Confocal laser endomicroscopy (CLE) is an emerging imaging method with a cellular resolution for obtaining histopathological images of structure information on the mucosa in real time. CLE can significantly improve the detection rate of early tumors. Usually, the fluorescence from the tissue pre-stained with exogenous fluorescent dyes can be detected by the photodetector, which further generates corresponding electrical signals by the photoelectric effect and internal multiplication. Then, the electrical signal can be amplified to form CLE images after analog-to-digital conversion. Meanwhile, the output signal of the photodetector is influenced by the noise, which will further affect the quality of the final image. The noise in CLE mostly includes the shot noise introduced by the inherent fluctuation of obtained photons and the dark noise of detectors. So far, many researchers have mainly proposed CLE using mainly two types of photodetectors, such as avalanche photodiode (APD) and photomultiplier tube (PMT). Usually, APD has the advantage of high quantum efficiency and low cost. PMT has the advantage of high gain and low noise. Therefore, the advantages of the selected photodetectors are very different. At present, no researcher has thoroughly investigated the performance differences between APD and PMT. This paper presents a detailed comparative study of APD and PMT performance in CLE.

Based on the different working principles of APD and PMT, the quantitative model for output signal-to-noise ratio (SNR) in CLE is obtained. The parameters include optical power, quantum efficiency and internal gain of the detector, system bandwidth, photocurrent, dark current, and amplifier noise. All parameters can be obtained from the user manual. Then, we implement a detector performance comparison test, and the light emitted from the LED of 525 nm is used to simulate the fluorescence of samples after passing through a neutral density filter and pinhole. This test not only provides a preliminary understanding of the differences between APD120A2 and PMTSS but also verifies the validity of the proposed model. Eventually, the most important CLE imaging experiment is carried out. For this experiment, a dual optical path CLE imaging system is established which allows simultaneous imaging with APD120A2 and PMTSS under the same optical system by using the beam splitter of 50∶50. The fluorescent samples for this experiment are fluorescein sodium solution, fluorescent beads with diameter of 13 μm, lens tissue (stained by sodium fluorescein), and Photinia serrulata leaves (stained by sodium fluorescein). Distribution of the number of pixels with different optical powers and image SNR is the quantitative evaluation index of the images in this experiment.

Validity of the detector output SNR model for CLE is verified by the detector performance comparison test. This test shows that the SNR of APD120A2 is 2.74 dB on average lower than that of PMTSS. The theoretical value of the model matches with the experimental results, and the correlation error is less than 4.8% (Fig. 5 and Fig. 6). In addition, there is an upper limit to increase the SNR of PMTSS by adding the internal gain of PMT. By imaging the fluorescein sodium solution, we find that due to fluorescence quenching and other factors, the image SNR difference between APD120A2 and PMTSS is 0.28 dB on average (Fig. 7), which is much smaller than the result in detector performance comparison test. The imaging results of other fluorescence samples indicate that the dark current of APD120A2 is higher than that of PMTSS, which affects the low light detection capability of APD120A2. When the average fluorescence optical power is less than 10 nW, the difference in image SNR between APD120A2-based CLE and PMTSS-based CLE increases with decreasing optical power. APD120A2-based CLE imaging performance is comparable to PMTSS-based CLE imaging performance when the average fluorescence optical power is greater than 10 nW. Under this condition, the difference of image SNR is less than 0.67 dB (Fig. 9). Thus, when the optical power is lower than 10 nW, APD120A2 is not recommended for CLE because of the higher dark noise.

This paper presents a detailed study to compare the influence of the selected differences in photodetector on the imaging performance of CLE systems. Firstly, a quantitative model for APD-based and PMT-based CLE SNR is proposed considering different parameters. Secondly, a comparison test for detector performance is carried out to verify the model, which aims to understand the characteristics of APD and PMT. In the end, a dual optical path CLE experimental system is built to evaluate the practical effects of APD120A2 and PMTSS in CLE. The experimental results show that the imaging performance of the APD120A2-based CLE is comparable to that of the PMTSS-based CLE when the average fluorescence optical power is higher than 10 nW. Therefore, in the design of the CLE, it is necessary to determine the detected optical power range for CLE. Selecting APD instead of PMT as the CLE photodetector can obtain more cost-effective imaging results. In the future, based on the SNR model in this paper, we will further improve the imaging quality of APD-based CLE systems by selecting APDs with lower noise factors and designing low-noise APD peripheral circuits. The research results can also provide guidance for selecting photodetectors in low-light detection applications.