Optical coherence tomography (OCT) is a mainstream optical imaging modality that enables the non-invasive imaging of the morphological structure of the target sample in real-time [1,2]. From the initial stage of development, OCT has been widely utilized in ophthalmology as a primary application [3–5] and has been actively applied to various applications including otolaryngology [6,7], dentistry [8,9], agriculture [10,11], and even in industrial applications [12,13]. Since this imaging technique has high versatility according to the experimental requirements, the conventional bench-top type of OCT is able to be adapted to many different system configurations, such as those handheld [14,15], backpack [16], endoscopic catheter [17,18], and integrated with a surgical microscope [19,20].

In conventional OCT systems, single-mode (SM) fiber has been utilized to deliver the light easily, however, the polarization state in the SM fiber is able to be arbitrarily changed by fiber imperfection, environmental perturbations, and stresses on the freely moving waveguide [21]. Since most modern OCT system configurations employ the polarized light source, the co-polarized backscattered term from the sample is able to be detected after interfering with the reference signal. In addition, because OCT primarily requires the equal polarization state of both reference and sample arms to generate the interference signal, this random polarization state change (i.e., polarization artifact) causes undesired image quality degradation (e.g., image contrast reduction, image resolution degradation, intensity flash, and signal loss in a certain area) [22]. Moreover, the arbitrary polarization state artifacts even more critically affect the signal distortion in freely moving probe-type OCT [23,24]. Therefore, it is necessary to develop additional techniques to minimize the artifact caused by randomly changed polarization states and compensate for the distorted OCT signal.

To address random polarization-state-related artifacts, various researches have been conducted by implementing polarization-sensitive OCT (PS-OCT), a polarization-diverse detection unit, and an optical switching method. In the case of the PS-OCT, it utilizes two different interfered signals from mutually orthogonal planes in the oscillated electromagnetic field and has been widely applied to various applications by providing the depth-resolved sample birefringence and other polarization information. However, PS-OCT requires multiple cameras or detectors to simultaneously obtain the two orthogonally polarized interfered signals, and in addition, the polarization-diverse detection unit, which measures the orthogonally polarized interference signal from the reference and sample arms [25–27]. However, the multi-detectors were essentially used for polarization-diverse detection-unit-based optical demodulation to reconstruct the polarization-independent image, which is hard to use in spectral-domain OCT (SD-OCT). Moreover, the optical switch has also been applied to remove the polarization artifact by sequentially detecting the two orthogonally polarized OCT signals [28]. Although the optical switch enables obtaining the OCT signal in a single spectrometer, it is vulnerable to motion artifacts during the imaging because of the time gap between the two sequentially obtained images and reduces the imaging speed by half.

In terms of the different types of optical fiber for minimizing the polarization artifact, polarization maintaining (PM) fiber has been implemented in the OCT system [29,30]. By parallelly installing the two stress elements to the fiber core along the cladding, the refractive indexes in the PM fiber are enhanced and it allows the light to propagate in two linear orthogonal channels (fast and slow axes). Based on the polarization-independent characteristic of PM fiber, it has been used in a fiber-based PS-OCT configuration [21,29,30] and for measuring the depth-resolved birefringence [31]. Although PM-fiber-based OCT systems have been implemented well, as an aspect of the OCT image quality, the ghost images were generated and even overlapped with the real image, because of the light propagation time differences along with the two polarization modes with different phase velocities [21,29]. There are presented methods to remove the ghost image from the real signal by splicing the PM fiber [21], using longer PM fiber to increase the distance between the ghost and real images, and applying post-processing software [29]. However, splicing inevitably causes signal losses and requires an additional manual process, using the longer fiber is not cost-effective, and the post-processing method imposes computational load.

In this study, we introduce a PM-fiber-based polarization-insensitive OCT (PM-PI-OCT) system, in which the interfered signal is insensitive to random polarization state changes. To provide theoretical proof of the proposed method, we conducted mathematical derivations. We also implemented the method in the system by incorporating the PM fiber with simple optical components such as a polarization controller and linear polarizer, and examined the actual compensated images. Additionally, to quantitatively evaluate the influence of external factors such as fiber rotation angles, we fabricated a 3D-printed apparatus and conducted evaluations to assess the robustness of our proposed method in terms of arbitrary polarization state variations compared with the single-mode-fiber-based conventional OCT system. Finally, using the developed PM-PI-OCT system, we performed in-vivo imaging of the rat retina and human retina as a sample to confirm its applicability under actual experimental conditions.

The optical configuration of the proposed PM-PI-OCT system is shown in Fig. 1. A superluminescent light-emitting diode (EXS210068-01, Exalos, Switzerland), which has 854 nm of center wavelength and 53 nm of full-width at half-maximum, was used as a source of PM-PI-OCT. The output beam of the source, which was naturally polarized (i.e., unpolarized), was distributed to the sample and reference arms according to the coupling ratio of the fiber coupler (TW850R3A2, Thorlabs, USA). To control the polarization state of the input beam to PM fiber (P3-780PM-FC-2, Thorlabs, USA), the polarization controller (FPC023, Thorlabs, USA) was implemented in both the reference and sample arms. The output beam of PM fiber was collimated through a collimator (F240-APC-850, Thorlabs, USA) and linearly polarized by a linear polarizer (LPVIS050, Thorlabs, USA) at both the reference and sample arms. In terms of other optical components in the reference arm, an achromatic doublet lens (AC254-050-AB, Thorlabs, USA) and a protected silver mirror (PF10-03-P01, Thorlabs, USA) were utilized. The passed light through the linear polarizer in the sample arm was reflected by a two-axis galvanometer scanner (GVS002, Thorlabs, USA) for three-dimensional (3D) imaging and focused by an objective lens. The generated interference signal from the sample and reference arms was transferred to the laboratory-based customized spectrometer [32,33]. Among the optical fibers used, the most temperature-sensitive component is the fiber coupler. Therefore, to isolate the optical fiber from temperature influences, a thermal insulator was placed onto the optical table and secured the fiber coupler on top of the insulator. In addition, during experiments, the room temperature was maintained as a constant value to minimize the effect of temperature variance. The purpose of using PM fiber and a linear polarizer, which are key components in PM-PI-OCT and are able to be easily implemented in the conventional SD-OCT system setup, is to compensate for the randomly changed polarization state. A detailed description of the polarization state compensating process is described in Section 2.B.

In this section, complex notation of optical waves for electric or magnetic fields is used and optical intensity is calculated by the squared absolute of a field. The reflected beam from the reference (ER→(z)) and sample (ES→(z)) arms is able to be expressed as ER→(z)=x^ERxe−jkz+y^ERye−j(kz+Δϕr),(1)ES→(z)=x^ESxe−jk(z−z0)+y^ESye−j(k(z−z0)+Δϕs),(2)where ERx and ERy are amplitude coefficients of the reflected reference beam of the x- and y-axes and ESx and ESy are amplitudes of the reflected sample beam of the x- and y-axes. z0 is the path delay of the sample arm to the reference, which is varied according to the sample position. Δϕr and Δϕs are phase differences between the x- and y-axes of reference and sample arms. Through the square-law detection, the measured intensity of the interference signal (I) is expressed as I=|ER+ES|2=IR+IS+(ER*ES+ERES*),(3)where IR and IS are intensities of the reference and sample arms. In addition, the complex conjugate of the field (E) is denoted as E* and the absolute of the field is denoted as |E|. Since we considered the field in both x- and y-axes as shown in Eqs. (1) and (2), the intensities of reference (IR) and sample (IS) arms turn out to be IR=ERx2+ERy2=IRx+IRy,(4)IS=ESx2+ESy2=ISx+ISy,(5)where IRx and IRy are the intensities of the reference beams of the x- and y-axes and ISx and ISy are the intensities of the sample beams of the x- and y-axes. Then, the measured intensity is expressed as I=IR+IS+2Re{ERES*},(6)where Re{ERES*} takes the real part of a complex value. Equation (6) can be rewritten as I=IR+IS+2IRxISxcos(kz0)+2IRyISycos(kz0)cos(Δϕ),(7)where Δϕ is the phase difference between the sample and reference fields (i.e., Δϕs−Δϕr). Expressing the polarization effect in a simplified expression, Eq. (7) turns out to be I=IR+IS+2γIRIScos(kz0),(8)where γ is the polarization-involved factor, which quantifies the polarization-related efficiency. In addition, following Eqs. (6) and (7), γ is able to be denoted as γ=(IRxISxIRIS+IRyISyIRIScos(Δϕ)).(9)

Since the polarization-involved factor (γ) is continuously changed according to the polarization state, the obtained intensities along with the z-axis were not constant (i.e., external banding, resistance, and temperature change crucially affect the intensity), which seems like intensity blinking.

Δz is calculated as the following equation: Δz=λ×LB,(10)where λ is the center wavelength of the source, L is the length of PM fiber, and B is the beat length of PM fiber. In this paper, Δz is calculated as 0.7 mm with λ=854nm, L=2m, and B=2.4mm. By involving the PM fiber in the experimental setup, Eqs. (1) and (2) can be rewritten as ER→(z)=ER1e−jkz+ER2e−jk(z−Δz)+ER3e−jk(z−2Δz),(11)ES→(z)=ES1e−jk(z−z0)+ES2e−jk((z−z0)−Δz)+ES3e−jk((z−z0)−2Δz),(12)where ER1, ER2, and ER3 are amplitude coefficients of the reflected reference beams of fast-fast (FF), fast-slow/slow-fast (FS/SF), and slow-slow (SS) and ES1, ES2, and ES3 are amplitude coefficients of the reflected sample beams of FF, FS/SF, and SS. In addition, intensities of reference (IR) and sample (IS) arms can be expressed as IR=ER12+ER22+ER32=IR1+IR2+IR3,(13)IS=ES12+ES22+ES32=IS1+IS2+IS3.(14)

Then, following Eq. (8), the measured intensity of the interference signal (I) is rewritten as I=IR+IS+2(γ1I1cos(kz0)+γ2I2cos(k(z0−Δz))+γ3I3cos(k(z0+Δz))+γ4I4cos(k(z0−2Δz))+γ5I5cos(k(z0+2Δz))),(15)where γ1 to γ5 and I1 to I5 are polarization-related factors and intensities of each case (case 1 to case 5). Based on Eq. (15), it is proved that five duplicated images are inevitably generated at an interval Δz. Therefore, it is required to develop the compensation method of polarization-related factors to maintain the polarization-insensitive state while utilizing PM fiber in the OCT system.

In our proposed PM-PI-OCT system, as shown in Fig. 1, polarization state control is accomplished by a polarization controller, a PM fiber, and a linear polarizer. The polarization controllers in the reference and sample arms aligned the polarization state of the input beam to the fast or slow axis of PM fiber. In addition, PM fiber maintained the polarization of the input beam, and the linear polarizer delicately controlled the polarization state of transmitted beam through PM fiber by filtering other beams in different states (i.e., remaining light aligned to the fast or slow axis). By doing so, it is able to co-align the input and output polarization states of PM fiber for both sample and reference arms in a linear-polarization state, which means that the polarization states of each arm are solely selectable in fast or slow axis (i.e., FF or SS state). As a result of this process, the five different cases (as shown in Table 1) are simplified into a single state (FFFF or FFSS or SSSS). Therefore, Eq. (15) is simplified in three different states as IFFFF=IR+IS+2γ5I5cos(k(z0+2Δz)),(16)IFFSS=IR+IS+2γ1I1cos(kz0),(17)ISSSS=IR+IS+2γ4I4cos(k(z0−2Δz)),(18)which represents an interfered signal of each combined state (FFFF, FFSS, and SSSS), respectively. The advantages of choosing the specific polarization state are largely divided into two. First, by eliminating the influence of external factors, it is able to obtain a polarization-insensitive state and avoid the image overlapping from different polarization states without missing internal structural information. Next, using duplicated images with five different polarization states occupies around 80% of the space along the depth direction. Therefore, it is hard to acquire additional information by modifying the hardware setup using different sample arms (e.g., space-division method). In contrast, our proposed method offers the advantage of increasing the amount of information obtainable in the depth direction by selecting a single polarization state. The specific alignment procedure of selecting the desired polarization state and changing sequence from one state to another state is described in Section 3.A.

To quantitatively analyze the effect of polarization state change in OCT images according to the rotation angle (i.e., motion artifact of fiber caused by the stresses on the freely moving waveguide and environmental perturbations), we customized the mount using a 3D printer for precise angle control, as shown in Fig. 3. The main purpose of making the customized mount is to quantitatively generate and measure the polarization artifact with the twisting fiber to mimic the actual experimental condition for both SM and PM fibers. The customized mounts are installed at SM and PM fibers, which is illustrated in Fig. 3(a) using 3D modeling software, and both fibers were wound in three turns around a circle with an equal radius. Figures 3(b) and 3(c) are photographs of the 3D-printed customized mount with and without a cover, respectively. In addition, according to the angle mark engraved on the reference plate, the effect of polarization artifact is able to be measured by rotating the wound fiber part at intervals of 22.5° from 0° to 180°, and the representative cases of rotation are shown in Figs. 3(d)–3(g) (0°, 45°, 135°, and 180°, respectively). The measured results of the polarization artifact of each angle are described in Section 3.B in detail. To qualitatively compare polarization variance robustness of SM fiber and PM fiber, we used the structural similarity index (SSIM) and intensity histogram as analysis methods.

To simultaneously measure the human retina image in-vivo and compare the robustness of polarization artifacts between SM fiber and PM fiber, we implemented a dual-OCT system with a single probe, which is composed of conventional SM-fiber-based OCT (SM-OCT) and PM-PI-OCT, as shown in Fig. 4. Since we integrated the dual-OCT system in a single handheld probe, it is able to measure the left and right eyes simultaneously. The used components in conventional OCT are almost identical to PM-PI-OCT, specifically described in Section 2.A and Fig. 1, except using SM fiber (P3-830A-FC-2, Thorlabs, USA) instead of PM fiber. In addition, the linear polarizer was also removed in the conventional OCT system setup. To alternately obtain the interfered signal of PM-PI-OCT and conventional OCT, the MEMS-fiber optical switch (OSW12-780-SM, Thorlabs, USA) was used before the collimator of the spectrometer. The beam path of the optical switch was converted at the end of one B-scan to sequentially acquire the interference signals of PM-PI-OCT and conventional OCT at an equal period. The obtained human retina in-vivo images using this system are described in Section 3.C in detail.

All experimental procedures in this study were performed in accordance with the Kyungpook National University Animal Laboratory Steering Committee (No. KNU-2021-0123). The illumination power of the OCT source on the sample conformed to the American National Standards Institute limits (72mW/cm2 at 860 nm). Normal healthy rats (male, 5 weeks old) were utilized for retinal imaging to evaluate the performance of PM-PI-OCT-based in-vivo imaging. Prior to imaging, the rats were anesthetized with 0.75% isoflurane and 1 L/min of oxygen using an isoflurane machine. The anesthetized rat was placed on an imaging stage equipped with a heating pad to maintain body temperature, and the condition of the rat was confirmed by monitoring the movements of the hands and feet during the experiment.

To clearly demonstrate the polarization-insensitive OCT images using our proposed method, we presented the compensated images of three different states utilizing the IR-card as a sample, as shown in Fig. 5. As described in Section 2.B, the employment of two PM fibers in the reference and sample arms inevitably leads to the generation of five different state images [Fig. 6(a)].

Step 1. Adjust the polarization controller of reference (rPC) and sample (sPC) and linear polarizer of reference (rLP) and sample (sLP) to make the random polarization state, which generates five different images [Fig. 5(a)].

Step 2. Choose the desired polarization state (FFFF, FFSS, and SSSS) and in this step we chose FFFF state to explain the procedure.

Step 3. First adjust sPC to maximize the FFFF signal with minimizing FFSS and SSSS signal, which matches the entrance polarization state of the sample arm for the fast axis.

Step 4. Rotate the sLP to remove the SSSS signal through making the polarization state of the sample arm as FF.

Step 5. Adjust the rPC and rLP with equal sequence of the sample arm (steps 4 and 5) to make the polarization state of reference as FF; then only the FFFF image remains as shown in Fig. 5(b).

Step 6. Based on the system configuration of the FFFF state, by rotating sLP with 90° and adjusting sPC, the polarization state of the sample arm is changed as SS, which makes FFSS images as shown in Fig. 5(c).

Step 7. Based on the system configuration of the FFSS state, by rotating rLP with 90° and adjusting rPC, the polarization state of the reference arm is changed as SS, which makes SSSS images as shown in Fig. 5(d).

Therefore, our proposed method not only removes the polarization sensitivity, but also freely regulates the polarization state from one to another. Since our proposed arbitrary polarization state compensation method involves simply adjusting the hardware state (angle of polarization controller and linear polarizer), the correction process takes only 10 s without introducing any sample position distortion or requiring additional software-based post-processing. Therefore, the top pixel positions along the depth direction for each compensated state are 58 (FFFF), 239 (FFSS), and 420 (SSSS), which are consistent with the locations of the uncompensated images shown in Figs. 2(b) and 2(c). The slight intensity differences observed among the compensated states are caused by the path length difference between the sample and reference paths, as we regulated the reference path length to match the position of the FFFF state.

To quantitatively analyze the robustness of our proposed method for random polarization state variance, we compared the degree of intensity fluctuation according to the different rotation angles of SM-OCT and PM-PI-OCT as shown in Fig. 6. Based on representative rat retina images obtained using SM fiber and PM fiber shown in Figs. 6(a) and 6(g), we rotated the fibers from 0° to 180° with 45° intervals using the apparatus in Fig. 3 and took 100 images at each angle for analysis. The intensity histograms for each angle are shown in Figs. 6(b)–6(f) for SM fiber and Figs. 6(h)–6(l) for PM fiber. To improve the accuracy of image comparison, instead of directly using the acquired original images, we conducted a primary step of cropping the same region in all images before proceeding with image analysis. Qualitatively, it can be observed that the extent of movement along the central axis of the histogram marked by the red line varies with angles for SM fiber, while it remains consistent for PM fiber. Furthermore, the robustness of PM-PI-OCT regarding angle-dependent polarization state changes outperforms that of SM fiber, as indicated by the structural similarity index (SSIM) measured with respect to the 0° image [Fig. 6(m)] and the intensity histogram comparison [Fig. 6(n)]. As shown in Figs. 6(m) and 6(n), PM-PI-OCT demonstrates consistent average and standard deviation values of SSIM and histogram comparison across angles, whereas SM fiber exhibits fluctuations. In addition, as demonstrated in Figs. 6(o) and 6(p), when observing the distribution of SSIM and histogram comparison values for each image, it is evident that PM-PI-OCT images consistently exhibit higher values with small variance. Moreover, analyzing the correlation between histogram comparison and SSIM through the simultaneously plotted Fig. 6(q), it is discernible that PM-PI-OCT images are able to be distinctly separated and grouped from SM-OCT images with higher values. Hence, these results validate that PM-PI-OCT is able to maintain image quality regardless of random fluctuations in the polarization state caused by external influences.

Following the analysis of image quality variances based on the representative five angles of polarization change, we evaluated the polarization-insensitivity of PM-PI-OCT where the angle continuously changes randomly from 0° to 180° during 1000-frame imaging, as shown in Fig. 7. Then, representative images extracted at intervals of 250 frames out of the 1000 frames are shown in Figs. 7(a)–7(e) for SM-OCT and Figs. 7(f)–7(j) for PM-PI-OCT. To examine the image intensity fluctuation due to random polarization state variance during 1000-frame imaging, we averaged the intensity of each image. The number of images corresponding to each average value is mapped as shown in Fig. 7(k). For PM-PI-OCT, the average intensity of the 1000 images is distributed between 52.5 and 54.5, while for SM-OCT, the distribution spans from 50 to 60, indicating vulnerability to polarization changes. Furthermore, the polarization insensitivity of PM-PI-OCT is more pronouncedly evident through the calculation of intensity differences between frames [Fig. 7(l)], based on the first image value as reference. Moreover, the robustness of our proposed method is corroborated by the graph depicting the correlation between averaged intensity and intensity difference [Fig. 7(m)], as well as the plotting of mean and standard deviation [Fig. 7(n)]. Along with the five representative angle-based analyses described in Fig. 6, it is evident from Figs. 7(o)–7(r) that the SSIM and histogram comparing values between PM-PI-OCT images obtained under random polarization states are higher than those of SM-OCT, while indicating a significantly higher image similarity among PM-PI-OCT images. These results prove that our proposed PM-PI-OCT not only minimizes the impact of polarization variance but also maintains a superior polarization-insensitive performance compared to the conventional SM-OCT.

To demonstrate the applicability of PM-PI-OCT, we conducted in-vivo human retina imaging by causing the random polarization state changes of optical fibers. As we mentioned in Section 2.D, the customized handheld-probe for simultaneous imaging of both eyes was used. The maximum light power on the pupil was controlled under 0.9 mW (near-infrared), which is safe enough following the American National Standards Institute (ANSI) standard of ophthalmic instrument. The acquired human retina images of PM-PI-OCT and SM-OCT while randomly changing the fiber angle from 0° to 180° during 500-frame imaging are shown in Fig. 8. Among the obtained 500 consecutive images, five representative images were extracted at intervals of 125 frames out of the 500 frames, shown in Figs. 8(a)–8(e) for SM-OCT and Figs. 8(f)–8(j) for PM-PI-OCT. It is able to identify the superb robustness of our proposed method compared with the conventional OCT system for the polarization changing effect. In addition to quantitatively analyzing the signal differences in each frame, we extracted the signal-to-noise ratio (SNR). We selected the sub-area of the retinal pigment epithelium (RPE) region as a signal value, which has the highest intensity, and then obtained SNR using background intensity as a noise value. The computed SNRs of SM-OCT images were 25.62, 32.28, 30.25, 24.12, and 25.74 dB, whose standard deviation is 3.11 dB. In contrast, the obtained SNRs of PM-PI-OCT images were 32.85, 32.17, 32.38, 33.23, and 31.82, whose standard deviation is 0.49 dB, six times smaller than that of SM-OCT. Therefore, based on these results, the polarization insensitivity of the proposed PM-PI-OCT was verified.

The advantage of our proposed PM-PI-OCT is that it is highly polarization insensitive at various conditions, for three reasons. First, it makes a free movement of a versatile OCT probe after initial compensation of the polarization state. Thus, no additional compensation process is required during system operation since it makes a polarization-insensitive state of the system. Second, whereas in SM-fiber OCT the intensity fluctuation of each frame hinders quantitative comparison between each frame (e.g., monitoring-based analysis), in PM-PI-OCT the intensity differences of each frame are minimized up to 3%, which enables precise comparison for multiple images stacked along in time. Therefore, by utilizing the polarization variance robustness of our proposed system, it is more suitable for consecutive monitoring and medical applications with quantitative analysis. Third, PM-PI-OCT requires no image post-processing, no computational load, no customized optical complex component, and only a simple and quick compensation process, which is able to be finished within 10 s. In addition, when compared to the method of removing inevitably generated ghost images in OCT using conventional PM fiber, our method has the advantage of not requiring additional processes such as splicing. In addition, it is able to use longer polarization maintaining fiber to increase the pixel interval of each polarization state. However, it is required to use longer polarization maintaining fibers of considerable length, which incurs additional coasts. Furthermore, since the images of different polarization states are retained along the depth direction, it is not able to utilize the whole region, limiting the depth range available for imaging. Thus, our proposed method for making a polarization-insensitive state of OCT is easily implementable for any system setup, resulting in high potential for utilization and applicability, because of the simple compensation process and the use of commercially available components. Therefore, when all the aforementioned advantages are summarized, the simplicity and ease of use of our proposed method should make it attractive for general OCT applications.

In this study, we proposed PM-PI-OCT, which used PM fiber and simple optical configurations, to compensate for the random polarization-variance-related artifacts in OCT systems. The concept of this study was theoretically analyzed through mathematical derivations. To compensate for the five different images resulting from the phase delay of the two axes in the PM fiber used to maintain the polarization state, we implemented a system using a polarization controller and a pair of linear polarizers. Furthermore, to quantitatively evaluate the performance of PM-PI-OCT, we measured the intensity variations by rotating the fiber from 0° to 180°. The results confirmed that PM-PI-OCT is not affected by polarization changes, unlike SM-OCT, where intensity varies with the angle. In addition, the applicability of the proposed PM-PI-OCT was proved through human retina imaging with a handheld OCT probe. Our proposed method can be directly applied to existing OCT configurations and offers advantages such as fast compensation within 10 s, and being without the need for additional software-based post-processing or computational load. Therefore, we believe that our PM-PI-OCT setup is able to be practically applied to various OCT systems including conventional bench-top systems, handheld probes, catheters, and endoscopes.

Acknowledgment. This research was supported by the MSIT (Ministry of Science and ICT), Korea, under the Innovative Human Resource Development for Local Intellectualization support program (IITP-2024-RS-2022-00156389) supervised by the IITP (Institute for Information & Communications Technology Planning & Evaluation).