1. INTRODUCTION

X-ray detection plays a vital role in various applications, including medical imaging, security checking, radioactive material detection, and industrial flaw inspections [1–3]. During the recent decade, dynamic and high-speed x-ray imaging has been rapidly developed to meet the enhanced criteria, such as dynamic imaging for medical diagnosis, high-speed detection in security checking, kilohertz–gigahertz x-ray imaging in a synchrotron facility, and nuclear reaction monitoring [4–7]. However, the advancements in high-speed x-ray imaging technology have also resulted in a substantial increase in data volume, which will impose system limitations on data transmission, temporal bandwidth, processing capability, and power consumption [8–10]. For example, current dynamic digital radiography could even generate data volumes up to hundreds of Gbps, which is much higher than the transmission abilities of interfaces such as CameraLink HS (10 Gbps) and GigE (6.8 Gbps).

Endeavors have been devoted to solving the issue of redundant data volume, such as random addressing of pixels in the image plane, pixel binning, photoelectric coupling, multiplexing, interface enhancements, and GPS upgrades [11–13]. However, the progress achieved so far is limited, and the data volume is still approaching the ceiling of the system’s capability.

In sharp contrast, the human retina can efficiently extract key information at an unfixed frame rate, from the input image received by the human eye, and then output it to the brain asynchronously. Inspired by the function of retina, event-based vision sensors (EVSs) have been intensively developed. EVSs can mimic human bipolar cells and generate potential changes only when there is a variation in photon intensity. The variation in photon outputs an “event,” and only such meaningful events are further stored as the registration of data [14,15]. This method shows grand advantages of high bandwidth, minimized data volume, and resilience to constant background. In principle, an EVS-like x-ray imager is highly promising to improve the aforementioned limitations of dynamic x-ray detection systems. Yet, as far as we are concerned, there is no study about x-ray imaging based on EVS technology.

Here, we propose, to the best of our knowledge, the first event-based x-ray imager. It is emphasized that one major obstacle to realizing the above task is the lack of ghosting-free scintillators. Conventional scintillators, such as CsI:Tl and NaI:Tl, always encounter the problems of serious afterglow and ghosting effect, and the limited approaches are mostly oriented toward doping or co-doping at the price of a substantially decreased scintillation yield. This work develops the new scintillator ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ film with a one-dimensional crystal structure, high light yield, as well as low afterglow. We quantitatively evaluated the ghosting of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ scintillator film as only 2.4% of CsI:Tl film. The negligible ghosting guarantees their successful application in combination with event-based sensors, and the imaging system is applied to liquid flow monitoring with high spatial resolution (shown in Supplement 1, Fig. S7) and no ghosting artifacts. Benefiting from the principle of asynchronous working, it achieves an impressive data compression ratio of 23.7%, significantly reducing power consumption. Furthermore, such imaging systems have been shown to possess x-ray-tracing capabilities that enable edge imaging under x-ray exposure [16]. The technique presented in this work opens up opportunities for the future realization of advanced x-ray detectors.

 

Fig. 1. Working mechanism of event-based x-ray imager. We demonstrate conventional and event-based x-ray imaging systems for imaging of moving objects. (a) Conventional x-ray imager outputs the contrast intensity of the whole pixel arrays. The gray square array represents a frame, with each small square representing a pixel. The color of the pixel indicates the corresponding gray value of the image. Lighter colors indicate higher gray values for that pixel’s output. (b) Event-based x-ray imager only outputs contrast changes. Decreases or increases in contrast correspond to triggering an “OFF” or “ON” event, resulting in the output of an image. Pixels in the background with no change, in contrast, remain in static mode and do not produce any data output. (c) Schematic of the operation of an EVS pixel, converting contrast intensity changes into events. Here, P represents the polarity. ${{\rm P}_ +}$ and ${{\rm P}_ -}$, respectively, represent the transition from low to high or from high to low in contrast intensity, also known as ON or OFF events. (d) Schematic responses are produced by a conventional frame sensor (CFS) and an EVS. Top: series of consecutive simulated frames of liquid flow monitoring. Middle: signal returned by a CFS with a 33 ms exposure time corresponding to the image frame. Bottom: signal returned by an EVS.

Download Full Size | PDF

2. RESULTS

The dominating x-ray detection mainly depends on scintillators to first convert xrays into visible photons, and vision sensors could capture the visible photons for signal extraction [Figs. 1(a) and 1(b)] [17]. In a conventional x-ray imager, the contrast intensity of the whole pixel arrays would be stored and transmitted to later processing units. The “absolute” contrast is measured at a constant rate, and the resulting image is returned as a complete frame [Fig. 1(a)]. Such conventional methods often suffer from data redundancy and contain a substantial amount of invalid information [18]. In sharp contrast, our event-based x-ray imager utilizes ghosting-free scintillation film to timely down-convert x rays and event-based vision sensors to asynchronously measure the contrast change of each pixel. In instances where no contrast changes, the sensor stays silent. The outcome is a sequence of visuals that exclusively encompass accurate data regarding the pixels responsible for the alteration in contrast. These images discard any irrelevant or static regions, focusing solely on the pixels that contribute to the detected events [19]. By isolating the pertinent information, the resulting image series provides a clear representation of the specific areas involved in the contrast changes, enabling accurate analysis and interpretation of the x-ray data.

Specifically, each pixel in the event sensor is independent of the other, which means each pixel senses x rays independently. We set a contrast change threshold C for pixels, so the pixels keep comparing the relative change value of the signal all the time rather than sampling the actual signal value. When the pixel senses a change in signal greater than the set threshold, it triggers an event signal [Fig. 1(c)]. The content of the event signal includes the coordinate address of the pixel, the polarity of the change in contrast, and the timestamp of the event, where polarity refers to an increase (called “ON”) or decrease (called “OFF”) in contrast. When the photocurrent sensed by the pixel remains constant, the pixel enters standby mode without signaling output. Therefore we say that event sensors are asynchronous sensing. Event-based imaging produces continuous streams of events without dynamic blurring, removing sampling rate limitations and achieving improved imaging effects, unlike conventional imaging. Figure 1(d) displays the differences between the responses of EVSs and CFSs for the stimulated signal of a liquid flow process without and with contrast agent. Furthermore, it allows for stable and clear imaging in both bright and dark scenes to achieve a high dynamic range. The reason is that the output signal is based on the relative change in brightness intensity. In addition, such an imaging method eliminates the storage of information about the background, which is equivalent to preprocessing the image to remove useless information about the background, thus realizing data volume saving.

 

Fig. 2. Scintillation performance of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$. (a) Absorption coefficients of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ and mainstream scintillator CsI:Tl as a function of x-ray energy (from 1 keV to ${{10}^3}\;{\rm MeV}$). (b) Radioluminescence spectrum of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$. The peak position is at 466 nm. (c) Principle schematic of the close-space sublimation method. (d) Cross-sectional SEM images of the columnar ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ film prepared by close-space sublimation method. (e) Photoluminescence decay profiles of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$. (f) Photoluminescence quantum yield of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ film. (g) Recorded x-ray responses of LYSO: Ce, CsI:Tl, and ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$. (h) Radioluminescence intensity of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ powder under x-ray irradiation with different dose rates.

Download Full Size | PDF

The scintillator film, as a core component of the imaging system, is critical for its imaging performance. To match high temporal and spatial resolution imaging, the scintillator film should meet the requirements of low afterglow, high luminescence efficiency, and low optical crosstalk [20]. Current methods to reduce afterglow are restricted by the sacrifice of scintillation light output. For instance, CsI:Tl exhibits a significant drawback of prolonged afterglow (${\gt}{2}\%$ at 3 ms) [21], and the doping of Bi could decrease the light yield to 55% of the original value [22]. In contrast, copper-based halides are excellent scintillators for x-ray detectors owing to their remarkable attributes such as minimal afterglow and high light output. For example, previous work reported ${{\rm Cs}_3}{{\rm Cu}_2}{{\rm I}_5}$ crystal afterglow of 0.03% at 10 ms as well as ${{\rm CsCu}_2}{{\rm I}_3}$ crystal afterglow of 0.008% at 10 ms [23,24]. Consequently, we choose a novel one-dimensional copper-based scintillator, ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$; crystal possesses a combination of low afterglow (0.1% at 10 ms) and high light output (65,000 photons/MeV). Notably, it can be manufactured in large areas and the film maintains the low afterglow levels (0.1% at 10 ms) observed in the crystal form, demonstrating its excellent performance and stability in maintaining minimal afterglow [25]. These features make it an ideal candidate for creating x-ray imagers with EVS. As previous work reported, the ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ scintillator is a one-dimensional crystal structure. It meets the advantage of being thermodynamically easy to grow into columnar structures, which is good for reducing light scattering, minimizing pixel crosstalk, and achieving high spatial resolution.

Here, we investigated the material properties of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$. The physical diagram of the ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ film is shown in Supplement 1, Fig. S1. Figure 2(a) shows the x-ray absorption coefficients of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ in the energy range from ${{10}^{- 3}}$ to ${{10}^3}\;{\rm MeV}$, close to that of commercial CsI:Tl. Figure 2(b) shows the radioluminescence (RL) spectra of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ powder. The emission peak of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ is at 466 nm, which is coupled with the spectral response range of the EVS. Figure 2(h) presents the RL of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$, showing a good linear dependence with the x-ray dose. The close-space sublimation (CSS) method, with lamp heating and a short deposition distance, has the advantages of rapid heating and evaporation and is commonly used for rapid evaporation of large-area thin-film processes [Fig. 2(c)]. We prepared ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ thin films using CSS, whose cross-sectional morphology showed a columnar crystalline structure [Fig. 2(d)], which is conducive to improve the spatial resolution of imaging. The time-resolved photoluminescence decay profiles of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ under 340 nm excitation were acquired [Fig. 2(e)], and decay profiles can be well-fitted with a short decay time of tens of 40 µs. In addition, we investigated the luminescence efficiency of the ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ film, which was experimentally demonstrated to have a high photoluminescence quantum yield of $92 \pm 1\%$ [Fig. 2(f)]. The scintillators with similar sizes were placed in the same position within an integrating sphere for scintillation light collection. The light response signal emitted by the material is detected by a wavelength-dependent detection efficiency of a photodetector. Finally, comparing their output light response signals, ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ is 1.6 times more efficient than CsI:Tl and 3.6 times more efficient than LYSO [Fig. 2(g)]. Therefore, ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ has excellent luminescence efficiency, which can realize high absorption of high-energy x-ray photons and efficiently convert them into visible photons to enhance spatial resolution.

Then we evaluated the ghosting effect of the scintillation film. Trapping in the layers is followed by the slow release over a broad range of time constants and finally gives rise to a ghosting effect [26–28], which will cause the sensor to output error events to affect the imaging results (Fig. S2, Supplement 1). As previously mentioned, the one-dimensional nature of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ ensures that the grain boundaries in the film maintain chain integrity, resulting in minimizing afterglow [29,30]. To investigate the issue of ghosting effects of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ scintillator films, we conducted tests comparing with commercially available CsI:Tl films. Specifically, in the experiment, as shown in Fig. 3(a) and Note 1 in Supplement 1, we employed tungsten sheets to partially cover the scintillator films, i.e., region A of the film. The film was then exposed to multiple x-ray exposures, allowing the afterglow traps in the uncovered portion, i.e., region B of the film, to be adequately filled. Subsequently, we removed the tungsten sheets and exposed the entire film again. At this stage, region B of the film exhibited not only fluorescence from the current exposure but also phosphorescent components emitted by the previously filled afterglow traps. If a scintillator exhibits significant afterglow, characterized by a large number of afterglow traps, the brightness of region B during the final exposure will be much higher than the fluorescence emitted by region A. By comparing the percentage difference in brightness between region A and region B during the final exposure, we can assess the extent of ghosting artifacts present in the scintillator film.

 

Fig. 3. Ghosting effect evaluation of ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ and CsI:Tl scintillator films. (a) Schematic diagram of physical placement in tests. The scintillator film is divided into two regions: region A and the remaining region B. During the test, we used tungsten sheets to shield region A of the film from radiation exposure. Images of (b) ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ film and (c) CsI:Tl film after x-ray exposure.

Download Full Size | PDF

To mitigate the impact of film uniformity on the test results, we captured initial images of the films before conducting the tests. When calculating the brightness difference during the test, we subtracted the original difference between the initial images from the obtained difference. This approach ensures that the measured brightness difference accurately reflects the contribution of afterglow while minimizing the impact of any inherent nonuniformity in the film. Figures S3 and S4 (see Supplement 1) recorded the signals of the film with and without masking exposed to x ray, and Table S1 in Note 1 recorded the detailed signal values. Figures 3(b) and 3(c) illustrate the imaging results of the final exposure for both the ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ and CsI:Tl films. The striped patterns observed in the images are related to the coupling between the scintillator films and the CMOS panel used in the backend. From the analysis, it can be observed that the brightness levels between the covered and uncovered portions of the ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ film are relatively similar, with a difference of only 0.1% (recording to Eq. S1, Supplement 1). The ghosting artifacts are almost negligible. However, in the case of CsI:Tl, a significant boundary can be observed, with a striking brightness difference of 4.1% (recording to Eq. S1, Supplement 1). This indicates that during dynamic imaging, the pronounced ghosting artifacts caused by CsI:Tl can have a substantial impact on discerning the motion of the target object. The “false increase” in brightness can trigger imager events, resulting in image artifacts and significant residual effects on the observed motion process.

We used such an EVS in an optical setup designed for event-based x-ray imager, as presented in Fig. 4(a). In this imaging framework, x rays are transformed into visible light by a coupled scintillator and detected by an EVS. Each pixel of the EVS consists of a photodetection unit and a brightness detection unit. The photodetection unit converts the visible light signal into a voltage value, while the differential detection circuit in the brightness detection unit monitors the changes between the reference voltage and the incident light voltage. If the change exceeds a predetermined threshold in either the positive or negative direction, the comparator recognizes it as an event and outputs the associated data. Afterward, the circuit resets using the detected brightness of the previous event as the baseline. The threshold voltage is adjusted accordingly, using this new baseline as a reference, to facilitate the subsequent processing and event detection of the incoming light signal.

 

Fig. 4. Application of event-based x-ray imager. (a) Optical setup used for event-based x-ray imager. The inset shows a physical image of the imager. (b) Series of images for moving spring tested by an event-based x-ray imager, and the moving speed is 2 cm/s. (c) Image processing procedure of conventional x-ray imagers and event-based x-ray imagers. Comparison of the results of (d) conventional and (e) event-based imaging for liquid flow monitoring. The conventional one includes imaging of liquid without and with a contrast agent, subtracting the two to get the final image. The event-based one shows the image without any processing. (f) Distribution of gray values in images for liquid flow monitoring. The upper one represents the conventional final image after subtraction, and the lower one represents the image generated by event-based imaging. Gray values are divided into 256 levels, and count represents the number of pixels that output a specific gray value. (g) Total gray value counts for the conventional image for liquid flow monitoring before and after contrast subtraction, as well as for the event-based image for liquid flow monitoring.

Download Full Size | PDF

During the motion of an object, it passes over a specific position on the scintillator film at a given moment. This information is represented using timestamps (${t_i}$) and positional coordinates (${x_i},\;{y_i}$). The subsequent moment in time is denoted by (${t_{i + 1}}$, ${x_{i + 1}}$, ${y_{i + 1}}$). When exposed to x rays, the object obstructs a portion of the x-ray energy, causing a decrease in the emitted brightness intensity from the corresponding position on the scintillator film. Therefore, within the time interval $\Delta {t_i}$ from ${t_i}$ to ${t_{i + 1}}$, the EVS detects changes in the contrast intensity sensed by specific pixels. At the moment of ${t_{i + 1}}$, the pixel at coordinates (${x_i},\;{y_i}$) senses an increase in contrast intensity, triggering an “ON” event, represented as a white color in the image. Simultaneously, the pixel at coordinates (${x_{i + 1}},\;{y_{i + 1}}$) senses a decrease in contrast intensity, triggering an “OFF” event, depicted as blue in the image.

Figure 4(b) displays a clear imaging result of a moving spring at a speed of 2 cm/s, presented in the form of a pseudo-color image. As the spring moves from left to right, the right end of the spring under the x ray is shown in blue, while the left part is depicted in white. The size of the white area in the image reflects the response speed of the scintillator film. A slower response speed means that the scintillator film takes a longer time to absorb high-energy x-ray photons and convert them into visible photons, leading to significant tailing in the images obtained through the event-based x-ray imager. The tailing in the blue part of the image reflects the afterglow level of the scintillator film. Prolonged afterglow can lead to residual shadows of moving objects, causing ghosting artifacts. In event-based x-ray imaging, this effect manifests as smaller changes in brightness contrast that do not trigger events. This can result in blurred edges and reduced contrast in the resulting image. Figure 4(b) does not exhibit significant tailing in the blue and white images, and the edge of the spring can be clearly seen, indicating that the ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ scintillator film possesses a fast response speed and low afterglow level.

In addition to imaging monitoring of moving solid objects, we also conducted imaging of a simulated liquid flow monitoring by radiography. Liquid flow monitoring captures clear images by detecting the contrast agent within the fluid. Here, we use a heavy metal ion ${{\rm Pb}^{2 +}}$ solution that blocks x rays to replace the contrast agent in the simulated test. Figure 4(c) illustrates the image processing procedure for x-ray imaging using both a CFS and an EVS. Traditional x-ray detection methods rely on CFSs to detect the absolute values of brightness intensity. CFS-based subtraction radiography subtracts two x-ray images taken before and after contrast agent injection to eliminate static background tissue, resulting in a clear fluid image (Fig. S5 in Supplement 1). Therefore, image subtraction processing is required to obtain effective images.

In sharp contrast, the event-based x-ray imager is sensitive to changes in contrast intensity and can directly capture the imaging process during contrast enhancement. Before the introduction of ${{\rm Pb}^{2 +}}$, the EVS did not detect changes in contrast intensity, resulting in no effective image output. As shown in Fig. S6 (see Supplement 1), the injection of ${{\rm Pb}^{2 +}}$ resulted in a noticeable decrease in contrast intensity, which in turn triggered “OFF” events. However, once the injection of ${{\rm Pb}^{2 +}}$ was halted and the contrast intensity began to return, it triggered “ON” events. The liquid flow path is visible, and the contrast agent’s effect is visually apparent. Figures 4(d) and 4(e) display the imaging results of the two methods for liquid flow monitoring. The image based on the CFS is difficult to distinguish before contrast subtraction, whereas the image from the EVS is straightforward and clear. Furthermore, the direct imaging capability of the EVS achieves significant data compression. The gray value distributions of the conventional image after background subtraction and event-based image are shown in Fig. 4(f). Figure 4(g) shows the total gray value counts for the conventional image before and after contrast subtraction, as well as for the event-based image, which are 312,725, 86,440, and 20,474, respectively. The total count value represents the number of pixels that generated a brightness response in a single frame and can be understood as the data transmission quantity. Thus, the event-based x-ray imager achieves a 23.7% compression in data volume compared to traditional detection systems (Note 2 in Supplement 1), which alleviates the burden on data transmission, processing, and storage. The smaller the area in which the mask generates changes in brightness intensity, the greater the data savings in event-based x-ray detection systems.

3. CONCLUSION

In sum, we have successfully developed, to the best of our knowledge, the first event-based x-ray imager, combining the novel low afterglow copper-based scintillator film with event-based vision sensor. Compared to the conventional frame sensor, the event-based sensor measures changes in contrast intensity rather than “absolute” brightness intensity values. As a result, it offers high sensitivity to dynamic objects and does not respond to static backgrounds, thus achieving data volume compression. As a result, such an event-based x-ray imager achieves a high image quality with substantial data volume compression, such as achieving 23.7% compression of data volume in applications such as liquid flow monitoring. In addition, benefiting from the coupled ${{\rm Cs}_5}{{\rm Cu}_3}{{\rm Cl}_6}{{\rm I}_2}$ scintillator film with negligible ghosting (0.1%) and excellent columnar structure, the imager has achieved ghosting-free effect and high spatial resolution in dynamic x-ray imaging. Furthermore, the concept of the event-based x-ray imager can be further applied to other ghosting-free and low power consumption x-ray imaging applications.