three-dimensional thermophotonic super-resolution imaging by spatiotemporal diffusion reversal method

Principle of eTC-PCT imaging modality

In eTC-PCT imaging modality (6), a linear frequency-modulated chirp of short optical pulses is introduced on the surface of a sample, and the IR responses emitted by the sample through an IR camera are monitored as shown in Fig. 1A. The in-phase (R₀) and quadrature (R₉₀) reference signals, which are synthesized from the excitation chirp and the IR response at a specific pixel location using a pulse chirp with frequency sweep range of 0.2 to 0.6 Hz, duration of 12 s, pulse duration of 40 ms, and wavelength of 808 nm as shown in Fig. 1B, are cross-correlated at each pixel location. The resulting time delay measurement determines the depth of the absorber, as illustrated schematically in eTC-PCT signal processing diagram (Fig. 1C). The cross-correlation of R₀ and R₉₀ reference signals with the IR signal is depicted in Fig. 1 (D and E, respectively). The amplitude cross-correlation (ACC) and phase cross-correlation (ØCC) responses were calculated using equations provided in Fig. 1C. The final correlation responses (ACC and ØCC) were then truncated at consecutive time intervals by a time-gating filter to extract depth-resolved amplitude and phase information. The truncation region is shown in Fig. 1 (D and E) determined by the number of camera frames available between the quadrature component of the (n − 1)th pulse and the in-phase component of the nth pulse, where n is the number of the excitation pulse (6). The responses outside the truncation region were discarded, as they contain higher harmonic components. The width of the gated interval (d) can be adjusted up to d_min (=1/f_s), where f_s is the sampling frequency of IR camera (104 Hz). The data combined from all pixel locations at a specific time interval lead to tomographic amplitude and phase slice images (2D) of the sample. Amplitude and phase channels at the same pixel and delay time were combined (cross-multiplied) to obtain a new output for each delayed slice as our experimental results showed that the combined channel features higher spatial resolution and dynamic range than either channel alone, revealing more detailed structures of biological samples such as small vascular capillaries (14). This was done in analogy to photoacoustic frequency-domain imaging, which also showed that the amplitude provides increased SNR, while the phase exhibits higher spatial resolution (15). These combined amplitude and phase output images from each slice were adopted throughout the optimization of eTC-PCT images. A deconvolution method based on experimentally calculated PSF (text S1) followed by spatial second-derivative formation (text S2) and a unique spatial gradient adaptive filter (text S3) was applied to reconstruct thermophotonic super-resolution tomography images.

Imaging of hairline dental crack in extracted human tooth

An anonymous extracted human tooth with a hairline crack on the mesial area of the tooth (shown by a red arrow in Fig. 2A) was selected. The crack was found to be of a relatively uniform width of 100 μm across the tooth. This was further confirmed using μCT (Fig. 2B). The depth of the crack is shown in the tomography image (Fig. 2C), which shows that the crack is extended along much of the enamel surface. Depths of the cracks were estimated at distances 0, 346, and 489 μm from the surface using Eq. 1 provided in Materials and Methods corresponding to delay times 0, 200, and 400 ms, respectively. Figure 2D shows conventional eTC-PCT images. Because of the diffuse nature of heat conduction in enamel, the resolution of the crack degrades rapidly with distance and time from the surface of the tooth. As can be seen in the 200- to 400-ms delay time images, the crack is diffused in the lateral direction and the contrast is substantially degraded. However, the size of the deep crack is almost the same in the μCT slice (Fig. 2C). The eTC-PCT images after applying the PSF-mediated Richardson-Lucy deconvolution, second-derivative, and spatial gradient adaptive filter are shown in Fig. 2E. The process reversed the lateral diffusion and extensively improved the spatial resolution of the crack. The deep crack (400-ms delay time), which deteriorated in the conventional eTC-PCT image (Fig. 2D), is clearly visible in the filtered image (Fig. 2E).

The extracted two-dimensional eTC-PCT slices were processed by ImageJ software to generate 3D volume reconstruction (16). Figure 2 (F and G) present the transverse cross section of the 3D volume images from the original eTC-PCT tomograms and the processed eTC-PCT datasets, respectively. The transverse cross-sectional image of eTC-PCT channels (Fig. 2F) shows that the crack is diffused inside the tooth, and it is almost impossible to determine its extent, whereas the transverse cross-sectional image of filtered eTC-PCT channels (Fig. 2G) shows that the crack is extended inside the tooth over the cross section down to a depth of 0.6 mm from the surface of the tooth (black arrow in Fig. 2G). These results show that the 3D reconstruction of teeth from the diffusion-corrected tomographic slices provides greatly improved visualization of the boundaries of the crack while its extent inside the tooth can be observed in a transverse plane across the tooth very similar to the μCT image (Fig. 2C) but with much better contrast.

In vivo imaging of mouse brain

In vivo eTC-PCT imaging of a mouse brain was performed by placing the mouse under anesthesia. Figure 3A shows a photograph of a mouse head; the imaged region of interest is shown by the red rectangle. The conventional eTC-PCT images were deteriorated largely by the diffusive spread as shown in delay times 0 ms, 100 ms, and 0.8 s (Fig. 3B). As expected, the near-subsurface eTC-PCT images (0 and 100 ms) are much less blurred compared to the deeper image (0.8 s). Therefore, deconvolution and second-derivative algorithms were sufficient for restoring the true spatial content of the biological features, such as near-subsurface blood vessels. With combined deconvolution and second-derivative processing, the near-subsurface blood vessel network (purple arrows at 100 ms) that appeared lumped or could hardly be distinguished in the original eTC-PCT images can now be clearly resolved in the 0- to 100-ms delay time slices (Fig. 3C). These small blood vessels disappeared in deeper slices and only large blood vessels (orange arrows) could be observed at 0.8-s delay time (Fig. 3C). Such deep subsurface features became visible only by applying deconvolution followed by spatial second-derivative and spatial adaptive filtering. The other brain anatomical features, such as cerebrum (white arrow), hemisphere of cerebellum (red arrow), superior sagittal sinus (green arrow), and transverse sinus (yellow arrows) are distinctly visible in the diffusion-corrected image at 0.8-s delay time (Fig. 3C).

Ex vivo imaging of mouse brain

eTC-PCT imaging of the mouse brain was subsequently performed ex vivo after the removal of the scalp of the euthanized mouse. Figure 4A shows an open-skull photograph of the cortex vasculature. The conventional eTC-PCT images at 0-ms, 100-ms, and 0.8-s delay times are depicted in Fig. 4B. The diffusion-reversal process yielded the blur suppressed/high-contrast images shown in Fig. 4C after the implementation of deconvolution and second-derivative processing of the near-surface slices (0- and 100-ms delay times) and all three algorithms on the deeper slice (0.8-s delay time). It is interesting to note that the blood vessel network in Fig. 3C, associated with the skin of the skull (0- to 100-ms delay times), is not present in this open-skull image sequence shown in Fig. 4C. The remaining features are similar to the in vivo depth selective images of the mouse brain: the interior blood vessels (orange arrow), cerebrum (white arrow), hemisphere of cerebellum (red arrow), superior sagittal sinus (green arrow), and transverse sinus (yellow arrow) became visible within a 100-ms delay time, although they were notably distinct at 0.8-s delay time. In Fig. 4, individual structures appear from various subsurface layers at shorter (by ~50%) delay times than in Fig. 3, owing to the open-skull inspection with no skin and bone interfaces to act like thermal delay lines. None of the interior brain structures can be seen in the visible light photograph in Fig. 4A.

In vivo monitoring of tumor in mouse thigh

In vivo eTC-PCT imaging of a healthy mouse’s thigh was acquired by placing the mouse under anesthesia, as shown in the photograph (Fig. 5A). The red rectangle on the mouse’s thigh indicates the imaged area. The conventional eTC-PCT images at 400-ms and 1.5-s delay times are shown in Fig. 5B. These two images provide different depth information of the mouse thigh: The former provides the small saphenous veins/arteries, while the latter predominantly images the great saphenous vein, a large vein of the legs and thigh. The blood vessels are visible in the eTC-PCT images, as they provide higher photothermal signals compared to the surrounding tissues because of the increased absorption coefficient of hemoglobin. The small blood vessels (400-ms delay) are much more prominent in the eTC-PCT images after applying deconvolution and second-derivative processing (Fig. 5C). Similarly, the true spatial content of the great saphenous vein (1.5-s delay) was achieved by applying deconvolution followed by spatial second-derivative and spatial adaptive filtering.

The eTC-PCT imaging of the anesthetized mouse thigh was subsequently performed after injecting cancer cells. Figure 6 presents the results on the ninth day after the injection of cancer cells. The conventional eTC-PCT images at 300-ms and 1.5-s delay times are shown in Fig. 6A, while the eTC-PCT image after deconvolution and the second derivative at 300-ms delay time and after applying all three algorithms at 1.5-s delay time are shown in Fig. 6B. The eTC-PCT images at 300-ms delay time show the formation of new blood vessels (angiogenesis) in the form of a “vessel ring” surrounding the injection location to support the rapid proliferation of cancer cells. This is a critical indicator of the onset of cancerous lesion growth. The vasculature network around the tumor becomes much more visible because of enhanced contrast against the blood-deficient background after the implementation of deconvolution and second derivative on the conventional eTC-PCT image (Fig. 6B). The formation of new blood vessels and their connections with the existing vessels are not visible in the deeper slices (after 1.5-s delay time; Fig. 6B), which predominantly show the blood-rich great saphenous vein. Nevertheless, the contrast and spatial resolution of the great saphenous vein have been greatly improved, and the true spatial extent of the vein and surrounding vessels, which had been obliterated from the original diffusive image, has been restored following spatiotemporal filtering.

After producing the eTC-PCT images on the ninth day of injection, the mouse was euthanized, and the leg skin was removed. Figure 6C shows the visible light photograph of the skinless mouse thigh. The tumor is clearly visible in the photograph after skin removal. To further confirm the presence of a tumor, the tumorous tissues were extracted and sent for histological imaging. Figure 6 (D and E) shows the hematoxylin and eosin (H&E) images of the tumor cells of uniform appearance characteristically arranged in nests and covered by fibrous bands. The surrounding normal tissues, which are pink/red in color, showed eosinophilic hyaline staining (Fig. 6E). The two green quadrangles in the middle of the tumor (Fig. 6D) indicate the in-software measured sizes. The size of the tumor measured with the software was 2.8 mm by 1.6 mm. The geometry of the tumor cell in the H&E staining (purplish blue color in Fig. 6D) is consistent with the visible light photograph of the tumorous thigh (Fig. 6C).

Dental sample preparation and eTC-PCT imaging

Anonymous extracted human teeth were collected from local oral surgeons in Toronto following the ethical guidelines in place at the University of Toronto (protocol number 00041610). The collected teeth were subsequently cleaned, visually inspected, and kept in a tightly capped container containing 0.1% thymol solution. For the experiments, a tooth with a hairline crack was selected. The tooth was washed off in running water, dried in the air, and mounted on a LEGO block to ensure repeatable and accurate positioning of the sample while taking eTC-PCT images. To produce thermophotonic images, raw camera frames (sampling frequency, f = 104 Hz) were collected from the eTC-PCT system with a fixed-width excitation pulse in the frequency range between 0.2 and 0.6 Hz, a chirp duration of 12 s, and pulse duration of the excitation chirp of 40 ms. The energy density on the sample surface was 0.66 J/cm². This value complies with the maximum permissible exposure (MPE) requirements (22), operating at 82% of the allowable maximum energy for clinical uses. After the eTC-PCT signal processing, tomographic images of size 80 × 64 pixels were generated at 20-ms intervals. The depth of the crack (z) at delay time (t) was calculated using Eq. 1

where α = 4.7 × 10⁻⁷ m²s⁻¹ is the thermal diffusivity of enamel (23).

Mouse brain and thigh imaging

Two adult CD-1Nude mice (Charles River Breeding Laboratories) were used for brain and thigh imaging following protocol 20011804 approved by the Division of Comparative Medicine of the Faculty of Medicine, University of Toronto. Laboratory animal care guidelines were followed while handling the animal. For the in vivo brain imaging, one mouse was anesthetized by administrating 1 liter/min of oxygen mixed with isoflurane gas (3% for induction and 1 to 1.5% for maintenance). A heating IR lamp was used to regulate the animal temperature. The mouse was laid on an anesthetic machine inclined normally to the camera, and the selected area was illuminated and imaged using the eTC-PCT system under anesthesia. The homogenized laser beam diameter was 3.6 cm, and the size of the IR images was 1.4 cm by 1.12 cm. For ex vivo brain imaging, the skin of the mouse was removed after the mouse was euthanized and imaged using the eTC-PCT system. eTC-PCT imaging of a mouse brain was performed using a 0.03- to 0.1-Hz chirp bandwidth, 81-s duration, and 140-ms pulse width. The energy density on the sample surface was 1.1 J/cm². The tomographic images were extracted with a time slice width of 10 ms for achieving the highest axial resolution and SNR.

For the thigh imaging, another mouse was used following the same protocol as in the brain imaging. The eTC-PCT imaging was acquired using a 0.07- to 0.1-Hz chirp bandwidth, 81-s duration, and 70-ms pulse width light source as an external illumination. The laser beam diameter was 3 cm, and the IR image size was 1.35 cm by 1.08 cm. The energy density on the sample surface was 0.79 J/cm², which was less than the maximum permissible exposure, which is an important feature toward clinical applications of the eTC-PCT technique. The tomographic images were extracted with a time slice width of 100 ms to achieve higher signal quality.

Cancer cell preparation

In this study, FaDu cell lines isolated from human patient with hypopharyngeal head and neck squamous cell carcinoma (American Type Culture Collection, Manassas, VA) and cultured in minimal essential medium (MEM) F-15 supplemented with 10% fetal bovine serum were used for developing a tumor in the mouse thigh. The right thigh of the nude mouse was injected subcutaneously with 4.8 × 10⁶ cultured cells/30 μl and imaged consecutively with the eTC-PCT system to monitor the progression of the tumor.

Histopathology imaging of the tumor

Nine days after cancer cell inoculation, the mouse was euthanized, and tumorous tissues were extracted and sent to Princess Margaret Hospital, Toronto, ON, for histopathology analysis. The extracted tissues were fixed in 4% of neutral-buffered formalin and routinely processed in paraffin wax. H&E staining was conducted using standard procedures and reviewed under a microscope.

Experimental thermophotonic imaging system

The eTC-PCT imaging modality (Fig. 1, A and C) was used to implement the high-resolution image processing method. The system uses a function generator (Keysight 33500B, USA) to generate a linear frequency modulation (LFM) excitation chirp that controls the diode laser (808 nm; Jenoptic JOLD-120-QPXF-2P) through a laser driver (PCO-6131; Directed Energy, CO, USA). A high-speed DAQ device (NI PCI-6281) receives the excitation signal, which is used to generate a reference linear frequency modulation waveform (LFM chirp). The laser beam passes through a collimator (F22SMA-B; Thorlabs Inc., NJ, USA), and a diffuser (ED1-C20; Thorlabs Inc., NJ, USA) for collimation, expansion, and homogenization. The final component is the IR camera (A6700sc, FLIR, USA; 3- to 5-μm spectral response), which documents the thermal profile of the laser-irradiated sample at a frame rate of 104 Hz. The entire system is controlled by a computer.

Micro–computed tomography

A μCT system (Skyscan1172 μCT system, Bruker MicroCT, Kontich, Belgium) was used to take images of teeth. The tooth was fixed on a LEGO block and placed in the field of view of the μCT detector. X-rays were generated at a voltage of 100 kV with an exposure time of 590 ms per frame. The tooth was rotated 360° with an angular step of 0.4°. μCT images (n = 987) of size 1724 × 1392 were reconstructed using NRecon software (version: 1.7.4; Skyscan, Kontich, Belgium).

This PDF file includes: