Cerenkov luminescence imaging (CLI) can provide information of medical radionuclides used in nuclear imaging based on Cerenkov radiation, which makes it possible for optical means to image clinical radionuclide labeled probes. However, the exceptionally weak Cerenkov luminescence (CL) from Cerenkov radiation is susceptible to lots of impulse noises introduced by high energy gamma rays generating from the decays of radionuclides. In this work, a temporal median filter is proposed to remove this kind of impulse noises. Unlike traditional CLI collecting a single CL image with long exposure time and smoothing it using median filter, the proposed method captures a temporal sequence of CL images with shorter exposure time and employs a temporal median filter to smooth a temporal sequence of pixels. Results of in vivo experiments demonstrated that the proposed temporal median method can effectively remove random pulse noises induced by gamma radiation and achieve a robust CLI image.
Cerenkov luminescence imaging (CLI) has the ability to optically visualize radioactive decay signals from medical isotopes using optical imaging instruments and attracts more and more attention. CLI utilizes a type of electromagnetic radiation called Cerenkov radiation produced when a charged particle travels faster than the speed of light through an insulating medium [
Robertson et al. first performed CLI on mice using positron-emitting radiotracers [
However, the very weak CL from Cerenkov radiation of radionuclides is susceptible to high level of impulse noises introduced by high energy gamma rays. To obtain strong enough CL signals for imaging, traditional CLI collects one image of CL with long exposure time about several minutes. At the same time, a great number of gamma rays may reach the CCD sensor and produce heavy random impulse noises. So, using shorter exposure time is beneficial to reduce the random impulse noises induced by gamma radiation on the premise of obtaining observable CL signals.
In the current study, single CL image acquisition with minute-long exposure time is substituted by a temporal sequence of CL images with second-long exposure time to suppress these random impulse noises. Accordingly, a temporal median (TM) filter is proposed to remove the high level impulse noises induced by gamma radiation in CLI. Unlike traditional median filter smoothing pixels within spatial domain, TM filter smooths pixels within time domain, which is more effective to remove this kind of additive random noises.
Figure
Schematic diagram of temporal median filter.
We select 9 pixels containing eight neighbor pixels of
A Monte Carlo simulation using MOSE software based on digital mouse atlas is conducted to get the CL image. A point light source mimicking CL source generated from medical isotopes is subcutaneously embedded on back of the digital mouse with the depth about 5 mm as shown in Figure
Numerical simulation of CL image. (a) The digital mouse and CL source location. (b) The simulated CLI image. (c) The fusion image of simulated CL and CT data. (d) The fusion image of simulated CL with impulse noises and CT data.
18F-FDG was provided from the Department of Nuclear Medicine, Xijing Hospital, Fourth Military Medical University. Animal was cared for in accordance with a protocol approved by the Xidian University Animal Care and Use Committee. A Kunming mouse with abdomen unhairing using depilatory cream was used as the imaging object. A pseudotumor was provided using 100 uCi of 18F-FDG with the volume 20 uL mixed with 20 uL matrigel (BD Biosciences, Sparks, MD) injected to abdomen of the mouse.
All CLI images were acquired using a home-made in vivo animal optical imaging system, which includes an Andor Ixon Ultra 897 EMCCD with a Schneider 25 mm f/0.95 lens. After the mouse was anesthetized with intraperitoneal injection of 100 uL anesthetic, the mouse was put into a light-tight chamber of the imaging system and the collection of CLI images started.
Figure
The traditional denoising method for simulated CLI. (a) Simulated CLI image. (b) Simulated CLI image with noises. (c–e) Filtered CLI images using median filter with filter window size = 5, 10, and 15. (f, g) Root mean squared error of the filtered CLI image and ROI of that.
Figure
A comparison between median filter and TM filter. (a) CLI image with no noise. (b) Result of median filter. (c) Result of TM filter. (d) Comparison of RMSE for the two methods.
CL signals are so weak that a long exposure time is necessary for CLI. During the long exposure process, a great number of gamma rays may reach CCD sensor and produce heavy random impulse noises. Figure
The traditional denoising method for CLI. (a) Original CLI image captured with 60 s exposure time. (b–d) Filtered CLI images using median filter with filter window size = 5, 10, and 15. A large filter window size achieves a smooth CLI image with less impulse noises. (e, f) Quantitative analysis of the relations between signal values in region of interest (ROI) marked with red circle and region of background (ROB) marked with red rectangle. The signal values in both ROI and ROB decrease with the increase of filter window size.
Figure
Original CLI images and mean and standard deviation of a CLI image sequence containing 30 CLI images with exposure time of 10 s. (a) is the photograph of the mouse from supine view, and the red dashed box indicates the location of pseudotumor. (b) is detailed view of red dashed box for number 1, 6, 11, 16, 21, and 26 CLI images. (c) is mean image of the CLI image sequence. (d) is the standard deviation image for the CLI image sequence.
A sequence of 30 CLI images with exposure time of 10 s for each one was used to test TM filter. To compare with median filter, we randomly chose a CLI image from the sequence shown in Figure
A randomly chosen original CLI image from a sequence of 30 CLI images with exposure time of 10 s for each one (a), result of median filter fused with white light image (b), and result of TM filter fused with white light image (c).
To further illustrate noise suppression of TM filter, we investigate the background region outside of the mouse after filtering. If TM filter is effective for suppressing these impulse noises, there will be no noises in the background region of the filtered CL image. Due to median filter works on a single CLI image, we obtain mean and standard deviation images of 30 filtered CLI images using median filter as shown in Figures
Mean image of 30 CLI images filtered using median filter (a), standard deviation image of 30 CLI images filtered using median filter (b), and filtered CLI image using TM filter (c). The top is the enlarged image corresponding to the red rectangle region.
Generally, a large sequence containing more CLI images has greater ability to remove random impulse noises, but that will cost more time for data acquisition. Figure
The influence of sequence length to TM filter. (a), (b), and (c) are results of TM filter for
Short exposure time is helpful in reducing the random pulse noise induced by gamma radiation, but CL signals cannot be collected using unduly short exposure time when radioactivity is rather low. In this experiment, the radioactivity of 18F-FDG in the pseudotumor was 100 uCi. We use different lengths of exposure time to investigate the characteristic of TM filter. Figure
The results for TM filter with different exposure time
In this work, a TM filter is proposed to remove the random impulse noises induced by high energy gamma rays generated from radioactive decay of medical nuclides. This method synthesizes characteristics of randomness and pulse for random pulse noises in CL images and employs a temporal median-like filter in a temporal sequence of CL images. Several simulation and in vivo experiments were presented to verify the proposed TM filter, and these results demonstrated that TM filter can effectively remove random impulse noises induced by gamma radiation and achieve a robust CL image. Several key properties of TM filter were also investigated based on in vivo experimental data to fully understand robustness and usability of this method.
In terms of extremely weak CL signal, traditional CLI collects a single CL image with minutes-long exposure time, which often contains lots of random impulse noises due to the arrival of gamma rays to CCD sensor. With the increase of the exposure time, the original CL image contains more and more impulse noises. In order to filter out these random pulse noises, traditional median filter with a large filter window size is employed. The filtered CL image of this single acquisition strategy shows instability and oversmoothness. The proposed method can overcome these drawbacks by collecting a temporal sequence of CLI images with seconds-long exposure time and a TM filter. With a multiple CLI images strategy, the proposed method is useful for suppressing these random pulse noises and getting robust CLI image. The smooth effect of TM filter is to work on pixels of CLI images collected at different time points located in the same coordinate, so the proposed method avoids smoothing CLI image on spatial domain when removing the random pulse noises. Finally, the proposed method can acquire high quality CLI images with high robustness.
To concentrate on the validity of TM for CLI, we perform an in vivo experiment based on a pseudotumor model. The nonspecific distributions of medical nuclides often make the actual CLI applications contain complicated circumstances. So the proposed method needs to be tested by different actual CLI applications.
The authors declare that they have no competing interests.
This work was supported by the Program of the National Natural Science Foundation of China under Grant nos. 81227901, 61405149, 61471279, 31371006, and 81571725, the Natural Science Basic Research Plan in Shaanxi Province of China under Grant no. 2015JZ019, the Beijing Municipal Natural Science Foundation under Grant no. 7142012, and the Fundamental Research Funds for the Central Universities.