Principle of Lock-in Thermography (LIT)

The principle of LIT is based on the introduction of a periodically modulated light source on the surface of the sample and monitoring the resulting local surface temperature of the sample via an IR camera⁴. Absorption of modulated light excitation creates a modulated temperature field (aka thermal-wave field) inside the sample extending axially to a depth proportional to the thermal diffusion length⁴². In this configuration, the presence of defects alters the centroid of the induced thermal-wave field in the defective region, resulting in depth-integrated radiometric signals different from those registered in intact areas. The lock-in demodulation of the depth-integrated signals leads to the calculation of phase and amplitude images⁶. Since the intensity-modulation of excitation in LIT is carried out at a single frequency (f_m), the LIT demodulation of temporal radiometric responses reveals the defective areas in both amplitude and phase images/channels⁴³. The presence of subsurface defects, normally, results in an increase in amplitude values as the thermal impedance introduced by defects results in an additional diffusive thermal contribution to sample surface temperature. Defects also can be identified in phase images because of the alteration of the center of gravity of the thermal-wave field in the defective zone by the introduction of thermal impedance. While generally, amplitude images offer better SNR, LIT phase images are inherently emissivity-normalized and as such not prone to errors caused by variations in emissivity of sample surface⁴⁴. Another key advantage of LIT is its depth profilometric nature. That is, inspection depth in LIT is controlled via the intensity-modulation frequency of external excitation, fm, and in accordance with the definition of thermal diffusion length (i.e., μ=α/πfm with α as thermal diffusivity). As a result, probing deeper into the sample can be achieved by reducing the modulation frequency, while LIT inspection at high modulation frequencies is suitable for the detection of superficial defects (e.g., coatings defects).

A schematic diagram of the LIT imaging system built in our lab using a low-cost cellphone attachment IR camera is shown in Fig. 6a. A fiber-coupled (core diameter = 200 um), continuous-wave near-infrared laser with a center wavelength of 808 nm (Jenoptik, Jena, Germany) was used as an excitation light source. The intensity of excitation light was collimated and homogenized using a collimator and optical diffuser (Thorlabs, Newton, New Jersey, USA, F220SMA-780 and ED1-C20-MD). The sample was secured on a LEGO-jig and mounted on a 3-axis translation stage and excited by the light source. A multifunctional data acquisition unit (National Instruments, Austin, Texas, NI USB-6363 BNC) was used to modulate laser intensity at the desired modulation frequency (fm). The thermal responses of the sample were registered by using a low-cost cellphone attachment IR camera (SEEK THERMAL COMPACT; Seek Thermal Inc.; Android; 156 × 207 Pixels) in conjunction with a low-cost (~$10) CO2 laser cutting Zinc Selenide objective lens (f = 1′′).

(a) Schematic diagram of the lock-in thermography system (b) A use case diagram that depicts a blueprint of the functionalities provided by our developed SDK. (c) A flow chart showing the signal processing method applied to the waveform at each pixel of the image.

While the nominal frame rate of the camera through its standard applet is less than 9 fps, we have managed to utilize USB 2.0 documentation and Microsoft Windows native application programming interfaces (APIs, such as WinUSB and SetupAPI) in order to set up packets of information and send them to the camera’s default endpoint address and, subsequently, acquire frame data from the camera through a corresponding pipe. As such, the developed platform has not only the ability to control camera attributes through a simple USB interface but also can achieve a stable high frame rate of 33 fps through a circular buffer hierarchy and multi-threading. In addition, the program controls the camera shutter and calibrate the camera before capturing the frames, providing stable frames up to a maximum frame rate of 33 fps after calibration. In order to compensate for the non-uniform sampling of data, in our previous work²⁹ we recorded timing information of each frame captured by the camera and employed spline interpolation at a uniform sampling frequency before applying LIT demodulation. The developed SDK, on the other hand, has the ability to capture frames at a constant frame rate, so interpolation is not necessary before the LIT demodulation of thermal signals.

Figure 6b is a use case diagram that depicts a blueprint of the functionalities provided by our developed SDK for controlling the camera and capturing camera frames at a constant and uniform frame rate. First, the user initiates the SDK and checks if the IR camera is connected to the system. Then, the user adjusts the camera setting and initializes the camera for display and visualization of raw frames. Once these initialization processes are carried out, the SKD allows the system to take raw frames in two modes: calibration mode and fast mode. In the calibration mode, camera frames are captured while the shutter is closed and the camera’s internal non-uniformity correction map is recalibrated. The user can adjust the number of frames required for the calibration of the camera. Once the camera calibration is done, the system is set to take images in fast mode at the desired camera frame rate up to a maximum of 33 fps and without any disruption in the acquisition. Capturing frames in fast mode are carried out by sending a request for a frame acquisition followed by a read buffer command to move the acquired frame and time stamp information from camera memory. Once all images are captured, the camera is removed from the system and these captured frames are processed for LIT demodulation as described in Fig. 6c.

Figure 6c depicts the signal processing steps applied to the temporal signal of each camera pixel for LIT demodulation. That is, after acquiring frames, first the effect of bulk heating is removed from the temporal signal by fitting a polynomial curve of order seven. Then, a fast Fourier transform (FFT) is applied to the waveform, followed by determination of the magnitude and phase of the complex number corresponding to the laser modulation frequency in the Fourier domain. At last, the transfer function of the system is applied to the amplitude and phase images to correct imperfections caused by nonuniformities in the illumination and collection optical sub-systems. The transfer function was calculated by performing lock-in imaging on a thick Aluminum block painted with matt black paint on the front surface that resembles a semi-infinite blackbody sample. The theoretical amplitude and phase responses of such samples are known. Therefore, we found the experimental amplitude and phase images at desired modulation frequencies and then calculated the compensation factor for each camera pixel that once applied to the experimental pixel value, yield the theoretical prediction. With this practice, any systematic error caused by illumination and/or acquisition sub-systems can be compensated for. Once the amplitude and phase transfer functions at given modulation frequency are calculated based on the semi-infinite sample, they can directly be applied to amplitude and phase images of any sample to remove the effects of system systematic errors.

A spectral-domain OCT system that comprised of a 1315 nm near-infrared superluminescent diode (Exalos; Switzerland) light source (maximum power 30 mW) was used to take images of demineralized tooth. The OCT system is based on Michelson interferometry in which 50/50 of light is split into the reference and sample arms by a fiber coupler. A polarization controller is used in the reference arm to adjust polarization to the cross-polarization state. A-line scan camera in the spectrometer was composed of 2048 pixels and could acquire A-lines up to a maximum scan/acquisition rate of 140 kHz. The theoretical axial and lateral resolution of the system in teeth is approximately 5 μm and 10 μm, respectively. A GPU (Graphics Processing Unit) based processing was developed for real-time display of OCT B-scans (i.e., cross-sectional image). The B-scan captured from the system were used for studying the effects of demineralization. To find the depth of caries using OCT images, the location of the surface where the scattering occurs and the location of the surface where the scattering ends is thresholded in each vertical line inside the demineralization windows. The color-coded areas in Fig. 3j indicate the thresholded area. Since the depth of each vertical line is unequal, average depth is reported as the depth of caries.

Micro-Computed Tomography (SKYSCAN 1272 high-resolution µCT system, Bruker MicroCT, Kontich, Belgium) was used to take images of the demineralized teeth. The sample was placed in a LEGO block such that the treatment window would fall in the field of view of the µCT detector. A 0.5-mm Al/0.038-mm Cu filter was used to minimize the effects of beam hardening. The tooth was scanned at a rate of 87 kV with camera pixel size of 7.4 µm and exposure time of 2000 ms per frame and a rotation angle of 0.1 degrees. The µCT images (n = 2813) were reconstructed from 1920 projections using NRECON software (Version 1.7.1.6., Skyscan, Kontich, Belgium).

To examine acquisition disruptions and frame rate of manufacturer’s applet and developed SDK (Fig. 1) LIT experiments were carried out on thick Aluminum block (40 mm × 30 mm × 30 mm) with no internal defects. To test the advantage of a higher frame rate for better spatial resolution, an aluminum pin fin thermal heat sink sample was used (Fig. 2a; Wakefield-Vette, New Hampshire, United States). The dimension of the heat sink is 40.6 mm × 40.6 × 13.3 mm and consists of pin fins of cross-section size 1.4 mm × 2.4 mm beneath the 2-mm thick plate, as shown in Fig. 2a. LIT experiments were conducted on plate surface of the thermal heat sink (area shown by the red dashed rectangle in Fig. 2a). The area imaged on the 156 × 207 pixels of the camera was 7.43 × 9.86 mm², respectively. The spatial pixel size of images from the interrogated surface is, therefore, ~48 µm. Raw frames were collected for 90 seconds and those frames were demodulated according to lock-in principles to compute amplitude images. To study the subsurface defect (Fig. 3a), a custom-made Aluminium block (dimensions 45 mm × 25 mm × 30 mm) was used. Three subsurface circular holes of diameter 5 mm were created by drilling the block from the back side. The aforementioned simulates circular defects at 300 µm, 500 µm, and 800 µm below the intact interrogation surface. The thermal diffusivity and conductivity of the aluminum were 9.1 × 10⁻⁵ m²/s⁴⁵and 205 W/mK⁴⁶, respectively. In all LIT experiments carried out on aluminium samples the average optical intensity on the sample surface was 1.6 W/cm² and temporal temperature responses were recorded for 60 seconds.

The anonymous human teeth were collected from local oral surgeons and in accordance to the bio- and laser safety guidelines in place at York University. These specimens were stored in distilled water inside the fume hood to prevent dehydration. In order to induce demineralization in specimens in a controlled manner, an acidified gel was prepared by mixing 0.100 M lactic acid and 0.100 M sodium hydroxide to give a pH value of 4.5 and then adding 6% w/v hydroxyethyl cellulose^18,19.

A tooth with no visible defect or white spot lesion was selected, rinsed thoroughly with distilled water and dried in the air before exposure to the demineralizing gel. In order to achieve localized demineralization, the surface of the tooth was covered by transparent nail polish, leaving small windows (aka treatment windows) exposed. The sample was then submerged upside down in a test tube containing 25 ml of acidified gel for specified durations of demineralization. Two treatment windows were created on the tooth by exposing the left and right treatment windows to the gel for 3 and 7 days, respectively. At the conclusion of artificial demineralization, the sample was removed from the test tube, rinsed under running water, nail polish was removed using Acetone and rinsed again with water before conducting LIT imaging. To take LIT images of the dental sample, the cellphone attachment camera was focused on the surface of the sample that is securely mounted on a LEGO block. LIT imaging was carried out on dental sample by recording the thermal responses for 30 seconds at modulation frequency of 2 Hz. The average optical intensity on the sample surface was 1 W/cm².

OCT imaging was conducted at scan/acquisition rate of 100 KHz. Polarization controller in reference arm was adjusted to the cross-polarization state to minimize surface reflections. To segment the demineralized areas from the healthy area in B-mode images, mean and STD of the amplitude image was calculated and a thresholding operation was applied to the entire image. The global threshold value was calculated by computing the average pixel plus one STD of all the pixels of the image.

To spike LFAs at different THC concentration, saliva samples containing Δ⁹-THC was prepared by adding known volumes of Δ⁹-THC stock solution (MilliporeSigma; Oakville, Canada) to non-stabilized artificial saliva (Pickering Laboratories, Inc, Mountain View, California, USA). The THC solution was prepared following the standardized procedure suggested by the Canadian Society of Forensic Science Drugs and Driving Committee was followed⁴⁷. Commercially available oral fluid LFA strips (NARCOCHECK Saliva Test Strips, Kappa City Biotech SAS, Montluçon, France) were used. Six different concentrations (25, 10, 7.5, 5, 2 and 0 ng/ml) of THC-saliva solution was prepared. In each experiment, 150 µL of the solution was pipetted to the sample pad of the LFA strip. Eight LFAs were spiked at each concentration to investigate reproducibility and each LFA was interrogated five times using the developed low-cost LIT system to investigate repeatability. All LIT experiments were carried out with average optical intensity of 1.6 W/cm² and temporal temperature responses were recorded for 60 seconds.